"Counting sheep PSG": EEGLAB-compatible open-source matlab software for signal processing, visualization, event marking and staging of polysomnographic data

BACKGROUND

Progress in advancing sleep research employing polysomnography (PSG) has been negatively impacted by the limited availability of widely available, open-source sleep-specific analysis tools.

NEW METHOD

Here, we introduce Counting Sheep PSG, an EEGLAB-compatible software for signal processing, visualization, event marking and manual sleep stage scoring of PSG data for MATLAB.

RESULTS

Key features include: (1) signal processing tools including bad channel interpolation, down-sampling, re-referencing, filtering, independent component analysis, artifact subspace reconstruction, and power spectral analysis, (2) customizable display of polysomnographic data and hypnogram, (3) event marking mode including manual sleep stage scoring, (4) automatic event detections including movement artifact, sleep spindles, slow waves and eye movements, and (5) export of main descriptive sleep architecture statistics, event statistics and publication-ready hypnogram.

COMPARISON WITH EXISTING METHODS

Counting Sheep PSG was built on the foundation created by sleepSMG (https://sleepsmg.sourceforge.net/). The scope and functionalities of the current software have made significant advancements in terms of EEGLAB integration/compatibility, preprocessing, artifact correction, event detection, functionality and ease of use. By comparison, commercial software can be costly and utilize proprietary data formats and algorithms, thereby restricting the ability to distribute and share data and analysis results.

CONCLUSIONS

The field of sleep research remains shackled by an industry that resists standardization, prevents interoperability, builds-in planned obsolescence, maintains proprietary black-box data formats and analysis approaches. This presents a major challenge for the field of sleep research. The need for free, open-source software that can read open-format data is essential for scientific advancement to be made in the field.

Automatic detection of midfacial fractures in facial bone CT images using deep learning-based object detection models

BACKGROUND

Midfacial fractures are among the most frequent facial fractures. Surgery is recommended within 2 weeks of injury, but this time frame is often extended because the fracture is missed on diagnostic imaging in the busy emergency medicine setting. Using deep learning technology, which has progressed markedly in various fields, we attempted to develop a system for the automatic detection of midfacial fractures. The purpose of this study was to use this system to diagnose fractures accurately and rapidly, with the intention of benefiting both patients and emergency room physicians.

METHODS

One hundred computed tomography images that included midfacial fractures (e.g., maxillary, zygomatic, nasal, and orbital fractures) were prepared. In each axial image, the fracture area was surrounded by a rectangular region to create the annotation data. Eighty images were randomly classified as the training dataset (3736 slices) and 20 as the validation dataset (883 slices). Training and validation were performed using Single Shot MultiBox Detector (SSD) and version 8 of You Only Look Once (YOLOv8), which are object detection algorithms.

RESULTS

The performance indicators for SSD and YOLOv8 were respectively: precision, 0.872 and 0.871; recall, 0.823 and 0.775; F1 score, 0.846 and 0.82; average precision, 0.899 and 0.769.

CONCLUSIONS

The use of deep learning techniques allowed the automatic detection of midfacial fractures with good accuracy and high speed. The system developed in this study is promising for automated detection of midfacial fractures and may provide a quick and accurate solution for emergency medical care and other settings.

Strategies for enhancing automatic fixation detection in head-mounted eye tracking

Moving through a dynamic world, humans need to intermittently stabilize gaze targets on their retina to process visual information. Overt attention being thus split into discrete intervals, the automatic detection of such fixation events is paramount to downstream analysis in many eye-tracking studies. Standard algorithms tackle this challenge in the limiting case of little to no head motion. In this static scenario, which is approximately realized for most remote eye-tracking systems, it amounts to detecting periods of relative eye stillness. In contrast, head-mounted eye trackers allow for experiments with subjects moving naturally in everyday environments. Detecting fixations in these dynamic scenarios is more challenging, since gaze-stabilizing eye movements need to be reliably distinguished from non-fixational gaze shifts. Here, we propose several strategies for enhancing existing algorithms developed for fixation detection in the static case to allow for robust fixation detection in dynamic real-world scenarios recorded with head-mounted eye trackers. Specifically, we consider (i) an optic-flow-based compensation stage explicitly accounting for stabilizing eye movements during head motion, (ii) an adaptive adjustment of algorithm sensitivity according to head-motion intensity, and (iii) a coherent tuning of all algorithm parameters. Introducing a new hand-labeled dataset, recorded with the Pupil Invisible glasses by Pupil Labs, we investigate their individual contributions. The dataset comprises both static and dynamic scenarios and is made publicly available. We show that a combination of all proposed strategies improves standard thresholding algorithms and outperforms previous approaches to fixation detection in head-mounted eye tracking.

Towards the automated detection of interictal epileptiform discharges with magnetoencephalography

BACKGROUND

The analysis of clinical magnetoencephalography (MEG) in patients with epilepsy traditionally relies on visual identification of interictal epileptiform discharges (IEDs), which is time consuming and dependent on subjective criteria.

NEW METHOD

Here, we explore the ability of Independent Components Analysis (ICA) and Hidden Markov Modeling (HMM) to automatically detect and localize IEDs. We tested our pipelines on resting-state MEG recordings from 10 school-aged children with (multi)focal epilepsy.

RESULTS

In focal epilepsy patients, both pipelines successfully detected visually identified IEDs, but also revealed unidentified low-amplitude IEDs. Success was more mitigated in patients with multifocal epilepsy, as our automated pipeline missed IED activity associated with some foci-an issue that could be alleviated by post-hoc manual selection of epileptiform ICs or HMM states.

COMPARISON WITH EXISTING METHODS

We compared our results with visual IED detection by an experienced clinical magnetoencephalographer, getting heightened sensitivity and requiring minimal input from clinical practitioners.

CONCLUSIONS

IED detection based on ICA or HMM represents an efficient way to identify IED localization and timing. The development of these automatic IED detection algorithms provide a step forward in clinical MEG practice by decreasing the duration of MEG analysis and enhancing its sensitivity.

Performance of artificial intelligence-based software for the automatic detection of lung lesions on chest radiographs of patients with suspected lung cancer

PURPOSE

This study aimed to evaluate the performance of the commercially available artificial intelligence-based software CXR-AID for the automatic detection of pulmonary nodules on the chest radiographs of patients suspected of having lung cancer.

MATERIALS AND METHODS

This retrospective study included 399 patients with clinically suspected lung cancer who underwent CT and chest radiography within 1 month between June 2020 and May 2022. The candidate areas on chest radiographs identified by CXR-AID were categorized into target (properly detected areas) and non-target (improperly detected areas) areas. The non-target areas were further divided into non-target normal areas (false positives for normal structures) and non-target abnormal areas. The visibility score, characteristics and location of the nodules, presence of overlapping structures, and background lung score and presence of pulmonary disease were manually evaluated and compared between the nodules detected or undetected by CXR-AID. The probability indices calculated by CXR-AID were compared between the target and non-target areas.

RESULTS

Among the 450 nodules detected in 399 patients, 331 nodules detected in 313 patients were visible on chest radiographs during manual evaluation. CXR-AID detected 264 of these 331 nodules with a sensitivity of 0.80. The detection sensitivity increased significantly with the visibility score. No significant correlation was observed between the background lung score and sensitivity. The non-target area per image was 0.85, and the probability index of the non-target area was lower than that of the target area. The non-target normal area per image was 0.24. Larger and more solid nodules exhibited higher sensitivities, while nodules with overlapping structures demonstrated lower detection sensitivities.

CONCLUSION

The nodule detection sensitivity of CXR-AID on chest radiographs was 0.80, and the non-target and non-target normal areas per image were 0.85 and 0.24, respectively. Larger, solid nodules without overlapping structures were detected more readily by CXR-AID.

A novel approach for diabetic foot diagnosis: Deep learning-based detection of lower extremity arterial stenosis

PURPOSE OF THE STUDY

Assessing the lower extremity arterial stenosis scores (LEASS) in patients with diabetic foot ulcer (DFU) is a challenging task that requires considerable time and efforts from physicians, and it may yield varying results. The presence of vascular wall calcification and other irrelevant tissue information surrounding the vessel can further compound the difficulties of this evaluation. Automatic detection of lower extremity arterial stenosis (LEAS) is expected to help doctors develop treatment plans for patients faster.

METHODS

In this paper, we first reconstructed the 3D model of blood vessels by medical digital image processing and then utilized it as the training data for deep learning (DL) in conjunction with the non-calcified part of blood vessels in the original data. We proposed an improved model of vascular stenosis small target detection based on YOLOv5. We added Convolutional Block Attention Module (CBAM) in backbone, replaced Path Aggregation Network (PANET) with Bidirectional Feature Pyramid Network (BiFPN) and replaced C3 with GhostC3 in neck to improve the recognition of three types of stenosis targets (I: <50 %, II: 51 % - 99 %, III: completely occluded). Additionally, we utilized K-Means++ instead of K-Means for better algorithm convergence performance, and enhanced the Complete-IoU (CIoU) loss function to Alpha-Scylla-IoU (ASIoU) loss for faster reasoning and convergence. Lastly, we conducted comparisons between our approach and five other prominent models.

RESULT

Our method had the best average ability to detect three types of stenosis with 85.40% mean Average Precision (mAP) and 74.60 Frames Per Second (FPS) and explored the possibility of applying DL to the detection of LEAS in diabetic foot. The code is available at github.com/wuchongxin/yolov5_LEAS.git.

Deep Learning Application to Detect Glaucoma with a Mixed Training Approach: Public Database and Expert-Labeled Glaucoma Population

INTRODUCTION

Artificial intelligence has real potential for early identification of ocular diseases such as glaucoma. An important challenge is the requirement for large databases properly selected, which are not easily obtained. We used a relatively original strategy: a glaucoma recognition algorithm trained with fundus images from public databases and then tested and retrained with a carefully selected patient database.

METHODS

The study's supervised deep learning method was an adapted version of the ResNet-50 architecture previously trained from 10,658 optic head images (glaucomatous or non-glaucomatous) from seven public databases. A total of 1,158 new images labeled by experts from 616 patients were added. The images were categorized after clinical examination including visual fields in 304 (26%) control images or those with ocular hypertension and 347 (30%) images with early, 290 (25%) with moderate, and 217 (19%) with advanced glaucoma. The initial algorithm was tested using 30% of the selected glaucoma database and then re-trained with 70% of this database and tested again.

RESULTS

The results in the initial sample showed an area under the curve (AUC) of 76% for all images, and 66% for early, 82% for moderate, and 84% for advanced glaucoma. After retraining the algorithm, the respective AUC results were 82%, 72%, 89%, and 91%.

CONCLUSION

Using combined data from public databases and data selected and labeled by experts facilitated improvement of the system's precision and identified interesting possibilities for obtaining tools for automatic screening of glaucomatous eyes more affordably.

Computerized detection of cyclic alternating patterns of sleep: A new paradigm, future scope and challenges

BACKGROUND AND OBJECTIVES

Sleep quality is associated with wellness, and its assessment can help diagnose several disorders and diseases. Sleep analysis is commonly performed based on self-rating indices, sleep duration, environmental factors, physiologically and polysomnographic-derived parameters, and the occurrence of disorders. However, the correlation that has been observed between the subjective assessment and objective measurements of sleep quality is small. Recently, a few automated systems have been suugested to measure sleep quality to address this challenge. Sleep quality can be assessed by evaluating macrostructure-based sleep analysis via the examination of sleep cycles, namely Rapid Eye Movement (REM) and Non Rapid Eye Movement (NREM) with N1, N2, and N3 stages. However, macrostructure sleep analysis does not consider transitory phenomena like K-complexes and transient fluctuations, which are indispensable in diagnosing various sleep disorders. The CAP, part of the microstructure of sleep, may offer a more precise and relevant examination of sleep and can be considered one of the candidates to measure sleep quality and identify sleep disorders such as insomnia and apnea. CAP is characterized by very subtle changes in the brain's electroencephalogram (EEG) signals that occur during the NREM stage of sleep. The variations among these patterns in healthy subjects and subjects with sleep disorders can be used to identify sleep disorders. Studying CAP is highly arduous for human experts; thus, developing automated systems for assessing CAP is gaining momentum. Developing new techniques for automated CAP detection installed in clinical setups is essential. This paper aims to analyze the algorithms and methods presented in the literature for the automatic assessment of CAP and the development of CAP-based sleep markers that may enhance sleep quality assessment, helping diagnose sleep disorders.

METHODS

This literature survey examined the automated assessment of CAP and related parameters. We have reviewed 34 research articles, including fourteen ML, nine DL, and ten based on some other techniques.

RESULTS

The review includes various algorithms, databases, features, classifiers, and classification performances and their comparisons, advantages, and limitations of automated systems for CAP assessment.

CONCLUSION

A detailed description of state-of-the-art research findings on automated CAP assessment and associated challenges has been presented. Also, the research gaps have been identified based on our review. Further, future research directions are suggested for sleep quality assessment using CAP.

Airborne pollen grain detection from partially labelled data utilising semi-supervised learning

Airborne pollen monitoring has been conducted for more than a century now, as knowledge of the quantity and periodicity of airborne pollen has diverse use cases, like reconstructing historic climates and tracking current climate change, forensic applications, and up to warning those affected by pollen-induced respiratory allergies. Hence, related work on automation of pollen classification already exists. In contrast, detection of pollen is still conducted manually, and it is the gold standard for accuracy. So, here we used a new-generation, automated, near-real-time pollen monitoring sampler, the BAA500, and we used data consisting of both raw and synthesised microscope images. Apart from the automatically generated, commercially-labelled data of all pollen taxa, we additionally used a manually created partial classification test set of bounding boxes and pollen taxa, so as to more accurately evaluate the real-life performance. For the pollen detection, we employed two-stage deep neural network object detectors. We explored a semi-supervised training scheme to remedy the partial labelling. Using a teacher-student approach, the model can add pseudo-labels to complete the labelling during training. To evaluate the performance of our deep learning algorithms and to compare them to the commercial algorithm of the BAA500, we created a manual test set, in which an expert aerobiologist corrected automatically annotated labels. For the novel manual test set, both the supervised and semi-supervised approaches clearly outperform the commercial algorithm with an F1 score of up to 76.9 % compared to 61.3 %. On an automatically created and partially labelled test dataset, we obtain a maximum mAP of 92.7 %. Additional experiments on raw microscope images show comparable performance for the best models, which potentially justifies reducing the complexity of the image generation process. Our results bring automatic pollen monitoring a step forward, as they close the gap in pollen detection performance between manual and automated procedure.

Automated spike and seizure detection: Are we ready for implementation?

OBJECTIVE

Automated detection of spikes and seizures has been a subject of research for several decades now. There have been important advances, yet automated detection in EMU (Epilepsy Monitoring Unit) settings has not been accepted as standard practice. We intend to implement this software at our EMU and so carried out a qualitative study to identify factors that hinder ('barriers') and facilitate ('enablers') implementation.

METHOD

Twenty-two semi-structured interviews were conducted with 14 technicians and neurologists involved in recording and reporting EEGs and eight neurologists who receive EEG reports in the outpatient department. The study was reported according to the Consolidated Criteria for Reporting Qualitative Studies (COREQ).

RESULTS

We identified 14 barriers and 14 enablers for future implementation. Most barriers were reported by technicians. The most prominent barrier was lack of trust in the software, especially regarding seizure detection and false positive results. Additionally, technicians feared losing their EEG review skills or their jobs. Most commonly reported enablers included potential efficiency in the EEG workflow, the opportunity for quantification of EEG findings and the willingness to try the software.

CONCLUSIONS

This study provides insight into the perspectives of users and offers recommendations for implementing automated spike and seizure detection in EMUs.

Automatic detection of glaucoma via fundus imaging and artificial intelligence: A review

Glaucoma is a leading cause of irreversible vision impairment globally, and cases are continuously rising worldwide. Early detection is crucial, allowing timely intervention that can prevent further visual field loss. To detect glaucoma an examination of the optic nerve head via fundus imaging can be performed, at the center of which is the assessment of the optic cup and disc boundaries. Fundus imaging is noninvasive and low-cost; however, image examination relies on subjective, time-consuming, and costly expert assessments. A timely question to ask is: "Can artificial intelligence mimic glaucoma assessments made by experts?" Specifically, can artificial intelligence automatically find the boundaries of the optic cup and disc (providing a so-called segmented fundus image) and then use the segmented image to identify glaucoma with high accuracy? We conducted a comprehensive review on artificial intelligence-enabled glaucoma detection frameworks that produce and use segmented fundus images and summarized the advantages and disadvantages of such frameworks. We identified 36 relevant papers from 2011 to 2021 and 2 main approaches: 1) logical rule-based frameworks, based on a set of rules; and 2) machine learning/statistical modeling-based frameworks. We critically evaluated the state-of-art of the 2 approaches, identified gaps in the literature and pointed at areas for future research.

CMB-HUNT: Automatic detection of cerebral microbleeds using a deep neural network

Cerebral microbleeds (CMBs) are gaining increasing interest due to their importance in diagnosing cerebral small vessel diseases. However, manual inspection of CMBs is time-consuming and prone to human error. Existing automated or semi-automated solutions still have insufficient detection sensitivity and specificity. Furthermore, they frequently use more than one magnetic resonance imaging modality, but these are not always available. The majority of AI-based solutions use either numeric or image data, which may not provide sufficient information about the true nature of CMBs. This paper proposes a deep neural network with multi-type input data for automated CMB detection (CMB-HUNT) using only susceptibility-weighted imaging data (SWI). Combination of SWIs and radiomic-type numerical features allowed us to identify CMBs with high accuracy without the need for additional imaging modalities or complex predictive models. Two independent datasets were used: one with 304 patients (39 with CMBs) for training and internal system validation and one with 61 patients (21 with CMBs) for external validation. For the hold-out testing dataset, CMB-HUNT reached a sensitivity of 90.0%. As results of testing showed, CMB-HUNT outperforms existing methods in terms of the number of FPs per case, which is the lowest reported thus far (0.54 FPs/patient). The proposed system was successfully applied to the independent validation set, reaching a sensitivity of 91.5% with 1.9 false positives per patient and proving its generalization potential. The results were comparable to previous studies. Our research confirms the usefulness of deep learning solutions for CMB detection based only on one MRI modality.

Deep learning-based automatic-bone-destruction-evaluation system using contextual information from other joints

BACKGROUND

X-ray images are commonly used to assess the bone destruction of rheumatoid arthritis. The purpose of this study is to propose an automatic-bone-destruction-evaluation system fully utilizing deep neural networks (DNN). This system detects all target joints of the modified Sharp/van der Heijde score (SHS) from a hand X-ray image. It then classifies every target joint as intact (SHS = 0) or non-intact (SHS ≥ 1).

METHODS

We used 226 hand X-ray images of 40 rheumatoid arthritis patients. As for detection, we used a DNN model called DeepLabCut. As for classification, we built four classification models that classify the detected joint as intact or non-intact. The first model classifies each joint independently, whereas the second model does it while comparing the same contralateral joint. The third model compares the same joint group (e.g., the proximal interphalangeal joints) of one hand and the fourth model compares the same joint group of both hands. We evaluated DeepLabCut's detection performance and classification models' performances. The classification models' performances were compared to three orthopedic surgeons.

RESULTS

Detection rates for all the target joints were 98.0% and 97.3% for erosion and joint space narrowing (JSN). Among the four classification models, the model that compares the same contralateral joint showed the best F-measure (0.70, 0.81) and area under the curve of the precision-recall curve (PR-AUC) (0.73, 0.85) regarding erosion and JSN. As for erosion, the F-measure and PR-AUC of this model were better than the best of the orthopedic surgeons.

CONCLUSIONS

The proposed system was useful. All the target joints were detected with high accuracy. The classification model that compared the same contralateral joint showed better performance than the orthopedic surgeons regarding erosion.

Validation of a machine learning software tool for automated large vessel occlusion detection in patients with suspected acute stroke

PURPOSE

CT angiography (CTA) is the imaging standard for large vessel occlusion (LVO) detection in patients with acute ischemic stroke. StrokeSENS LVO is an automated tool that utilizes a machine learning algorithm to identify anterior large vessel occlusions (LVO) on CTA. The aim of this study was to test the algorithm's performance in LVO detection in an independent dataset.

METHODS

A total of 400 studies (217 LVO, 183 other/no occlusion) read by expert consensus were used for retrospective analysis. The LVO was defined as intracranial internal carotid artery (ICA) occlusion and M1 middle cerebral artery (MCA) occlusion. Software performance in detecting anterior LVO was evaluated using receiver operator characteristics (ROC) analysis, reporting area under the curve (AUC), sensitivity, and specificity. Subgroup analyses were performed to evaluate if performance in detecting LVO differed by subgroups, namely M1 MCA and ICA occlusion sites, and in data stratified by patient age, sex, and CTA acquisition characteristics (slice thickness, kilovoltage tube peak, and scanner manufacturer).

RESULTS

AUC, sensitivity, and specificity overall were as follows: 0.939, 0.894, and 0.874, respectively, in the full cohort; 0.927, 0.857, and 0.874, respectively, in the ICA occlusion cohort; 0.945, 0.914, and 0.874, respectively, in the M1 MCA occlusion cohort. Performance did not differ significantly by patient age, sex, or CTA acquisition characteristics.

CONCLUSION

The StrokeSENS LVO machine learning algorithm detects anterior LVO with high accuracy from a range of scans in a large dataset.

Sedimentary structure discrimination with hyperspectral imaging in sediment cores

Hyperspectral imaging (HSI) is a non-destructive, high-resolution imaging technique that is currently under significant development for analyzing geological areas with remote devices or natural samples in a laboratory. In both cases, the hyperspectral image provides several sedimentary structures that must be separated to temporally and spatially describe the sample. Sediment sequences are composed of successive deposits (strata, homogenite, flood) that are visible depending on sample properties. The classical methods to identify them are time-consuming, have a low spatial resolution (millimeters) and are generally based on naked-eye counting. In this study, we compare several supervised classification algorithms to discriminate sedimentological structures in lake sediments. Instantaneous events in lake sediments are generally linked to extreme geodynamical events (e.g., floods, earthquakes), so their identification and counting are essential to understand long-term fluctuations and improve hazard assessments. Identification and counting are done by reconstructing a chronicle of event layer occurrence, including estimation of deposit thicknesses. Here, we applied two hyperspectral imaging sensors (Visible Near-Infrared, VNIR, 60 μm, 400-1000 nm; Short Wave Infrared, SWIR, 200 μm, 1000-2500 nm) on three sediment cores from different lake systems. We highlight that the SWIR sensor is the optimal one for creating robust classification models with discriminant analyses (prediction accuracies of 0.87-0.98). Indeed, the VNIR sensor is impacted by the surface reliefs and structures that are not in the learning set, which causes mis-classification. These observations are also valid for the combined sensor (VNIR-SWIR) and the RGB images. Several spatial and spectral pre-processing were also compared and enabled one to highlight discriminant information specific to a sample and a sensor. These works show that the combined use of hyperspectral imaging and machine learning improves the characterization of sedimentary structures compared to conventional methods.

Establishing an integrated pipeline for automatic and efficient detection of trace DNA encountered in forensic applications

The analysis of trace DNA is a crucial component in forensic applications. Biological materials containing low-level DNA collected at crime scenes, such as fingerprints, can be valuable as evidence. Automatic detection of biological samples has been largely embraced in forensic applications to meet the increasing throughput requirements. However, the amount of DNA automatically retrieved from trace evidence often tends to be small and unstable, ultimately resulting in poor detection of DNA profiles. Thus, in this work, we introduced a robust DNA extraction and purification platform named Bionewtech® BN3200 (Bionewtech®, Shanghai, China) with the goal of constructing a rapid automatic detection system for trace DNA. The establishment of automatic detection system for trace DNA mainly encompassed two parts: assessing the sensitivity of automatic extraction platform and screening the optimal short tandem repeat (STR) typing kit. The sensitivity of Bionewtech® BN3200 platform based on Ultra-sensitive DNA Extraction kit was initially estimated, demonstrating that this extraction platform might contain large potential in the trace DNA extraction. For the amplification part, three sets of commercial multiplex STR typing kits were selected as candidates, and the amplified products were further genotyped on the Applied Biosystems 3500xl Genetic Analyzer. After comparation, SiFa™ 23 Plex Kit was determined as the most suitable amplification system for trace DNA. Eventually, the newly exploited trace DNA detection system was successfully implemented in the detection of fingerprints derived from glass surfaces with the five-seconds contact time. As a result, the DNA recovered from the fingerprints fluctuated approximately from 57.60 pg to 18.05 ng, in addition, over 70% of the total STR loci were detected in 75% of the fingerprint samples.

Automated seizure detection in an EMU setting: Are software packages ready for implementation?

PURPOSE

We assessed whether automated detection software, combined with live observation, enabled reliable seizure detection using three commercial software packages: Persyst, Encevis and BESA.

METHODS

Two hundred and eighty-six prolonged EEG records of individuals aged 16-86 years, collected between August 2019 and January 2020, were retrospectively processed using all three packages. The reference standard included all seizures mentioned in the clinical report supplemented with true detections made by the software and not previously detected by clinical physiologists. Sensitivity was measured for offline review by clinical physiologists and software seizure detection, both in combination with live monitoring in an EMU setting, for all three software packages at record and seizure level.

RESULTS

The database contained 249 seizures in 64 records. The sensitivity of seizure detection was 98% for Encevis and Persyst, and 95% for BESA, when a positive results was defined as detection at least one of the seizures occurring within an individual record. When positivity was defined as recognition of all seizures, sensitivity was 93% for Persyst, 88% for Encevis and 84% for BESA. Clinical physiologists' review had a sensitivity of 100% at record level and 98% at seizure level. The median false positive rate per record was 1.7 for Persyst, 2.4 for BESA and 5.5 for Encevis per 24 h.

CONCLUSION

Automated seizure detection software does not perform as well as technicians do. However, it can be used in an EMU setting when the user is aware of its weaknesses. This assessment gives future users helpful insight into these strengths and weaknesses. The Persyst software performs best.

Automated spike detection: Which software package?

PURPOSE

We assessed three commercial automated spike detection software packages (Persyst, Encevis and BESA) to see which had the best performance.

METHODS

Thirty prolonged EEG records from people aged at least 16 years were collected and 30-minute representative epochs were selected. Interictal epileptiform discharges (IEDs) were marked by three human experts and by all three software packages. For each 30-minutes selection and for each 10-second epoch we measured whether or not IEDs had occurred. We defined the gold standard as the combined detections of the experts. Kappa scores, sensitivity and specificity were estimated for each software package.

RESULTS

Sensitivity for Persyst in the default setting was 95% for 30-minute selections and 82% for 10-second epochs. Sensitivity for Encevis was 86% (30-minute selections) and 61% (10-second epochs). The specificity for both packages was 88% for 30-minute selections and 96%-99% for the 10-second epochs. Interrater agreement between Persyst and Encevis and the experts was similar than between experts (0.67-0.83 versus 0.63-0.67). Sensitivity for BESA was 40% and specificity 100%. Interrater agreement (0.25) was low.

CONCLUSIONS

IED detection by the Persyst automated software is better than the Encevis and BESA packages, and similar to human review, when reviewing 30-minute selections and 10-second epochs. This findings may help prospective users choose a software package.

Effect of the recording condition on the quality of a single-lead electrocardiogram

Although many wearable single-lead electrocardiogram (ECG) monitoring devices have been developed, information regarding their ECG quality is limited. This study aimed to evaluate the quality of single-lead ECG in healthy subjects under various conditions (body positions and motions) and in patients with arrhythmias, to estimate requirements for automatic analysis, and to identify a way to improve ECG quality by changing the type and placement of electrodes. A single-lead ECG transmitter was placed on the sternum with a pair of electrodes, and ECG was simultaneously recorded with a conventional Holter ECG in 12 healthy subjects under various conditions and 35 patients with arrhythmias. Subjects with arrhythmias were divided into sinus rhythm (SR) and atrial fibrillation (AF) groups. ECG quality was assessed by calculating the sensitivity and positive predictive value (PPV) of the visual detection of QRS complexes (vQRS), automatic detection of QRS complexes (aQRS), and visual detection of P waves (vP). Accuracy was defined as a 100% sensitivity and PPV. We also measured the amplitude of the baseline, P wave, and QRS complex, and calculated the signal-to-noise ratio (SNR). We then focused on aQRS and estimated thresholds to obtain an accurate aQRS in more than 95% of the data. Finally, we sought to improve ECG quality by changing electrode placement using offset-type electrodes in 10 healthy subjects. The single-lead ECG provided 100% accuracy for vQRS, 87% for aQRS, and 74% for vP in healthy subjects under various conditions. Failure for accurate detection occurred in several motions in which the baseline amplitude was increased or in subjects with low QRS or P amplitude, resulting in low SNR. The single-lead ECG provided 97% accuracy for vQRS, 80% for aQRS in patients with arrhythmias, and 95% accuracy for vP in the SR group. The AF group showed higher baseline amplitude than the SR group (0.08 mV vs. 0.02 mV, P < 0.01) but no significant difference in accuracy for aQRS (79% vs. 81%, P = 1.00). The thresholds to obtain an accurate aQRS were a QRS amplitude > 0.42 mV and a baseline amplitude < 0.20 mV. The QRS amplitude was significantly influenced by electrode placement and body position (P < 0.01 for both, two-way analysis of variance), and the maximum reduction by changing body position was estimated as 30% compared to the sitting posture. The QRS amplitude significantly increased when the inter-electrode distance was extended vertically (1.51 mV for vertical extension vs. 0.93 mV for control, P < 0.01). The single-lead ECG provided at least 97% accuracy for vQRS, 80% for aQRS, and 74% for vP. To obtain stable aQRS in any body positions, a QRS amplitude > 0.60 mV and a baseline amplitude < 0.20 mV were required in the sitting posture considering the reduction induced by changing body position. Vertical extension of the inter-electrode distance increased the QRS amplitude.

Deep Learning-Based Automatic Detection of Ellipsoid Zone Loss in Spectral-Domain OCT for Hydroxychloroquine Retinal Toxicity Screening

PURPOSE

Retinal toxicity resulting from hydroxychloroquine use manifests photoreceptor loss and disruption of the ellipsoid zone (EZ) reflectivity band detectable on spectral-domain (SD) OCT imaging. This study investigated whether an automatic deep learning-based algorithm can detect and quantitate EZ loss on SD OCT images with an accuracy comparable with that of human annotations.

DESIGN

Retrospective analysis of data acquired in a prospective, single-center, case-control study.

PARTICIPANTS

Eighty-five patients (168 eyes) who were long-term hydroxychloroquine users (average exposure time, 14 ± 7.2 years).

METHODS

A mask region-based convolutional neural network (M-RCNN) was implemented and trained on individual OCT B-scans. Scan-by-scan detections were aggregated to produce an en face map of EZ loss per 3-dimensional SD OCT volume image. To improve the accuracy and robustness of the EZ loss map, a dual network architecture was proposed that learns to detect EZ loss in parallel using horizontal (horizontal mask region-based convolutional neural network [M-RCNN_H]) and vertical (vertical mask region-based convolutional neural network [M-RCNN_V]) B-scans independently. To quantify accuracy, 10-fold cross-validation was performed.

MAIN OUTCOME MEASURES

Precision, recall, intersection over union (IOU), F1-score metrics, and measured total EZ loss area were compared against human grader annotations and with the determination of toxicity based on the recommended screening guidelines.

RESULTS

The combined projection network demonstrated the best overall performance: precision, 0.90 ± 0.09; recall, 0.88 ± 0.08; and F1 score, 0.89 ± 0.07. The combined model performed superiorly to the M-RCNN_H only model (precision, 0.79 ± 0.17; recall, 0.96 ± 0.04; IOU, 0.78 ± 0.15; and F1 score, 0.86 ± 0.12) and M-RCNN_V only model (precision, 0.71 ± 0.21; recall, 0.94 ± 0.06; IOU, 0.69 ± 0.21; and F1 score, 0.79 ± 0.16). The accuracy was comparable with the variability of human experts: precision, 0.85 ± 0.09; recall, 0.98 ± 0.01; IOU, 0.82 ± 0.12; and F1 score, 0.91 ± 0.06. Automatically generated en face EZ loss maps provide quantitative SD OCT metrics for accurate toxicity determination combined with other functional testing.

CONCLUSIONS

The algorithm can provide a fast, objective, automatic method for measuring areas with EZ loss and can serve as a quantitative assistance tool to screen patients for the presence and extent of toxicity.

Artificial intelligence (AI) real-time detection vs. routine colonoscopy for colorectal neoplasia: a meta-analysis and trial sequential analysis

GOALS AND BACKGROUND

Studies analyzing artificial intelligence (AI) in colonoscopies have reported improvements in detecting colorectal cancer (CRC) lesions, however its utility in the realworld remains limited. In this systematic review and meta-analysis, we evaluate the efficacy of AI-assisted colonoscopies against routine colonoscopy (RC).

STUDY

We performed an extensive search of major databases (through January 2021) for randomized controlled trials (RCTs) reporting adenoma and polyp detection rates. Odds ratio (OR) and standardized mean differences (SMD) with 95% confidence intervals (CIs) were reported. Additionally, trial sequential analysis (TSA) was performed to guard against errors.

RESULTS

Six RCTs were included (4996 participants). The mean age (SD) was 51.99 (4.43) years, and 49% were females. Detection rates favored AI over RC for adenomas (OR 1.77; 95% CI: 1.570-2.08) and polyps (OR 1.91; 95% CI: 1.68-2.16). Secondary outcomes including mean number of adenomas (SMD 0.23; 95% CI: 0.18-0.29) and polyps (SMD 0.23; 95% CI: 0.17-0.29) detected per procedure favored AI. However, RC outperformed AI in detecting pedunculated polyps. Withdrawal times (WTs) favored AI when biopsies were included, while WTs without biopsies, cecal intubation times, and bowel preparation adequacy were similar.

CONCLUSIONS

Colonoscopies equipped with AI detection algorithms could significantly detect previously missed adenomas and polyps while retaining the ability to self-assess and improve periodically. More effective clearance of diminutive adenomas may allow lengthening in surveillance intervals, reducing the burden of surveillance colonoscopies, and increasing its accessibility to those at higher risk. TSA ruled out the risk for false-positive results and confirmed a sufficient sample size to detect the observed effect. Currently, these findings suggest that AI-assisted colonoscopy can serve as a useful proxy to address critical gaps in CRC identification.

The RITHMI study: diagnostic ability of a heart rhythm monitor for automatic detection of atrial fibrillation

INTRODUCTION AND OBJECTIVES

Early detection of atrial fibrillation (AF) is a priority to reduce embolic events by initiating oral anticoagulation therapy. The aim of this study was to evaluate the diagnostic ability of a wrist device designed for automatic AF detection.

METHODS

RITHMI is a prospective, comparative, observational study that included 167 patients referred to a cardiology outpatient clinic for a general consultation or for electrical cardioversion. The study evaluated the ability of a wrist monitor that uses a photoplethysmography (PPG) signal and an electrocardiographic lead to automatically detect AF compared with diagnosis established by 2 cardiologists using the 12-lead electrocardiogram.

RESULTS

The AF detection algorithm based on the PPG signal had a sensitivity of 91% and a specificity of 96% (diagnostic accuracy: 93%). The automatic algorithm based on the electrocardiographic signal had a sensitivity of 94% and a specificity of 96% (diagnostic accuracy: 95%). The 2 algorithms concurred in the diagnosis in 96% of the cases. Overall, the monitor had a sensitivity and specificity of 95% (diagnostic accuracy: 95% and Kappa index: 0.98).

CONCLUSIONS

This study shows that automatic AF detection through the use of a heart rhythm monitor incorporating sensors and algorithms that analyze the PPG signal and the electrocardiographic signal corresponding to lead I is feasible and has high diagnostic accuracy.

MEG detection of high frequency oscillations and intracranial-EEG validation in pediatric epilepsy surgery

OBJECTIVE

To assess the feasibility of automatically detecting high frequency oscillations (HFOs) in magnetoencephalography (MEG) recordings in a group of ten paediatric epilepsy surgery patients who had undergone intracranial electroencephalography (iEEG).

METHODS

A beamforming source-analysis method was used to construct virtual sensors and an automatic algorithm was applied to detect HFOs (80-250 Hz). We evaluated the concordance of MEG findings with the sources of iEEG HFOs, the clinically defined seizure onset zone (SOZ), the location of resected brain structures, and with post-operative outcome.

RESULTS

In 8/9 patients there was good concordance between the sources of MEG HFOs and iEEG HFOs and the SOZ. Significantly more HFOs were detected in iEEG relative to MEG t(71) = 2.85, p < .05. There was good concordance between sources of MEG HFOs and the resected area in patients with good and poor outcome, however HFOs were also detected outside of the resected area in patients with poor outcome.

CONCLUSION

Our findings demonstrate the feasibility of automatically detecting HFOs non-invasively in MEG recordings in paediatric patients, and confirm compatibility of results with invasive recordings.

SIGNIFICANCE

This approach provides support for the non-invasive detection of HFOs to aid surgical planning and potentially reduce the need for invasive monitoring, which is pertinent to paediatric patients.

Automatic and sensitive detection of West Nile virus non-structural protein 1 with a portable SERS-LFIA detector

A portable surface-enhanced Raman scattering (SERS)-lateral flow immunoassay (LFIA) detector has been developed for the automatic and highly sensitive detection of West Nile virus (WNV) non-structural protein 1 (NS1) and actual WNV samples. Au@Ag nanoparticles (Au@Ag NPs) labeled with double-layer Raman molecules were used as SERS tags to prepare WNV-specific SERS-LFIA strips. On this platform, the WNV-specific antigen NS1 protein was quantitatively and sensitively detected. The detection limit for the WNV NS1 protein was 0.1 ng/mL, which was 100-fold more sensitive than visual signals. The detection limit for inactivated WNV virions was 0.2 × 10² copies/μL. The sensitivity of the SERS-LFIA detector was comparable to that of the fluorescence quantitative reverse transcription-polymerase chain reaction assay. The prepared SERS-LFIA strips exhibited high sensitivity and good specificity for WNV. Thus, the strips developed herein have clinical application value. Moreover, the portable SERS-LFIA detector enabled automatic and rapid detection of the SERS-LFIA strips. The platform established herein is expected to make a substantial contribution to the diagnosis and control of outbreaks of emerging infectious diseases, including WNV.

Automatic segmentation of corneal dystrophy on photographic images based on texture analysis

PURPOSE

To develop an automatic algorithm to analyze dystrophic lesions on photographic images of corneal dystrophy.

METHODS

The dataset included 32 images of corneal dystrophy. The dystrophic area was manually segmented twice. Manually labeled dystrophy areas were compared with automatically segmented images. First, we manually removed the light reflex from the image of the cornea. Using an automatic approach, we extracted the brown color of the iris. Then, the program detected the circular region of the pupil and the corneal surface. A whitish dystrophy area was defined based on the image intensity on the iris and the pupil. The sliding square kernel was applied to clearly define the dystrophic region.

RESULTS

For the manual analysis and the twice automatic approach, the Dice similarity was 0.804 and 0.801, respectively. The Pearson correlation coefficient was 0.807 and 0.806, respectively. The total number of distinct dystrophic areas showed no significant difference between the manual and automatic approaches according to the Wilcoxon signed-rank test (p < 0.0001, both).

CONCLUSIONS

We proposed an automatic algorithm for detecting the dystrophy areas on photographic images with an accuracy of approximately 0.80. This system can be applied to detect and predict the progression of corneal dystrophy.

A novel method for automatic pharmacological evaluation of sucrose preference change in depression mice

Sucrose preference test (SPT) is a most frequently applied method for measuring anhedonia, a core symptom of depression, in rodents. However, the method of SPT still remains problematic mainly due to the primitive, irregular, and inaccurate various types of home-made equipment in laboratories, causing imprecise, inconsistent, and variable results. To overcome this issue, we devised a novel method for automatic detection of anhedonia in mice using an electronic apparatus with its program for automated detecting the behavior of drinking of mice instead of manual weighing the water bottles. In this system, the liquid surface of the bottles was monitored electronically by infrared monitoring elements which were assembled beside the plane of the water surface and the information of times and duration of each drinking was collected to the principal machine. A corresponding computer program was written and installed in a computer connected to the principal machine for outputting and analyzing the data. This new method, based on the automated system, was sensitive, reliable, and adaptable for evaluation of stress- or drug-induced anhedonia, as well as taste preference and effects of addictive drugs. Extensive application of this automated apparatus for SPT would greatly improve and standardize the behavioral assessment method of anhedonia, being instrumental in novel antidepressant screening and depression researching.

Automatic wavelet-based assessment of behavioral sleep using multichannel electrocorticography in rats

PURPOSE

During the last decade, the reported prevalence of sleep-disordered breathing in adults has been rapidly increasing. Therefore, automatic methods of sleep assessment are of particular interest. In a framework of translational neuroscience, this study introduces a reliable automatic detection system of behavioral sleep in laboratory rats based on the signal recorded at the cortical surface without requiring electromyography.

METHODS

Experimental data were obtained in 16 adult male WAG/Rij rats at the age of 9 months. Electrocorticographic signals (ECoG) were recorded in freely moving rats during the entire day (22.5 ± 2.2 h). Automatic wavelet-based assessment of behavioral sleep (BS) was proposed. The performance of this wavelet-based method was validated in a group of rats with genetic predisposition to absence epilepsy (n=16) based on visual analysis of their behavior in simultaneously recorded video.

RESULTS

The accuracy of automatic sleep detection was 98% over a 24-h period. An automatic BS assessment method can be adjusted for detecting short arousals during sleep (microarousals) with various duration.

CONCLUSIONS

These findings suggest that automatic wavelet-based assessment of behavioral sleep can be used for assessment of sleep quality. Current analysis indicates a temporal relationship between microarousals, sleep, and epileptic discharges in genetically prone subjects.

An Automatic HFO Detection Method Combining Visual Inspection Features with Multi-Domain Features

As an important promising biomarker, high frequency oscillations (HFOs) can be used to track epileptic activity and localize epileptogenic zones. However, visual marking of HFOs from a large amount of intracranial electroencephalogram (iEEG) data requires a great deal of time and effort from researchers, and is also very dependent on visual features and easily influenced by subjective factors. Therefore, we proposed an automatic epileptic HFO detection method based on visual features and non-intuitive multi-domain features. To eliminate the interference of continuous oscillatory activity in detected sporadic short HFO events, the iEEG signals adjacent to the detected events were set as the neighboring environmental range while the number of oscillations and the peak-valley differences were calculated as the environmental reference features. The proposed method was developed as a MatLab-based HFO detector to automatically detect HFOs in multi-channel, long-distance iEEG signals. The performance of our detector was evaluated on iEEG recordings from epileptic mice and patients with intractable epilepsy. More than 90% of the HFO events detected by this method were confirmed by experts, while the average missed-detection rate was < 10%. Compared with recent related research, the proposed method achieved a synchronous improvement of sensitivity and specificity, and a balance between low false-alarm rate and high detection rate. Detection results demonstrated that the proposed method performs well in sensitivity, specificity, and precision. As an auxiliary tool, our detector can greatly improve the efficiency of clinical experts in inspecting HFO events during the diagnosis and treatment of epilepsy.

Automatic Detection and Analysis of Swallowing Sounds in Healthy Subjects and in Patients with Pharyngolaryngeal Cancer

Assessment of swallowing function is often invasive or involves irradiation. Analysis of swallowing sounds is a noninvasive method for assessment of swallowing but is not used in daily medical practice. Dysphagia could be the first symptom that occurs in head and neck cancer. This study evaluated a method for the automatic detection and analysis of swallowing sounds in healthy subjects and in patients with pharyngolaryngeal cancer. A smartphone application, developed for automatic detection and analysis of swallowing sounds was developed and tested in 12 healthy volunteers and in 26 patients with pharyngolaryngeal cancer. Swallowing sounds were recorded with a laryngophone during a standardized meal (100 mL mashed potatoes, 100 mL water, and 100 mL yogurt). Swallowing number and duration were noted; the results were compared to a standard swallowing sound analysis using the software AUDACITY^®. There were no statistically significant differences in swallowing number or duration between the two analysis methods for the three types of foods in healthy volunteers and only for water in patients. In healthy volunteers, the results of our automatic analysis were comparable with those obtained with the standard analysis. However, a better discrimination of swallowing sounds is necessary for the algorithm to obtain reliable results with thicker food in patients with head and neck cancer.

Clinical Performance of New Software to Automatically Detect Angioectasias in Small Bowel Capsule Endoscopy

Background

Video capsule endoscopy (VCE) revolutionized the diagnosis and management of obscure gastrointestinal bleeding, though the rate of detection of small bowel lesions by the physician is still disappointing. Our group developed a novel algorithm (CMEMS-Uminho) to automatically detect angioectasias which display greater accuracy in VCE static frames than other methods previously published. We aimed to evaluate the algorithm overall performance and assess its diagnostic yield and usability in clinical practice.

Methods

Algorithm overall performance was determined using 54 full-length VCE recordings. To assess its diagnostic yield and usability in clinical practice, 38 VCE examinations with the clinical diagnosis of angioectasias consecutively performed (2017-2018) were evaluated by three physicians with different experiences. The CMEMS-Uminho algorithm was also applied. The performance of the CMEMS-Uminho algorithm was defined by a positive concordance between a frame automatically selected by the software and a study independent capsule endoscopist.

Results

Overall performance in complete VCE recordings was 77.7%, and diagnostic yield was 94.7%. There were significant differences between physicians in regard to global detection rate (p < 0.001), detection rate per capsule (p < 0.001), diagnostic yield (p = 0.007), true positive rate (p < 0.001), time (p < 0.001), and speed viewing (p < 0.001). The application of CMEMS-Uminho algorithm significantly enhanced all readers' global detection rate (p < 0.001) and the differences between them were no longer observed.

Conclusion

The CMEMS-Uminho algorithm detained a good overall performance and was able to enhance physicians' performance, suggesting a potential usability of this tool in clinical practice.

Using sampled visual EEG review in combination with automated detection software at the EMU

PURPOSE

Complete visual review of prolonged video-EEG recordings at an EMU (Epilepsy Monitoring Unit) is time consuming and can cause problems in times of paucity of educated personnel. In this study we aimed to show non inferiority for electroclinical diagnosis using sampled review in combination with EEG analysis softreferware (P13 software, Persyst Corporation), in comparison to complete visual review.

METHOD

Fifty prolonged video-EEG recordings in adults were prospectively evaluated using sampled visual EEG review in combination with automated detection software of the complete EEG record. Visually assessed samples consisted of one hour during wakefulness, one hour during sleep, half an hour of wakefulness after wake-up and all clinical events marked by the individual and/or nurses. The final electro-clinical diagnosis of this new review approach was compared with the electro-clinical diagnosis after complete visual review as presently used.

RESULTS

The electro-clinical diagnosis based on sampled visual review combined with automated detection software did not differ from the diagnosis based on complete visual review. Furthermore, the detection software was able to detect all records containing epileptiform abnormalities and epileptic seizures.

CONCLUSION

Sampled visual review in combination with automated detection using Persyst 13 is non-inferior to complete visual review for electroclinical diagnosis of prolonged video-EEG at an EMU setting, which makes this approach promising.

Automatic detection of cortical arousals in sleep and their contribution to daytime sleepiness

OBJECTIVE

Significant interscorer variability is found in manual scoring of arousals in polysomnographic recordings (PSGs). We propose a fully automatic method, the Multimodal Arousal Detector (MAD), for detecting arousals.

METHODS

A deep neural network was trained on 2,889 PSGs to detect cortical arousals and wakefulness in 1-second intervals. Furthermore, the relationship between MAD-predicted labels on PSGs and next day mean sleep latency (MSL) on a multiple sleep latency test (MSLT), a reflection of daytime sleepiness, was analyzed in 1447 MSLT instances in 873 subjects.

RESULTS

In a dataset of 1,026 PSGs, the MAD achieved an F1 score of 0.76 for arousal detection, while wakefulness was predicted with an accuracy of 0.95. In 60 PSGs scored by nine expert technicians, the MAD performed comparable to four and significantly outperformed five expert technicians for arousal detection. After controlling for known covariates, a doubling of the arousal index was associated with an average decrease in MSL of 40 seconds (p = 0.0075).

CONCLUSIONS

The MAD performed better or comparable to human expert scorers. The MAD-predicted arousals were shown to be significant predictors of MSL.

SIGNIFICANCE

This study validates a fully automatic method for scoring arousals in PSGs.

Automatic detection of perforators for microsurgical reconstruction

The deep inferior epigastric perforator (DIEP) is the most commonly used free flap in mastectomy reconstruction. Preoperative imaging techniques are routinely used to detect location, diameter and course of perforators, with direct intervention from the imaging team, who subsequently draw a chart that will help surgeons choosing the best vascular support for the reconstruction. In this work, the feasibility of using a computer software to support the preoperative planning of 40 patients proposed for breast reconstruction with a DIEP flap is evaluated for the first time. Blood vessel centreline extraction and local characterization algorithms are applied to identify perforators and compared with the manual mapping, aiming to reduce the time spent by the imaging team, as well as the inherent subjectivity to the task. Comparing with the measures taken during surgery, the software calibre estimates were worse for vessels smaller than 1.5 mm (P = 6e-4) but better for the remaining ones (P = 2e-3). Regarding vessel location, the vertical component of the software output was significantly different from the manual measure (P = 0.02), nonetheless that was irrelevant during surgery as errors in the order of 2-3 mm do not have impact in the dissection step. Our trials support that a reduction of the time spent is achievable using the automatic tool (about 2 h/case). The introduction of artificial intelligence in clinical practice intends to simplify the work of health professionals and to provide better outcomes to patients. This pilot study paves the way for a success story.

A Survey on Machine Learning Approaches for Automatic Detection of Voice Disorders

The human voice production system is an intricate biological device capable of modulating pitch and loudness. Inherent internal and/or external factors often damage the vocal folds and result in some change of voice. The consequences are reflected in body functioning and emotional standing. Hence, it is paramount to identify voice changes at an early stage and provide the patient with an opportunity to overcome any ramification and enhance their quality of life. In this line of work, automatic detection of voice disorders using machine learning techniques plays a key role, as it is proven to help ease the process of understanding the voice disorder. In recent years, many researchers have investigated techniques for an automated system that helps clinicians with early diagnosis of voice disorders. In this paper, we present a survey of research work conducted on automatic detection of voice disorders and explore how it is able to identify the different types of voice disorders. We also analyze different databases, feature extraction techniques, and machine learning approaches used in these research works.

Chronset: An automated tool for detecting speech onset

The analysis of speech onset times has a longstanding tradition in experimental psychology as a measure of how a stimulus influences a spoken response. Yet the lack of accurate automatic methods to measure such effects forces researchers to rely on time-intensive manual or semiautomatic techniques. Here we present Chronset, a fully automated tool that estimates speech onset on the basis of multiple acoustic features extracted via multitaper spectral analysis. Using statistical optimization techniques, we show that the present approach generalizes across different languages and speaker populations, and that it extracts speech onset latencies that agree closely with those from human observations. Finally, we show how the present approach can be integrated with previous work (Jansen & Watter Behavior Research Methods, 40:744–751, 2008) to further improve the precision of onset detection. Chronset is publicly available online at www.bcbl.eu/databases/chronset.

Automatic detection and visualisation of MEG ripple oscillations in epilepsy

High frequency oscillations (HFOs, 80–500 Hz) in invasive EEG are a biomarker for the epileptic focus. Ripples (80–250 Hz) have also been identified in non-invasive MEG, yet detection is impeded by noise, their low occurrence rates, and the workload of visual analysis. We propose a method that identifies ripples in MEG through noise reduction, beamforming and automatic detection with minimal user effort. We analysed 15 min of presurgical resting-state interictal MEG data of 25 patients with epilepsy. The MEG signal-to-noise was improved by using a cross-validation signal space separation method, and by calculating ~ 2400 beamformer-based virtual sensors in the grey matter. Ripples in these sensors were automatically detected by an algorithm optimized for MEG. A small subset of the identified ripples was visually checked. Ripple locations were compared with MEG spike dipole locations and the resection area if available. Running the automatic detection algorithm resulted in on average 905 ripples per patient, of which on average 148 ripples were visually reviewed. Reviewing took approximately 5 min per patient, and identified ripples in 16 out of 25 patients. In 14 patients the ripple locations showed good or moderate concordance with the MEG spikes. For six out of eight patients who had surgery, the ripple locations showed concordance with the resection area: 4/5 with good outcome and 2/3 with poor outcome. Automatic ripple detection in beamformer-based virtual sensors is a feasible non-invasive tool for the identification of ripples in MEG. Our method requires minimal user effort and is easily applicable in a clinical setting.

•

Cross-validation signal space separation and beamformer increase the SNR in MEG.

•

Automatic detection of MEG ripples in the time domain is feasible.

•

Our method identifies ripples with minimal user effort and is clinically applicable.

•

Automatically detected ripples are concordant with MEG spikes in 14/16 patients.

•

Automatically detected ripples are concordant with resection area in 6/8 patients.

High-frequency oscillations detected in ECoG recordings correlate with cavernous malformation and seizure-free outcome in a child with focal epilepsy: A case report

Epilepsy associated with cavernous malformation (CM) often requires surgical resection of seizure focus to achieve seizure‐free outcome. High‐frequency oscillations (HFOs) in intracranial electroencephalogram (EEG) are reported as potential biomarkers of epileptogenic regions, but to our knowledge there are no data on the existence of HFOs in CM‐caused epilepsy. Here we report our experience of the identification of the seizure focus in a 3‐year‐old pediatric patient with intractable epilepsy associated with CM. The electrocorticographic recordings were obtained from a 64‐contact grid over 2 days in the epilepsy monitoring unit (EMU). The spatial distribution of HFOs and epileptic spikes were estimated from recording segments right after the electrode placement, during sleep and awake states separately. The HFO distribution showed consistency with the perilesional region; the location of spikes varied over days and did not correlate with the lesion. The HFO spatial distribution was more compact in sleep state and pinpointed the contacts sitting on the CM border. Following the resection of the CM and the hemosiderin ring, the patient became seizure‐free. This is the first report describing HFOs in a pediatric patient with intractable epilepsy associated with CM and shows their potential in identifying the seizure focus.

[Advances in automatic detection technology for images of thin blood film of malaria parasite]

This paper reviews the computer vision and image analysis studies aiming at automated diagnosis or screening of malaria in microscope images of thin blood film smears. On the basis of introducing the background and significance of automatic detection technology, the existing detection technologies are summarized and divided into several steps, including image acquisition, pre-processing, morphological analysis, segmentation, count, and pattern classification components. Then, the principles and implementation methods of each step are given in detail. In addition, the promotion and application in automatic detection technology of thick blood film smears are put forwarded as questions worthy of study, and a perspective of the future work for realization of automated microscopy diagnosis of malaria is provided.

Optimizing detection and analysis of slow waves in sleep EEG

BACKGROUND

Analysis of individual slow waves in EEG recording during sleep provides both greater sensitivity and specificity compared to spectral power measures. However, parameters for detection and analysis have not been widely explored and validated.

NEW METHOD

We present a new, open-source, Matlab based, toolbox for the automatic detection and analysis of slow waves; with adjustable parameter settings, as well as manual correction and exploration of the results using a multi-faceted visualization tool.

RESULTS

We explore a large search space of parameter settings for slow wave detection and measure their effects on a selection of outcome parameters. Every choice of parameter setting had some effect on at least one outcome parameter. In general, the largest effect sizes were found when choosing the EEG reference, type of canonical waveform, and amplitude thresholding.

COMPARISON WITH EXISTING METHOD

Previously published methods accurately detect large, global waves but are conservative and miss the detection of smaller amplitude, local slow waves. The toolbox has additional benefits in terms of speed, user-interface, and visualization options to compare and contrast slow waves.

CONCLUSIONS

The exploration of parameter settings in the toolbox highlights the importance of careful selection of detection METHODS: The sensitivity and specificity of the automated detection can be improved by manually adding or deleting entire waves and or specific channels using the toolbox visualization functions. The toolbox standardizes the detection procedure, sets the stage for reliable results and comparisons and is easy to use without previous programming experience.

Automatic detection of ventilatory modes during invasive mechanical ventilation

BACKGROUND

Expert systems can help alleviate problems related to the shortage of human resources in critical care, offering expert advice in complex situations. Expert systems use contextual information to provide advice to staff. In mechanical ventilation, it is crucial for an expert system to be able to determine the ventilatory mode in use. Different manufacturers have assigned different names to similar or even identical ventilatory modes so an expert system should be able to detect the ventilatory mode. The aim of this study is to evaluate the accuracy of an algorithm to detect the ventilatory mode in use.

METHODS

We compared the results of a two-step algorithm designed to identify seven ventilatory modes. The algorithm was built into a software platform (BetterCare® system, Better Care SL; Barcelona, Spain) that acquires ventilatory signals through the data port of mechanical ventilators. The sample analyzed compared data from consecutive adult patients who underwent >24 h of mechanical ventilation in intensive care units (ICUs) at two hospitals. We used Cohen's kappa statistics to analyze the agreement between the results obtained with the algorithm and those recorded by ICU staff.

RESULTS

We analyzed 486 records from 73 patients. The algorithm correctly labeled the ventilatory mode in 433 (89 %). We found an unweighted Cohen's kappa index of 84.5 % [CI (95 %) = (80.5 %: 88.4 %)].

CONCLUSIONS

The computerized algorithm can reliably identify ventilatory mode.

Automatic detection of noisy channels in fNIRS signal based on correlation analysis

BACKGROUND

fNIRS signals can be contaminated by distinct sources of noise. While most of the noise can be corrected using digital filters, optimized experimental paradigms or pre-processing methods, few approaches focus on the automatic detection of noisy channels.

METHODS

In the present study, we propose a new method that detect automatically noisy fNIRS channels by combining the global correlations of the signal obtained from sliding windows (Cui et al., 2010) with correlation coefficients extracted experimental conditions defined by triggers.

RESULTS

The validity of the method was evaluated on test data from 17 participants, for a total of 16 NIRS channels per subject, positioned over frontal, dorsolateral prefrontal, parietal and occipital areas. Additionally, the detection of noisy channels was tested in the context of different levels of cognitive requirement in a working memory N-back paradigm.

COMPARISON WITH EXISTING METHOD(S)

Bad channels detection accuracy, defined as the proportion of bad NIRS channels correctly detected among the total number of channels examined, was close to 91%. Under different cognitive conditions the area under the Receiver Operating Curve (AUC) increased from 60.5% (global correlations) to 91.2% (local correlations).

CONCLUSIONS

Our results show that global correlations are insufficient for detecting potentially noisy channels when the whole data signal is included in the analysis. In contrast, adding specific local information inherent to the experimental paradigm (e.g., cognitive conditions in a block or event-related design), improved detection performance for noisy channels. Also, we show that automated fNIRS channel detection can be achieved with high accuracy at low computational cost.

Application of the relative wavelet energy to heart rate independent detection of atrial fibrillation

BACKGROUND AND OBJECTIVES

Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia and a growing healthcare burden worldwide. It is often asymptomatic and may appear as episodes of very short duration; hence, the development of methods for its automatic detection is a challenging requirement to achieve early diagnosis and treatment strategies. The present work introduces a novel method exploiting the relative wavelet energy (RWE) to automatically detect AF episodes of a wide variety in length.

METHODS

The proposed method analyzes the atrial activity of the surface electrocardiogram (ECG), i.e., the TQ interval, thus being independent on the ventricular activity. To improve its performance under noisy recordings, signal averaging techniques were applied. The method's performance has been tested with synthesized recordings under different AF variable conditions, such as the heart rate, its variability, the atrial activity amplitude or the presence of noise. Next, the method was tested with real ECG recordings.

RESULTS

Results proved that the RWE provided a robust automatic detection of AF under wide ranges of heart rates, atrial activity amplitudes as well as noisy recordings. Moreover, the method's detection delay proved to be shorter than most of previous works. A trade-off between detection delay and noise robustness was reached by averaging 15 TQ intervals. Under these conditions, AF was detected in less than 7 beats, with an accuracy higher than 90%, which is comparable to previous works.

CONCLUSIONS

Unlike most of previous works, which were mainly based on quantifying the irregular ventricular response during AF, the proposed metric presents two major advantages. First, it can perform successfully even under heart rates with no variability. Second, it consists of a single metric, thus turning its clinical interpretation and real-time implementation easier than previous methods requiring combined indices under complex classifiers.

Automatic detection of high frequency oscillations during epilepsy surgery predicts seizure outcome

OBJECTIVE

High frequency oscillations (HFOs) and in particular fast ripples (FRs) in the post-resection electrocorticogram (ECoG) have recently been shown to be highly specific predictors of outcome of epilepsy surgery. FR visual marking is time consuming and prone to observer bias. We validate here a fully automatic HFO detector against seizure outcome.

METHODS

Pre-resection ECoG dataset (N=14 patients) with visually marked HFOs were used to optimize the detector's parameters in the time-frequency domain. The optimized detector was then applied on a larger post-resection ECoG dataset (N=54) and the output was compared with visual markings and seizure outcome. The analysis was conducted separately for ripples (80-250Hz) and FRs (250-500Hz).

RESULTS

Channel-wise comparison showed a high association between automatic detection and visual marking (p<0.001 for both FRs and ripples). Automatically detected FRs were predictive of clinical outcome with positive predictive value PPV=100% and negative predictive value NPV=62%, while for ripples PPV=43% and NPV=100%.

CONCLUSIONS

Our automatic and fully unsupervised detection of HFO events matched the expert observer's performance in both event selection and outcome prediction.

SIGNIFICANCE

The detector provides a standardized definition of clinically relevant HFOs, which may spread its use in clinical application.

Interaction with slow waves during sleep improves discrimination of physiologic and pathologic high-frequency oscillations (80-500 Hz)

OBJECTIVE

To characterize the interaction between physiologic and pathologic high-frequency oscillations (HFOs) and slow waves during sleep, and to evaluate the practical significance of these interactions by automatically classifying channels as recording from normal or epileptic brain regions.

METHODS

We automatically detected HFOs in intracerebral electroencephalography (EEG) recordings of 45 patients. We characterized the interaction between the HFOs and the amplitude and phase of automatically detected slow waves during sleep. We computed features associated with HFOs, and compared classic features such as rate, amplitude, duration, and frequency to novel features related to the interaction between HFOs and slow waves. To quantify the practical significance of the difference in these features we classified the channels as recording from normal/epileptic regions using logistic regression. We assessed the results in different brain regions to study differences in the HFO characteristics at the lobar level.

RESULTS

We found a clear difference in the coupling between the phase of slow waves during sleep and the occurrence of HFOs. In channels recording physiologic activity, the HFOs tend to occur after the peak of the deactivated state of the slow wave, and in channels with epileptic activity, the HFOs occur more often before this peak. This holds for HFOs in the ripple (80-250 Hz) and fast ripple (250-500 Hz) bands, and different regions of the brain. When using this interaction to automatically classify channels as recording from normal/epileptic brain regions, the performance is better than when using other HFO characteristics. We confirmed differences in the HFO characteristics in mesiotemporal structures and in the occipital lobe.

SIGNIFICANCE

We found the association between slow waves and HFOs to be different in normal and epileptic brain regions, emphasizing their different origin. This is also of practical significance, since it improves the separation between channels recording from normal and epileptic brain regions.

Monitoring burst suppression in critically ill patients: Multi-centric evaluation of a novel method

OBJECTIVE

To develop a computational method to detect and quantify burst suppression patterns (BSP) in the EEGs of critical care patients. A multi-center validation study was performed to assess the detection performance of the method.

METHODS

The fully automatic method scans the EEG for discontinuous patterns and shows detected BSP and quantitative information on a trending display in real-time. The method is designed to work without setting any patient specific parameters and to be insensitive to EEG artifacts and periodic patterns. For validation a total of 3982 h of EEG from 88 patients were analyzed from three centers. Each EEG was annotated by two reviewers to assess the detection performance and the inter-rater agreement.

RESULTS

Average inter-rater agreement between pairs of reviewers was κ=0.69. On average 22% of the review segments included BSP. An average sensitivity of 90% and a specificity of 84% were measured on the consensus annotations of two reviewers. More than 95% of the periodic patterns in the EEGs were correctly suppressed.

CONCLUSION

A fully automatic method to detect burst suppression patterns was assessed in a multi-center study. The method showed high sensitivity and specificity.

SIGNIFICANCE

Clinically applicable burst suppression detection method validated in a large multi-center study.

Detection and Magnetic Source Imaging of Fast Oscillations (40-160 Hz) Recorded with Magnetoencephalography in Focal Epilepsy Patients

We present a framework to detect fast oscillations (FOs) in magnetoencephalography (MEG) and to perform magnetic source imaging (MSI) to determine the location and extent of their generators in the cortex. FOs can be of physiologic origin associated to sensory processing and memory consolidation. In epilepsy, FOs are of pathologic origin and biomarkers of the epileptogenic zone. Seventeen patients with focal epilepsy previously confirmed with identified FOs in scalp electroencephalography (EEG) were evaluated. To handle data deriving from large number of sensors (275 axial gradiometers) we used an automatic detector with high sensitivity. False positives were discarded by two human experts. MSI of the FOs was performed with the wavelet based maximum entropy on the mean method. We found FOs in 11/17 patients, in only one patient the channel with highest FO rate was not concordant with the epileptogenic region and might correspond to physiologic oscillations. MEG FOs rates were very low: 0.02–4.55 per minute. Compared to scalp EEG, detection sensitivity was lower, but the specificity higher in MEG. MSI of FOs showed concordance or partial concordance with proven generators of seizures and epileptiform activity in 10/11 patients. We have validated the proposed framework for the non-invasive study of FOs with MEG. The excellent overall concordance with other clinical gold standard evaluation tools indicates that MEG FOs can provide relevant information to guide implantation for intracranial EEG pre-surgical evaluation and for surgical treatment, and demonstrates the important added value of choosing appropriate FOs detection and source localization methods.

Automatic detection of rhythmic and periodic patterns in critical care EEG based on American Clinical Neurophysiology Society (ACNS) standardized terminology

AIMS OF THE STUDY

Continuous EEG from critical care patients needs to be evaluated time efficiently to maximize the treatment effect. A computational method will be presented that detects rhythmic and periodic patterns according to the critical care EEG terminology (CCET) of the American Clinical Neurophysiology Society (ACNS). The aim is to show that these detected patterns support EEG experts in writing neurophysiological reports.

MATERIALS AND METHODS

First of all, three case reports exemplify the evaluation procedure using graphically presented detections. Second, 187 hours of EEG from 10 critical care patients were used in a comparative trial study. For each patient the result of a review session using the EEG and the visualized pattern detections was compared to the original neurophysiology report.

RESULTS

In three out of five patients with reported seizures, all seizures were reported correctly. In two patients, several subtle clinical seizures with unclear EEG correlation were missed. Lateralized periodic patterns (LPD) were correctly found in 2/2 patients and EEG slowing was correctly found in 7/9 patients. In 8/10 patients, additional EEG features were found including LPDs, EEG slowing, and seizures.

CONCLUSION

The use of automatic pattern detection will assist in review of EEG and increase efficiency. The implementation of bedside surveillance devices using our detection algorithm appears to be feasible and remains to be confirmed in further multicenter studies.

Temporal and spatial characteristics of high frequency oscillations as a new biomarker in epilepsy

OBJECTIVE

Interictal high frequency oscillations (HFOs) are a promising candidate as a biomarker in epilepsy as well as for defining the seizure-onset zone as for the prediction of the surgical outcome after epilepsy surgery. The purpose of the study is to investigate properties of HFOs in long-term recordings with respect to the sleep-wake cycle and anatomic regions to verify previous results based on observations from short intervals and patients mainly with temporal lobe epilepsy to the analysis of hours of recordings and focal epilepsies with extratemporal origin.

METHODS

Automatic HFO detection using a radial basis function neural network detector was performed in long-term recordings of 15 presurgical patients investigated with subdural strip, grid, and depth contacts. Periods with visual marked sleep stages based on parallel scalp recordings from two consecutive nights were compared to awake intervals. Statistical analysis was based on the Kruskal-Wallis test, Mann-Whitney U-test and Spearman's rank correlations.

RESULTS

HFO rates in seizure-onset contacts differed from other brain regions independent of the sleep-wake cycle. For temporal contacts, the HFO rate increased significantly with sleep stage. In addition, contacts covering the parietal lobe, including rolandic cortex, showed a significant increase of HFO rates during sleep. However, no significant HFO rate changes depending on the sleep-wake cycle were found for frontal contacts.

SIGNIFICANCE

The rate of interictal HFOs predicted the SOZ with statistical significance at the group level, but properties other than the HFO rate may need to be considered to improve the diagnostic utility of HFOs. This study gives evidence that the modulation of HFO rates by states of the sleep-wake cycle has particular characteristics within different neocortical regions and in mesiotemporal structures, and contributes to the establishment of HFOs as a biomarker in epilepsy.

Identification of seizure onset zone and preictal state based on characteristics of high frequency oscillations

OBJECTIVE

We investigate the relevance of high frequency oscillations (HFO) for biomarkers of epileptogenic tissue and indicators of preictal state before complex partial seizures in humans.

METHODS

We introduce a novel automated HFO detection method based on the amplitude and features of the HFO events. We examined intracranial recordings from 33 patients and compared HFO rates and characteristics between channels within and outside the seizure onset zone (SOZ). We analyzed changes of HFO activity from interictal to preictal and to ictal periods.

RESULTS

The average HFO rate is higher for SOZ channels compared to non-SOZ channels during all periods. Amplitudes and durations of HFO are higher for events within the SOZ in all periods compared to non-SOZ events, while their frequency is lower. All analyzed HFO features increase for the ictal period.

CONCLUSIONS

HFO may occur in all channels but their rate is significantly higher within SOZ and HFO characteristics differ from HFO outside the SOZ, but the effect size of difference is small.

SIGNIFICANCE

The present results show that based on accumulated dataset it is possible to distinguish HFO features different for SOZ and non-SOZ channels, and to show changes in HFO characteristics during the transition from interictal to preictal and to ictal periods.

Automatic detection of microcalcifications with multi-fractal spectrum

For improving the detection of micro-calcifications (MCs), this paper proposes an automatic detection of MC system making use of multi-fractal spectrum in digitized mammograms. The approach of automatic detection system is based on the principle that normal tissues possess certain fractal properties which change along with the presence of MCs. In this system, multi-fractal spectrum is applied to reveal such fractal properties. By quantifying the deviations of multi-fractal spectrums between normal tissues and MCs, the system can identify MCs altering the fractal properties and finally locate the position of MCs. The performance of the proposed system is compared with the leading automatic detection systems in a mammographic image database. Experimental results demonstrate that the proposed system is statistically superior to most of the compared systems and delivers a superior performance.