Artificial intelligence software for diagnosing intracranial arterial occlusion in patients with acute ischemic stroke

CT perfusion enhances accuracy of intracranial occlusion detection in acute stroke: effect of specialty and experience level

PURPOSE

Detection of intracranial arterial occlusions on CT angiography (CTA) can be challenging. We studied the value of CT perfusion (CTP) for arterial occlusion detection in the anterior circulation amongst radiologists and neurologists, both experienced and less experienced.

METHODS

Seven raters reviewed CTAs of 335 acute stroke patients with and without occlusions. We evaluated occlusion detection with and without CTP. We categorized the occlusions by location. Two experienced raters exposed to all baseline and follow-up imaging defined a consensus gold standard. We calculated sensitivity, specificity and accuracy for occlusion detection with and without CTP and compared the area under the curve (AUC). We also compared the performance of radiologists versus neurologists and of experienced and less experienced raters.

RESULTS

We included 260 patients with ≥1 occlusion and 75 without occlusions. The accuracy of occlusion detection was greater with CTP assistance compared to CTA only (AUC 0.93 vs 0.91, p= 0.03 for proximal and AUC 0.88 vs 0.81, p<0.001 for distal). Distal occlusion detection accuracy improved with CTP in neurologists and in radiologists, whereas improved proximal occlusion detection accuracy was only present in neurologists. Adding CTP improved distal occlusion detection in experienced and less experienced raters. Proximal occlusion detection accuracy improved with CTP in experienced raters, and trended towards improvement in less experienced raters.

CONCLUSION

Assistance of CTP maps may improve the accuracy of intracranial occlusion detection on CTA. In this study, the benefit was most profound for distal occlusions, regardless of experience level or specialty background of the rater.

Applications of Artificial Intelligence in Neurosurgery for Improving Outcomes Through Diagnostics, Predictive Tools, and Resident Education

BACKGROUND

Artificial intelligence (AI) has become an increasingly prominent tool in the field of neurosurgery, revolutionizing various aspects of patient care and surgical practices. AI-powered systems can provide real-time feedback to surgeons, enhancing precision and reducing the risk of complications during surgical procedures. The objective of this study is to review the role of AI in training neurosurgical residents, improving accuracy during surgery, and reducing complications.

METHODS

The literature search method involved searching PubMed using relevant keywords to identify English literature publications, including full texts, and concerning human subject matter from its inception until May 2024, initially generating 247,747 results. Articles were then screened for topic relevancy by abstract contents. Further articles were retrieved from the sources cited by the initially reviewed articles. A comprehensive review was then performed on various studies, including observational studies, case-control studies, cohort studies, clinical trials, meta-analyses, and reviews by 4 reviewers individually and then collectively.

RESULTS

Studies on AI in neurosurgery reach more than 4000 produced over a decade alone. The majority of studies regarding clinical diagnosis, risk prediction, and intraoperative guidance remain retrospective in nature. In its current form, AI-based paradigm performed inferiorly to neurosurgery residents in test taking.

CONCLUSIONS

AI has potential for broad applications in neurosurgery from use as a diagnostic, predictive, intraoperative, or educational tool. Further research is warranted for prospective use of AI-based technology for delivery of neurosurgical care.

Advancing personalised care in atrial fibrillation and stroke: The potential impact of AI from prevention to rehabilitation

Atrial fibrillation (AF) is a complex condition caused by various underlying pathophysiological disorders and is the most common heart arrhythmia worldwide, affecting 2 % of the European population. This prevalence increases with age, imposing significant financial, economic, and human burdens. In Europe, stroke is the second leading cause of death and the primary cause of disability, with numbers expected to rise due to ageing and improved survival rates. Functional recovery from AF-related stroke is often unsatisfactory, leading to prolonged hospital stays, severe disability, and high mortality. Despite advances in AF and stroke research, the full pathophysiological and management issues between AF and stroke increasingly need innovative approaches such as artificial intelligence (AI) and machine learning (ML). Current risk assessment tools focus on static risk factors, neglecting the dynamic nature of risk influenced by acute illness, ageing, and comorbidities. Incorporating biomarkers and automated ECG analysis could enhance pathophysiological understanding. This paper highlights the need for personalised, integrative approaches in AF and stroke management, emphasising the potential of AI and ML to improve risk prediction, treatment personalisation, and rehabilitation outcomes. Further research is essential to optimise care and reduce the burden of AF and stroke on patients and healthcare systems.

A multicenter validation and calibration of automated software package for detecting anterior circulation large vessel occlusion on CT angiography

PURPOSE

To validate JLK-LVO, a software detecting large vessel occlusion (LVO) on computed tomography angiography (CTA), within a multicenter dataset.

METHODS

From 2021 to 2023, we enrolled patients with ischemic stroke who underwent CTA within 24-hour of onset at six university hospitals for validation and calibration datasets and at another university hospital for an independent dataset for testing model calibration. The diagnostic performance was evaluated using area under the curve (AUC), sensitivity, and specificity across the entire study population and specifically in patients with isolated middle cerebral artery (MCA)-M2 occlusion. We calibrated LVO probabilities using logistic regression and by grouping LVO probabilities based on observed frequency.

RESULTS

After excluding 168 patients, 796 remained; the mean (SD) age was 68.9 (13.7) years, and 57.7% were men. LVO was present in 193 (24.3%) of patients, and the median interval from last-known-well to CTA was 5.7 h (IQR 2.5-12.1 h). The software achieved an AUC of 0.944 (95% CI 0.926-0.960), with a sensitivity of 89.6% (84.5-93.6%) and a specificity of 90.4% (87.7-92.6%). In isolated MCA-M2 occlusion, the AUROC was 0.880 (95% CI 0.824-0.921). Due to sparse data between 20 and 60% of LVO probabilities, recategorization into unlikely (0-20% LVO scores), less likely (20-60%), possible (60-90%), and suggestive (90-100%) provided a reliable estimation of LVO compared with mathematical calibration. The category of LVO probabilities was associated with follow-up infarct volumes and functional outcome.

CONCLUSION

In this multicenter study, we proved the clinical efficacy of the software in detecting LVO on CTA.

Introduction and accuracy assessment of Nicolab's StrokeViewer in a developing stroke thrombectomy UK service. a service development/improvement project

AIM

The aim of this study was to evaluate the implementation of artificial intelligence (AI) software in a quaternary stroke centre as well as assess the accuracy and efficacy of StrokeViewer software in large vessel occlusion detection and its potential impact on radiological workflow.

MATERIALS AND METHODS

Data were collected during two separate three-month periods comparing the accuracy rate of StrokeViewer in detection of large vessel occlusion to that of a junior registrar. During the first three months, 37 cases were identified and during the second leg, 47. The second leg of the study was performed due to a high number of technical failures during the first one and in an attempt to improve those via communication with the manufacturer and co-operation between allied healthcare professionals. Statistical analysis was performed using SPSS software.

RESULTS

Technical failure rate was 25% in the first leg and reduced to 17% in the second leg, showing a trend to statistical significance. Specificity and sensitivity of StrokeViewer were similar in the two legs of the study, measuring 91% and 93% initially and 94% and 93% finally, respectively. Efficacy was comparable to that of the junior registrar with StrokeViewer, demonstrating 92% accuracy during the first leg vs 95% by the junior registrar and 93% in the second leg vs 98% by the junior registrar. These did not show statistical significance.

CONCLUSION

This is a real-life analysis of StrokeViewer efficacy and its potential failures, showing a reduction in failure rate, accuracy rate of a junior registrar, and sensitivity and specificity values close to the advertised ones.

Current Stroke Solutions Using Artificial Intelligence: A Review of the Literature

INTRODUCTION

In recent years, artificial intelligence (AI) has emerged as a transformative tool for enhancing stroke diagnosis, aiding treatment decision making, and improving overall patient care. Leading AI-driven platforms such as RapidAI, Brainomix®, and Viz.ai have been developed to assist healthcare professionals in the swift and accurate assessment of stroke patients.

METHODS

Following the PRISMA guidelines, a comprehensive systematic review was conducted using PubMed, Embase, Web of Science, and Scopus. Characteristic descriptive measures were gathered as appropriate from all included studies, including the sensitivity, specificity, accuracy, and comparison of the available tools.

RESULTS

A total of 31 studies were included, of which 29 studies focused on detecting acute ischemic stroke (AIS) or large vessel occlusions (LVOs), and 2 studies focused on hemorrhagic strokes. The four main tools used were Viz.ai, RapidAI, Brainomix®, and deep learning modules.

CONCLUSIONS

AI tools in the treatment of stroke have demonstrated usefulness for diagnosing different stroke types, providing high levels of accuracy and helping to make quicker and more precise clinical judgments.

Automated detection of large vessel occlusion using deep learning: a pivotal multicenter study and reader performance study

BACKGROUND

To evaluate the stand-alone efficacy and improvements in diagnostic accuracy of early-career physicians of the artificial intelligence (AI) software to detect large vessel occlusion (LVO) in CT angiography (CTA).

METHODS

This multicenter study included 595 ischemic stroke patients from January 2021 to September 2023. Standard references and LVO locations were determined by consensus among three experts. The efficacy of the AI software was benchmarked against standard references, and its impact on the diagnostic accuracy of four residents involved in stroke care was assessed. The area under the receiver operating characteristic curve (AUROC), sensitivity, and specificity of the software and readers with versus without AI assistance were calculated.

RESULTS

Among the 595 patients (mean age 68.5±13.4 years, 56% male), 275 (46.2%) had LVO. The median time interval from the last known well time to the CTA was 46.0 hours (IQR 11.8-64.4). For LVO detection, the software demonstrated a sensitivity of 0.858 (95% CI 0.811 to 0.897) and a specificity of 0.969 (95% CI 0.943 to 0.985). In subjects whose symptom onset to imaging was within 24 hours (n=195), the software exhibited an AUROC of 0.973 (95% CI 0.939 to 0.991), a sensitivity of 0.890 (95% CI 0.817 to 0.936), and a specificity of 0.965 (95% CI 0.902 to 0.991). Reading with AI assistance improved sensitivity by 4.0% (2.17 to 5.84%) and AUROC by 0.024 (0.015 to 0.033) (all P<0.001) compared with readings without AI assistance.

CONCLUSIONS

The AI software demonstrated a high detection rate for LVO. In addition, the software improved diagnostic accuracy of early-career physicians in detecting LVO, streamlining stroke workflow in the emergency room.

A deep learning method to identify and localize large-vessel occlusions from cerebral digital subtraction angiography

BACKGROUND AND PURPOSE

An essential step during endovascular thrombectomy is identifying the occluded arterial vessel on a cerebral digital subtraction angiogram (DSA). We developed an algorithm that can detect and localize the position of occlusions in cerebral DSA.

METHODS

We retrospectively collected cerebral DSAs from a single institution between 2018 and 2020 from 188 patients, 86 of whom suffered occlusions of the M1 and proximal M2 segments. We trained an ensemble of deep-learning models on fewer than 60 large-vessel occlusion (LVO)-positive patients. We evaluated the model on an independent test set and evaluated the truth of its predicted localizations using Intersection over Union and expert review.

RESULTS

On an independent test set of 166 cerebral DSA frames with an LVO prevalence of 0.19, the model achieved a specificity of 0.95 (95% confidence interval [CI]: 0.90, 0.99), a precision of 0.7450 (95% CI: 0.64, 0.88), and a sensitivity of 0.76 (95% CI: 0.66, 0.91). The model correctly localized the LVO in at least one frame in 13 of the 14 LVO-positive patients in the test set. The model achieved a precision of 0.67 (95% CI: 0.52, 0.79), recall of 0.69 (95% CI: 0.46, 0.81), and a mean average precision of 0.75 (95% CI: 0.56, 0.91).

CONCLUSION

This work demonstrates that a deep learning strategy using a limited dataset can generate effective representations used to identify LVOs. Generating an expanded and more complete dataset of LVOs with obstructed LVOs is likely the best way to improve the model's ability to localize LVOs.

Current and future trends in acute ischemic stroke treatment: direct-to-angiography suite, middle vessel occlusion, large core, and minor strokes

Since the publication of the landmark thrombectomy trials in 2015, the field of endovascular therapy for ischemic stroke has been rapidly growing. The very low number needed to treat to provide functional benefits shown by the initial randomized trials has led clinicians and investigators to seek to translate the benefits of endovascular therapy to other patient subgroups. Even if the treatment effect is diminished, currently available data has provided sufficient information to extend endovascular therapy to large infarct core patients. Recently, published data have also shown that sophisticated imaging is not necessary for late time- window patients. As a result, further research into patient selection and the stroke pathway now focuses on dramatically reducing door-to-groin times and improving outcomes by circumventing classical imaging paradigms altogether and employing a direct-to-angio suite approach for selected large vessel occlusion patients in the early time window. While the results of this approach mainly concern patients with severe deficits, there are further struggles to provide evidence of the efficacy and safety of endovascular treatment in minor stroke and large vessel occlusion, as well as in patients with middle vessel occlusions. The current lack of good quality data regarding these patients provides significant challenges for accurately selecting potential candidates for endovascular treatment. However, current and future randomized trials will probably elucidate the efficacy of endovascular treatment in these patient populations.

Clinical applications of artificial intelligence in radiology

The rapid growth of medical imaging has placed increasing demands on radiologists. In this scenario, artificial intelligence (AI) has become an attractive partner, one that may complement case interpretation and may aid in various non-interpretive aspects of the work in the radiological clinic. In this review, we discuss interpretative and non-interpretative uses of AI in the clinical practice, as well as report on the barriers to AI's adoption in the clinic. We show that AI currently has a modest to moderate penetration in the clinical practice, with many radiologists still being unconvinced of its value and the return on its investment. Moreover, we discuss the radiologists' liabilities regarding the AI decisions, and explain how we currently do not have regulation to guide the implementation of explainable AI or of self-learning algorithms.

Duration and accuracy of automated stroke CT workflow with AI-supported intracranial large vessel occlusion detection

The Automation Platform (AP) is a software platform to support the workflow of radiologists and includes a stroke CT package with integrated artificial intelligence (AI) based tools. The aim of this study was to evaluate the diagnostic performance of the AP for the detection of intracranial large vessel occlusions (LVO) on conventional CT angiography (CTA), and the duration of CT processing in a cohort of acute stroke patients. The diagnostic performance for intracranial LVO detection on CTA by the AP was evaluated in a retrospective cohort of 100 acute stroke patients and compared to the diagnostic performance of five radiologists with different levels of experience. The reference standard was set by an independent neuroradiologist, with access to the readings of the different radiologists, clinical data, and follow-up. The data processing time of the AP for ICH detection on non-contrast CT, LVO detection on CTA, and the processing of CTP maps was assessed in a subset 60 patients of the retrospective cohort. This was compared to 13 radiologists, who were prospectively timed for the processing and reading of 21 stroke CTs. The AP showed shorter processing time of CTA (mean 60 versus 395 s) and CTP (mean 196 versus 243-349 s) as compared to radiologists, but showed lower sensitivity for LVO detection (sensitivity 77% of the AP vs mean sensitivity 87% of radiologists). If the AP would have been used as a stand-alone system, 1 ICA occlusion, 2 M1 occlusions and 8 M2 occlusions would have been missed, which would be eligible for mechanical thrombectomy. In conclusion, the AP showed shorter processing time of CTA and CTP as compared with radiologists, which illustrates the potential of the AP to speed-up the diagnostic work-up. However, its performance for LVO detection was lower as compared with radiologists, especially for M2 vessel occlusions.

AI-support for the detection of intracranial large vessel occlusions: One-year prospective evaluation

Purpose

Few studies have evaluated real-world performance of radiological AI-tools in clinical practice. Over one-year, we prospectively evaluated the use of AI software to support the detection of intracranial large vessel occlusions (LVO) on CT angiography (CTA).

Method

Quantitative measures (user log-in attempts, AI standalone performance) and qualitative data (user surveys) were reviewed by a key-user group at three timepoints. A total of 491 CTA studies of 460 patients were included for analysis.

Results

The overall accuracy of the AI-tool for LVO detection and localization was 87.6%, sensitivity 69.1% and specificity 91.2%. Out of 81 LVOs, 31 of 34 (91%) M1 occlusions were detected correctly, 19 of 38 (50%) M2 occlusions, and 6 of 9 (67%) ICA occlusions. The product was considered user-friendly. The diagnostic confidence of the users for LVO detection remained the same over the year. The last measured net promotor score was -56%. The use of the AI-tool fluctuated over the year with a declining trend.

Conclusions

Our pragmatic approach of evaluating the AI-tool used in clinical practice, helped us to monitor the usage, to estimate the perceived added value by the users of the AI-tool, and to make an informed decision about the continuation of the use of the AI-tool.

Automated detection of intracranial large vessel occlusions using Viz.ai software: Experience in a large, integrated stroke network

BACKGROUND AND PURPOSE

Endovascular thrombectomy is an evidence-based treatment for large vessel occlusion (LVO) stroke. Commercially available artificial intelligence has been designed to detect the presence of an LVO on computed tomography angiogram (CTA). We compared Viz.ai-LVO (San Francisco, CA, USA) to CTA interpretation by board-certified neuroradiologists (NRs) in a large, integrated stroke network.

METHODS

From January 2021 to December 2021, we compared Viz.ai detection of an internal carotid artery (ICA) or middle cerebral artery first segment (MCA-M1) occlusion to the gold standard of CTA interpretation by board-certified NRs for all code stroke CTAs. On a monthly basis, sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) were calculated. Trend analyses were conducted to evaluate for any improvement of LVO detection by the software over time.

RESULTS

3851 patients met study inclusion criteria, of whom 220 (5.7%) had an ICA or MCA-M1 occlusion per NR. Sensitivity and specificity were 78.2% (95% CI 72%-83%) and 97% (95% CI 96%-98%), respectively. PPV was 61% (95% CI 55%-67%), NPV 99% (95% CI 98%-99%), and accuracy was 95.9% (95% CI 95.3%-96.5%). Neither specificity or sensitivity improved over time in the trend analysis.

CONCLUSIONS

Viz.ai-LVO has high specificity and moderately high sensitivity to detect an ICA or proximal MCA occlusion. The software has the potential to streamline code stroke workflows and may be particularly impactful when emergency access to NRs or vascular neurologists is limited.

Agreement between estimated computed tomography perfusion ischemic core and follow-up infarct on diffusion-weighted imaging

BACKGROUND

Computed tomography perfusion (CTP) is frequently performed during the diagnostic workup of acute ischemic stroke patients. Yet, ischemic core estimates vary widely between different commercially available software packages. We assessed the volumetric and spatial agreement of the ischemic core on CTP with the follow-up infarct on diffusion-weighted imaging (DWI) using an automated software.

METHODS

We included successfully reperfused patients who underwent endovascular treatment (EVT) with CTP and follow-up DWI between November 2017 and September 2020. CTP data were processed with a fully automated software using relative cerebral blood flow (rCBF) < 30% to estimate the ischemic core. The follow-up infarct was segmented on DWI imaging data, which were acquired at approximately 24 h. Ischemic core on CTP was compared with the follow-up infarct lesion on DWI using intraclass correlation coefficient (ICC) and Dice similarity coefficient (Dice).

RESULTS

In 59 patients, the median estimated core volume on CTP was 16 (IQR 8-47) mL. The follow-up infarct volume on DWI was 11 (IQR 6-42) mL. ICC was 0.60 (95% CI 0.33-0.76), indicating moderate volumetric agreement. Median Dice was 0.20 (IQR 0.01-0.35). The median positive predictive value was 0.24 (IQR 0.05-0.57), and the median sensitivity was 0.3 (IQR 0.13-0.47). Severe core overestimation on computed tomography perfusion > 50 mL occurred in 4/59 (7%) of the cases.

CONCLUSIONS

In patients with successful reperfusion after EVT, CTP-estimated ischemic core showed moderate volumetric and spatial agreement with the follow-up infarct lesion on DWI, similar to the most used commercially available CTP software packages. Severe ischemic core overestimation was relatively uncommon.

Image level detection of large vessel occlusion on 4D-CTA perfusion data using deep learning in acute stroke

OBJECTIVES

Automated image-level detection of large vessel occlusions (LVO) could expedite patient triage for mechanical thrombectomy. A few studies have previously attempted LVO detection using artificial intelligence (AI) on CT angiography (CTA) images. To our knowledge this is the first study to detect LVO existence and location on raw 4D-CTA/ CT perfusion (CTP) images using neural network (NN) models.

MATERIALS AND METHODS

Retrospective study using data from a level-I stroke center was performed. A total of 306 (187 with LVO, and 119 without) patients were evaluated. Image pre-processing included co-registration, normalization and skull stripping. Five consecutive time-points for each patient were selected to provide variable contrast density in data. Additional data augmentation included rotation and horizonal image flipping. Our model architecture consisted of two neural networks, first for classification (based on hemispheric asymmetry), followed by second model for exact site of LVO detection. Only cases deemed positive by the classification model were routed to the detection model, thereby reducing false positives and improving specificity. The results were compared with expert annotated LVO detection.

RESULTS

Using a 80:20 split for training and validation, the combination of both classification and detection model achieved a sensitivity of 86.5%, a specificity of 89.5%, and an accuracy of 87.5%. A 5-fold cross-validation using the entire data achieved a mean sensitivity of 82.7%, a specificity of 89.8%, and an accuracy of 85.5% and a mean AUC of 0.89 (95% CI: 0.85-0.93).

CONCLUSION

Our findings suggest that accurate image-level LVO detection is feasible on CTP raw images.