Computer-aided diagnosis: how to move from the laboratory to the clinic

Predictive quantitative sonographic features on classification of hot and cold thyroid nodules

PURPOSE

This study investigated the potentiality of ultrasound imaging to classify hot and cold thyroid nodules on the basis of textural and morphological analysis.

METHODS

In this research, 42 hypo (hot) and 42 hyper-function (cold) thyroid nodules were evaluated through the proposed method of computer aided diagnosis (CAD) system. To discover the difference between hot and cold nodules, 49 sonographic features (9 morphological, 40 textural) were extracted. A support vector machine classifier was utilized for the classification of LNs based on their extracted features.

RESULTS

In the training set data, a combination of morphological and textural features represented the best performance with area under the receiver operating characteristic curve (AUC) of 0.992. Upon testing the data set, the proposed model could classify the hot and cold thyroid nodules with an AUC of 0.948.

CONCLUSIONS

CAD method based on textural and morphological features is capable of distinguishing between hot from cold nodules via 2-Dimensional sonography. Therefore, it can be used as a supplementary technique in daily clinical practices to improve the radiologists' understanding of conventional ultrasound imaging for nodules characterization.

Pulmonary quantitative CT imaging in focal and diffuse disease: current research and clinical applications

The frenetic development of imaging technology-both hardware and software-provides exceptional potential for investigation of the lung. In the last two decades, CT was exploited for detailed characterization of pulmonary structures and description of respiratory disease. The introduction of volumetric acquisition allowed increasingly sophisticated analysis of CT data by means of computerized algorithm, namely quantitative CT (QCT). Hundreds of thousands of CTs have been analysed for characterization of focal and diffuse disease of the lung. Several QCT metrics were developed and tested against clinical, functional and prognostic descriptors. Computer-aided detection of nodules, textural analysis of focal lesions, densitometric analysis and airway segmentation in obstructive pulmonary disease and textural analysis in interstitial lung disease are the major chapters of this discipline. The validation of QCT metrics for specific clinical and investigational needs prompted the translation of such metrics from research field to patient care. The present review summarizes the state of the art of QCT in both focal and diffuse lung disease, including a dedicated discussion about application of QCT metrics as parameters for clinical care and outcomes in clinical trials.

Radiomics in paediatric neuro-oncology: A multicentre study on MRI texture analysis

Brain tumours are the most common solid cancers in children in the UK and are the most common cause of cancer deaths in this age group. Despite current advances in MRI, non-invasive diagnosis of paediatric brain tumours has yet to find its way into routine clinical practice. Radiomics, the high-throughput extraction and analysis of quantitative image features (e.g. texture), offers potential solutions for tumour characterization and decision support. In the search for diagnostic oncological markers, the primary aim of this work was to study the application of MRI texture analysis (TA) for the classification of paediatric brain tumours. A multicentre study was carried out, within a supervised classification framework, on clinical MR images, and a support vector machine (SVM) was trained with 3D textural attributes obtained from conventional MRI. To determine the cross-centre transferability of TA, an assessment of how SVM performs on unseen datasets was carried out through rigorous pairwise testing. The study also investigated the nature of features that are most likely to train classifiers that can generalize well with the data. Finally, the issue of class imbalance, which arises due to some tumour types being more common than others, was explored. For each of the tests carried out through pairwise testing, the optimal area under the receiver operating characteristic curve ranged between 76% and 86%, suggesting that the model was able to capture transferable tumour information. Feature selection results suggest that similar aspects of tumour texture are enhanced by MR images obtained at different hospitals. Our results also suggest that the availability of equally represented classes has enabled SVM to better characterize the data points. The findings of the study presented here support the use of 3D TA on conventional MR images to aid diagnostic classification of paediatric brain tumours.

Promoting Collaborations Between Radiologists and Scientists

Radiology as a discipline thrives on the dynamic interplay between technological and clinical advances. Progress in almost all facets of the imaging sciences is highly dependent on complex tools sourced from physics, engineering, biology, and the clinical sciences to obtain, process, and view imaging studies. The application of these tools, however, requires broad and deep medical knowledge about disease pathophysiology and its relationship with medical imaging. This relationship between clinical medicine and imaging technology, nurtured and fostered over the past 75 years, has cultivated extraordinarily rich collaborative opportunities between basic scientists, engineers, and physicians. In this review, we attempt to provide a framework to identify both currently successful collaborative ventures and future opportunities for scientific partnership. This invited review is a product of a special working group within the Association of University Radiologists-Radiology Research Alliance.

Artificial intelligence and deep learning - Radiology's next frontier?

Tracing the use of computers in the radiology department from administrative functions through image acquisition, storage, and reporting, to early attempts at improved diagnosis, we begin to imagine possible new frontiers for their use in exam interpretation. Given their initially slow but ultimately substantial progress in the noninterpretive areas, we are left desiring and even expecting more in the interpretation realm. New technological advances may provide the next wave of progress and radiologists should be early adopters. Several potential applications are discussed and hopefully will serve to inspire future progress.

A fully automatic end-to-end method for content-based image retrieval of CT scans with similar liver lesion annotations

PURPOSE

The goal of medical content-based image retrieval (M-CBIR) is to assist radiologists in the decision-making process by retrieving medical cases similar to a given image. One of the key interests of radiologists is lesions and their annotations, since the patient treatment depends on the lesion diagnosis. Therefore, a key feature of M-CBIR systems is the retrieval of scans with the most similar lesion annotations. To be of value, M-CBIR systems should be fully automatic to handle large case databases.

METHODS

We present a fully automatic end-to-end method for the retrieval of CT scans with similar liver lesion annotations. The input is a database of abdominal CT scans labeled with liver lesions, a query CT scan, and optionally one radiologist-specified lesion annotation of interest. The output is an ordered list of the database CT scans with the most similar liver lesion annotations. The method starts by automatically segmenting the liver in the scan. It then extracts a histogram-based features vector from the segmented region, learns the features' relative importance, and ranks the database scans according to the relative importance measure. The main advantages of our method are that it fully automates the end-to-end querying process, that it uses simple and efficient techniques that are scalable to large datasets, and that it produces quality retrieval results using an unannotated CT scan.

RESULTS

Our experimental results on 9 CT queries on a dataset of 41 volumetric CT scans from the 2014 Image CLEF Liver Annotation Task yield an average retrieval accuracy (Normalized Discounted Cumulative Gain index) of 0.77 and 0.84 without/with annotation, respectively.

CONCLUSIONS

Fully automatic end-to-end retrieval of similar cases based on image information alone, rather that on disease diagnosis, may help radiologists to better diagnose liver lesions.

A machine learning approach to investigate the relationship between shape features and numerically predicted risk of ascending aortic aneurysm

Geometric features of the aorta are linked to patient risk of rupture in the clinical decision to electively repair an ascending aortic aneurysm (AsAA). Previous approaches have focused on relationship between intuitive geometric features (e.g., diameter and curvature) and wall stress. This work investigates the feasibility of a machine learning approach to establish the linkages between shape features and FEA-predicted AsAA rupture risk, and it may serve as a faster surrogate for FEA associated with long simulation time and numerical convergence issues. This method consists of four main steps: (1) constructing a statistical shape model (SSM) from clinical 3D CT images of AsAA patients; (2) generating a dataset of representative aneurysm shapes and obtaining FEA-predicted risk scores defined as systolic pressure divided by rupture pressure (rupture is determined by a threshold criterion); (3) establishing relationship between shape features and risk by using classifiers and regressors; and (4) evaluating such relationship in cross-validation. The results show that SSM parameters can be used as strong shape features to make predictions of risk scores consistent with FEA, which lead to an average risk classification accuracy of 95.58% by using support vector machine and an average regression error of 0.0332 by using support vector regression, while intuitive geometric features have relatively weak performance. Compared to FEA, this machine learning approach is magnitudes faster. In our future studies, material properties and inhomogeneous thickness will be incorporated into the models and learning algorithms, which may lead to a practical system for clinical applications.

Classifying symmetrical differences and temporal change for the detection of malignant masses in mammography using deep neural networks

We investigate the addition of symmetry and temporal context information to a deep convolutional neural network (CNN) with the purpose of detecting malignant soft tissue lesions in mammography. We employ a simple linear mapping that takes the location of a mass candidate and maps it to either the contralateral or prior mammogram, and regions of interest (ROIs) are extracted around each location. Two different architectures are subsequently explored: (1) a fusion model employing two datastreams where both ROIs are fed to the network during training and testing and (2) a stagewise approach where a single ROI CNN is trained on the primary image and subsequently used as a feature extractor for both primary and contralateral or prior ROIs. A "shallow" gradient boosted tree classifier is then trained on the concatenation of these features and used to classify the joint representation. The baseline yielded an AUC of 0.87 with confidence interval [0.853, 0.893]. For the analysis of symmetrical differences, the first architecture where both primary and contralateral patches are presented during training obtained an AUC of 0.895 with confidence interval [0.877, 0.913], and the second architecture where a new classifier is retrained on the concatenation an AUC of 0.88 with confidence interval [0.859, 0.9]. We found a significant difference between the first architecture and the baseline at high specificity with [Formula: see text]. When using the same architectures to analyze temporal change, we yielded an AUC of 0.884 with confidence interval [0.865, 0.902] for the first architecture and an AUC of 0.879 with confidence interval [0.858, 0.898] in the second setting. Although improvements for temporal analysis were consistent, they were not found to be significant. The results show our proposed method is promising and we suspect performance can greatly be improved when more temporal data become available.

A deep learning framework for supporting the classification of breast lesions in ultrasound images

In this research, we exploited the deep learning framework to differentiate the distinctive types of lesions and nodules in breast acquired with ultrasound imaging. A biopsy-proven benchmarking dataset was built from 5151 patients cases containing a total of 7408 ultrasound breast images, representative of semi-automatically segmented lesions associated with masses. The dataset comprised 4254 benign and 3154 malignant lesions. The developed method includes histogram equalization, image cropping and margin augmentation. The GoogLeNet convolutionary neural network was trained to the database to differentiate benign and malignant tumors. The networks were trained on the data with augmentation and the data without augmentation. Both of them showed an area under the curve of over 0.9. The networks showed an accuracy of about 0.9 (90%), a sensitivity of 0.86 and a specificity of 0.96. Although target regions of interest (ROIs) were selected by radiologists, meaning that radiologists still have to point out the location of the ROI, the classification of malignant lesions showed promising results. If this method is used by radiologists in clinical situations it can classify malignant lesions in a short time and support the diagnosis of radiologists in discriminating malignant lesions. Therefore, the proposed method can work in tandem with human radiologists to improve performance, which is a fundamental purpose of computer-aided diagnosis.

Deep neural network-based computer-assisted detection of cerebral aneurysms in MR angiography

BACKGROUND

The usefulness of computer-assisted detection (CAD) for detecting cerebral aneurysms has been reported; therefore, the improved performance of CAD will help to detect cerebral aneurysms.

PURPOSE

To develop a CAD system for intracranial aneurysms on unenhanced magnetic resonance angiography (MRA) images based on a deep convolutional neural network (CNN) and a maximum intensity projection (MIP) algorithm, and to demonstrate the usefulness of the system by training and evaluating it using a large dataset.

STUDY TYPE

Retrospective study.

SUBJECTS

There were 450 cases with intracranial aneurysms. The diagnoses of brain aneurysms were made on the basis of MRA, which was performed as part of a brain screening program.

FIELD STRENGTH/SEQUENCE

Noncontrast-enhanced 3D time-of-flight (TOF) MRA on 3T MR scanners.

ASSESSMENT

In our CAD, we used a CNN classifier that predicts whether each voxel is inside or outside aneurysms by inputting MIP images generated from a volume of interest (VOI) around the voxel. The CNN was trained in advance using manually inputted labels. We evaluated our method using 450 cases with intracranial aneurysms, 300 of which were used for training, 50 for parameter tuning, and 100 for the final evaluation.

STATISTICAL TESTS

Free-response receiver operating characteristic (FROC) analysis.

RESULTS

Our CAD system detected 94.2% (98/104) of aneurysms with 2.9 false positives per case (FPs/case). At a sensitivity of 70%, the number of FPs/case was 0.26.

DATA CONCLUSION

We showed that the combination of a CNN and an MIP algorithm is useful for the detection of intracranial aneurysms.

LEVEL OF EVIDENCE

4 Technical Efficacy: Stage 1 J. Magn. Reson. Imaging 2018;47:948-953.

Alzheimer's Disease Early Diagnosis Using Manifold-Based Semi-Supervised Learning

Alzheimer's disease (AD) is currently ranked as the sixth leading cause of death in the United States and recent estimates indicate that the disorder may rank third, just behind heart disease and cancer, as a cause of death for older people. Clearly, predicting this disease in the early stages and preventing it from progressing is of great importance. The diagnosis of Alzheimer's disease (AD) requires a variety of medical tests, which leads to huge amounts of multivariate heterogeneous data. It can be difficult and exhausting to manually compare, visualize, and analyze this data due to the heterogeneous nature of medical tests; therefore, an efficient approach for accurate prediction of the condition of the brain through the classification of magnetic resonance imaging (MRI) images is greatly beneficial and yet very challenging. In this paper, a novel approach is proposed for the diagnosis of very early stages of AD through an efficient classification of brain MRI images, which uses label propagation in a manifold-based semi-supervised learning framework. We first apply voxel morphometry analysis to extract some of the most critical AD-related features of brain images from the original MRI volumes and also gray matter (GM) segmentation volumes. The features must capture the most discriminative properties that vary between a healthy and Alzheimer-affected brain. Next, we perform a principal component analysis (PCA)-based dimension reduction on the extracted features for faster yet sufficiently accurate analysis. To make the best use of the captured features, we present a hybrid manifold learning framework which embeds the feature vectors in a subspace. Next, using a small set of labeled training data, we apply a label propagation method in the created manifold space to predict the labels of the remaining images and classify them in the two groups of mild Alzheimer's and normal condition (MCI/NC). The accuracy of the classification using the proposed method is 93.86% for the Open Access Series of Imaging Studies (OASIS) database of MRI brain images, providing, compared to the best existing methods, a 3% lower error rate.

Automated Segmentation of Light-Sheet Fluorescent Imaging to Characterize Experimental Doxorubicin-Induced Cardiac Injury and Repair

This study sought to develop an automated segmentation approach based on histogram analysis of raw axial images acquired by light-sheet fluorescent imaging (LSFI) to establish rapid reconstruction of the 3-D zebrafish cardiac architecture in response to doxorubicin-induced injury and repair. Input images underwent a 4-step automated image segmentation process consisting of stationary noise removal, histogram equalization, adaptive thresholding, and image fusion followed by 3-D reconstruction. We applied this method to 3-month old zebrafish injected intraperitoneally with doxorubicin followed by LSFI at 3, 30, and 60 days post-injection. We observed an initial decrease in myocardial and endocardial cavity volumes at day 3, followed by ventricular remodeling at day 30, and recovery at day 60 (P < 0.05, n = 7-19). Doxorubicin-injected fish developed ventricular diastolic dysfunction and worsening global cardiac function evidenced by elevated E/A ratios and myocardial performance indexes quantified by pulsed-wave Doppler ultrasound at day 30, followed by normalization at day 60 (P < 0.05, n = 9-20). Treatment with the γ-secretase inhibitor, DAPT, to inhibit cleavage and release of Notch Intracellular Domain (NICD) blocked cardiac architectural regeneration and restoration of ventricular function at day 60 (P < 0.05, n = 6-14). Our approach provides a high-throughput model with translational implications for drug discovery and genetic modifiers of chemotherapy-induced cardiomyopathy.

Quantitative computer-aided diagnostic algorithm for automated detection of peak lesion attenuation in differentiating clear cell from papillary and chromophobe renal cell carcinoma, oncocytoma, and fat-poor angiomyolipoma on multiphasic multidetector computed tomography

OBJECTIVE

To evaluate the performance of a novel, quantitative computer-aided diagnostic (CAD) algorithm on four-phase multidetector computed tomography (MDCT) to detect peak lesion attenuation to enable differentiation of clear cell renal cell carcinoma (ccRCC) from chromophobe RCC (chRCC), papillary RCC (pRCC), oncocytoma, and fat-poor angiomyolipoma (fp-AML).

MATERIALS AND METHODS

We queried our clinical databases to obtain a cohort of histologically proven renal masses with preoperative MDCT with four phases [unenhanced (U), corticomedullary (CM), nephrographic (NP), and excretory (E)]. A whole lesion 3D contour was obtained in all four phases. The CAD algorithm determined a region of interest (ROI) of peak lesion attenuation within the 3D lesion contour. For comparison, a manual ROI was separately placed in the most enhancing portion of the lesion by visual inspection for a reference standard, and in uninvolved renal cortex. Relative lesion attenuation for both CAD and manual methods was obtained by normalizing the CAD peak lesion attenuation ROI (and the reference standard manually placed ROI) to uninvolved renal cortex with the formula [(peak lesion attenuation ROI - cortex ROI)/cortex ROI] × 100%. ROC analysis and area under the curve (AUC) were used to assess diagnostic performance. Bland-Altman analysis was used to compare peak ROI between CAD and manual method.

RESULTS

The study cohort comprised 200 patients with 200 unique renal masses: 106 (53%) ccRCC, 32 (16%) oncocytomas, 18 (9%) chRCCs, 34 (17%) pRCCs, and 10 (5%) fp-AMLs. In the CM phase, CAD-derived ROI enabled characterization of ccRCC from chRCC, pRCC, oncocytoma, and fp-AML with AUCs of 0.850 (95% CI 0.732-0.968), 0.959 (95% CI 0.930-0.989), 0.792 (95% CI 0.716-0.869), and 0.825 (95% CI 0.703-0.948), respectively. On Bland-Altman analysis, there was excellent agreement of CAD and manual methods with mean differences between 14 and 26 HU in each phase.

CONCLUSION

A novel, quantitative CAD algorithm enabled robust peak HU lesion detection and discrimination of ccRCC from other renal lesions with similar performance compared to the manual method.

Effectiveness of computer-aided diagnosis (CADx) of breast pathology using immunohistochemistry results of core needle biopsy samples for synaptophysin, oestrogen receptor and CK14/p63 for classification of epithelial proliferative lesions of the breast

Aims

The aim of this study was to develop a computer-aided diagnosis (CADx) system for identifying breast pathology.

Methods

Two sets of 100 consecutive core needle biopsy (CNB) specimens were collected for test and validation studies. All 200 CNB specimens were stained with antibodies targeting oestrogen receptor (ER), synaptophysin and CK14/p63. All stained slides were scanned in a whole-slide imaging system and photographed. The photographs were analysed using software to identify the proportions of tumour cells that were positive and negative for each marker. In the test study, the cut-off values for synaptophysin (negative and positive) and CK14/p63 (negative and positive) were decided using receiver operating characteristic (ROC) analysis. For ER analysis, samples were divided into groups with <10% positive or >10% positive cells and decided using receiver operating characteristic (ROC) analysis. Finally, these two groups categorised as ER-low, ER-intermediate (non-low and non-high) and ER-high groups. In the validation study, the second set of immunohistochemical slides were analysed using these cut-off values.

Results

The cut-off values for synaptophysin, <10% ER positive, >10% ER positive and CK14/p63 were 0.14%, 2.17%, 77.93% and 18.66%, respectively. The positive predictive value for malignancy (PPV) was 100% for synaptophysin-positive/ER-high/(CK14/p63)-any or synaptophysin-positive/ER-low/(CK14/p63)-any. The PPV was 25% for synaptophysin-positive/ER-intermediate/(CK14/p63)-positive. For synaptophysin-negative/(CK14/p63)-negative, the PPVs for ER-low, ER-intermediate and ER-high were 100%, 80.0% and 95.8%, respectively. The PPV was 4.5% for synaptophysin-negative/ER-intermediate/(CK14/p63)-positive.

Conclusion

The CADx system was able to analyse sufficient data for all types of epithelial proliferative lesions of the breast including invasive breast cancer. This system may be useful for pathological diagnosis of breast CNB in routine investigations.

Computer-aided diagnosis program for classifying the risk of hepatocellular carcinoma on MR images following liver imaging reporting and data system (LI-RADS)

PURPOSE

To develop and evaluate a computer-aided diagnosis (CAD) program for liver lesions on magnetic resonance (MR) images for classification of the risk of hepatocellular carcinoma (HCC) following the liver imaging reporting and data system (LI-RADS).

MATERIALS AND METHODS

Liver MR images from 41 patients with hyperenhancing liver lesions categorized as LR 3, 4, and 5 were evaluated by two radiologists. The major LI-RADS features of each index liver lesion were recorded, including size (maximum transverse diameter), presence of hyperenhancement, washout appearance, and capsule appearance. A CAD program was implemented to register MR images at different contrast-enhancement phases, segment liver lesions, extract lesion features, and classify lesions according to LI-RADS. The LI-RADS features quantified by CAD were compared with those assessed by radiologists using the intraclass correlation coefficient (ICC) and receiver operator curve (ROC) analyses. The LI-RADS categorization between CAD and radiologists was evaluated using the weighted Cohen's kappa coefficient.

RESULTS

The mean and standard deviation of the lesion diameters were 21 ± 11 mm (range, 7-70 mm) by radiologists and 22 ± 11 mm (range, 8-72 mm) by CAD (ICC, 0.96-0.97). The area under the curve (AUC) for the washout assessment by CAD was 0.79-0.93 with sensitivity 0.69-0.82 and specificity 0.79-1. The AUC for the capsule assessment by CAD was 0.79-0.9 with sensitivity 0.75-0.9 and specificity 0.82-0.96. The classifications by the radiologists and CAD coincided in 76-83% lesions (k = 0.57-0.71), while the agreements between radiologists were in 78% lesions (k = 0.59).

CONCLUSION

We developed a CAD program for liver lesions on MR images and showed a substantial agreement in the LI-RADS-based classification of the risk of HCCs between the CAD and radiologists.

LEVEL OF EVIDENCE

1 Technical Efficacy: Stage 1 J. Magn. Reson. Imaging 2018;47:710-722.

Feasibility Study of a Generalized Framework for Developing Computer-Aided Detection Systems-a New Paradigm

We propose a generalized framework for developing computer-aided detection (CADe) systems whose characteristics depend only on those of the training dataset. The purpose of this study is to show the feasibility of the framework. Two different CADe systems were experimentally developed by a prototype of the framework, but with different training datasets. The CADe systems include four components; preprocessing, candidate area extraction, candidate detection, and candidate classification. Four pretrained algorithms with dedicated optimization/setting methods corresponding to the respective components were prepared in advance. The pretrained algorithms were sequentially trained in the order of processing of the components. In this study, two different datasets, brain MRA with cerebral aneurysms and chest CT with lung nodules, were collected to develop two different types of CADe systems in the framework. The performances of the developed CADe systems were evaluated by threefold cross-validation. The CADe systems for detecting cerebral aneurysms in brain MRAs and for detecting lung nodules in chest CTs were successfully developed using the respective datasets. The framework was shown to be feasible by the successful development of the two different types of CADe systems. The feasibility of this framework shows promise for a new paradigm in the development of CADe systems: development of CADe systems without any lesion specific algorithm designing.

Relevance of deep learning to facilitate the diagnosis of HER2 status in breast cancer

Tissue biomarker scoring by pathologists is central to defining the appropriate therapy for patients with cancer. Yet, inter-pathologist variability in the interpretation of ambiguous cases can affect diagnostic accuracy. Modern artificial intelligence methods such as deep learning have the potential to supplement pathologist expertise to ensure constant diagnostic accuracy. We developed a computational approach based on deep learning that automatically scores HER2, a biomarker that defines patient eligibility for anti-HER2 targeted therapies in breast cancer. In a cohort of 71 breast tumour resection samples, automated scoring showed a concordance of 83% with a pathologist. The twelve discordant cases were then independently reviewed, leading to a modification of diagnosis from initial pathologist assessment for eight cases. Diagnostic discordance was found to be largely caused by perceptual differences in assessing HER2 expression due to high HER2 staining heterogeneity. This study provides evidence that deep learning aided diagnosis can facilitate clinical decision making in breast cancer by identifying cases at high risk of misdiagnosis.

Technical Note: Image filtering to make computer-aided detection robust to image reconstruction kernel choice in lung cancer CT screening

PURPOSE

In lung cancer computed tomography (CT) screening, the performance of a computer-aided detection (CAD) system depends on the selection of the image reconstruction kernel. To reduce this dependence on reconstruction kernels, the authors propose a novel application of an image filtering method previously proposed by their group.

METHODS

The proposed filtering process uses the ratio of modulation transfer functions (MTFs) of two reconstruction kernels as a filtering function in the spatial-frequency domain. This method is referred to as MTFratio filtering. Test image data were obtained from CT screening scans of 67 subjects who each had one nodule. Images were reconstructed using two kernels: fSTD (for standard lung imaging) and fSHARP (for sharp edge-enhancement lung imaging). The MTFratio filtering was implemented using the MTFs measured for those kernels and was applied to the reconstructed fSHARP images to obtain images that were similar to the fSTD images. A mean filter and a median filter were applied (separately) for comparison. All reconstructed and filtered images were processed using their prototype CAD system.

RESULTS

The MTFratio filtered images showed excellent agreement with the fSTD images. The standard deviation for the difference between these images was very small, ∼6.0 Hounsfield units (HU). However, the mean and median filtered images showed larger differences of ∼48.1 and ∼57.9 HU from the fSTD images, respectively. The free-response receiver operating characteristic (FROC) curve for the fSHARP images indicated poorer performance compared with the FROC curve for the fSTD images. The FROC curve for the MTFratio filtered images was equivalent to the curve for the fSTD images. However, this similarity was not achieved by using the mean filter or median filter.

CONCLUSIONS

The accuracy of MTFratio image filtering was verified and the method was demonstrated to be effective for reducing the kernel dependence of CAD performance.

Fifty years of computer analysis in chest imaging: rule-based, machine learning, deep learning

Half a century ago, the term "computer-aided diagnosis" (CAD) was introduced in the scientific literature. Pulmonary imaging, with chest radiography and computed tomography, has always been one of the focus areas in this field. In this study, I describe how machine learning became the dominant technology for tackling CAD in the lungs, generally producing better results than do classical rule-based approaches, and how the field is now rapidly changing: in the last few years, we have seen how even better results can be obtained with deep learning. The key differences among rule-based processing, machine learning, and deep learning are summarized and illustrated for various applications of CAD in the chest.

Deep learning is effective for the classification of OCT images of normal versus Age-related Macular Degeneration

Objective

The advent of Electronic Medical Records (EMR) with large electronic imaging databases along with advances in deep neural networks with machine learning has provided a unique opportunity to achieve milestones in automated image analysis. Optical coherence tomography (OCT) is the most commonly obtained imaging modality in ophthalmology and represents a dense and rich dataset when combined with labels derived from the EMR. We sought to determine if deep learning could be utilized to distinguish normal OCT images from images from patients with Age-related Macular Degeneration (AMD).

Design

EMR and OCT database study.

Subjects

Normal and AMD patients who had a macular OCT.

Methods

Automated extraction of an OCT imaging database was performed and linked to clinical endpoints from the EMR. OCT macula scans were obtained by Heidelberg Spectralis, and each OCT scan was linked to EMR clinical endpoints extracted from EPIC. The central 11 images were selected from each OCT scan of two cohorts of patients: normal and AMD. Cross-validation was performed using a random subset of patients. Receiver operator curves (ROC) were constructed at an independent image level, macular OCT level, and patient level.

Main outcome measure

Area under the ROC.

Results

Of a recent extraction of 2.6 million OCT images linked to clinical datapoints from the EMR, 52,690 normal macular OCT images and 48,312 AMD macular OCT images were selected. A deep neural network was trained to categorize images as either normal or AMD. At the image level, we achieved an area under the ROC of 92.78% with an accuracy of 87.63%. At the macula level, we achieved an area under the ROC of 93.83% with an accuracy of 88.98%. At a patient level, we achieved an area under the ROC of 97.45% with an accuracy of 93.45%. Peak sensitivity and specificity with optimal cutoffs were 92.64% and 93.69% respectively.

Conclusions

Deep learning techniques achieve high accuracy and is effective as a new image classification technique. These findings have important implications in utilizing OCT in automated screening and the development of computer aided diagnosis tools in the future.

Computer-aided Characterization and Diagnosis of Diffuse Liver Diseases Based on Ultrasound Imaging: A Review

Diffuse liver diseases, such as hepatitis, fatty liver, and cirrhosis, are becoming a leading cause of fatality and disability all over the world. Early detection and diagnosis of these diseases is extremely important to save lives and improve effectiveness of treatment. Ultrasound imaging, a noninvasive diagnostic technique, is the most commonly used modality for examining liver abnormalities. However, the accuracy of ultrasound-based diagnosis depends highly on expertise of radiologists. Computer-aided diagnosis systems based on ultrasound imaging assist in fast diagnosis, provide a reliable "second opinion" for experts, and act as an effective tool to measure response of treatment on patients undergoing clinical trials. In this review, we first describe appearance of liver abnormalities in ultrasound images and state the practical issues encountered in characterization of diffuse liver diseases that can be addressed by software algorithms. We then discuss computer-aided diagnosis in general with features and classifiers relevant to diffuse liver diseases. In later sections of this paper, we review the published studies and describe the key findings of those studies. A concise tabular summary comparing image database, features extraction, feature selection, and classification algorithms presented in the published studies is also exhibited. Finally, we conclude with a summary of key findings and directions for further improvements in the areas of accuracy and objectiveness of computer-aided diagnosis.

Automatic Detection of Pectoral Muscle Region for Computer-Aided Diagnosis Using MIAS Mammograms

The computer-aided detection (CAD) systems have been developed to help radiologists with the early detection of breast cancer. This system provides objective and accurate information to reduce the misdiagnosis of the disease. In mammography, the pectoral muscle region is used as an index to compare the symmetry between the left and right images in the mediolateral oblique (MLO) view. The pectoral muscle segmentation is necessary for the detection of microcalcification or mass because the pectoral muscle has a similar pixel intensity as that of lesions, which affects the results of automatic detection. In this study, the mammographic image analysis society database (MIAS, 322 cases) was used for detecting the pectoral muscle segmentation. The pectoral muscle was detected by using the morphological method and the random sample consensus (RANSAC) algorithm. We evaluated the detected pectoral muscle region and compared the manual segmentation with the automatic segmentation. The results showed 92.2% accuracy. We expect that the proposed method improves the detection accuracy of breast cancer lesions using a CAD system.

Large scale deep learning for computer aided detection of mammographic lesions

Recent advances in machine learning yielded new techniques to train deep neural networks, which resulted in highly successful applications in many pattern recognition tasks such as object detection and speech recognition. In this paper we provide a head-to-head comparison between a state-of-the art in mammography CAD system, relying on a manually designed feature set and a Convolutional Neural Network (CNN), aiming for a system that can ultimately read mammograms independently. Both systems are trained on a large data set of around 45,000 images and results show the CNN outperforms the traditional CAD system at low sensitivity and performs comparable at high sensitivity. We subsequently investigate to what extent features such as location and patient information and commonly used manual features can still complement the network and see improvements at high specificity over the CNN especially with location and context features, which contain information not available to the CNN. Additionally, a reader study was performed, where the network was compared to certified screening radiologists on a patch level and we found no significant difference between the network and the readers.

Automatic detection of large pulmonary solid nodules in thoracic CT images

PURPOSE

Current computer-aided detection (CAD) systems for pulmonary nodules in computed tomography (CT) scans have a good performance for relatively small nodules, but often fail to detect the much rarer larger nodules, which are more likely to be cancerous. We present a novel CAD system specifically designed to detect solid nodules larger than 10 mm.

METHODS

The proposed detection pipeline is initiated by a three-dimensional lung segmentation algorithm optimized to include large nodules attached to the pleural wall via morphological processing. An additional preprocessing is used to mask out structures outside the pleural space to ensure that pleural and parenchymal nodules have a similar appearance. Next, nodule candidates are obtained via a multistage process of thresholding and morphological operations, to detect both larger and smaller candidates. After segmenting each candidate, a set of 24 features based on intensity, shape, blobness, and spatial context are computed. A radial basis support vector machine (SVM) classifier was used to classify nodule candidates, and performance was evaluated using ten-fold cross-validation on the full publicly available lung image database consortium database.

RESULTS

The proposed CAD system reaches a sensitivity of 98.3% (234/238) and 94.1% (224/238) large nodules at an average of 4.0 and 1.0 false positives/scan, respectively.

CONCLUSIONS

The authors conclude that the proposed dedicated CAD system for large pulmonary nodules can identify the vast majority of highly suspicious lesions in thoracic CT scans with a small number of false positives.

Progress in Fully Automated Abdominal CT Interpretation

OBJECTIVE

Automated analysis of abdominal CT has advanced markedly over just the last few years. Fully automated assessment of organs, lymph nodes, adipose tissue, muscle, bowel, spine, and tumors are some examples where tremendous progress has been made. Computer-aided detection of lesions has also improved dramatically.

CONCLUSION

This article reviews the progress and provides insights into what is in store in the near future for automated analysis for abdominal CT, ultimately leading to fully automated interpretation.

Computer-Aided Diagnosis with Deep Learning Architecture: Applications to Breast Lesions in US Images and Pulmonary Nodules in CT Scans

This paper performs a comprehensive study on the deep-learning-based computer-aided diagnosis (CADx) for the differential diagnosis of benign and malignant nodules/lesions by avoiding the potential errors caused by inaccurate image processing results (e.g., boundary segmentation), as well as the classification bias resulting from a less robust feature set, as involved in most conventional CADx algorithms. Specifically, the stacked denoising auto-encoder (SDAE) is exploited on the two CADx applications for the differentiation of breast ultrasound lesions and lung CT nodules. The SDAE architecture is well equipped with the automatic feature exploration mechanism and noise tolerance advantage, and hence may be suitable to deal with the intrinsically noisy property of medical image data from various imaging modalities. To show the outperformance of SDAE-based CADx over the conventional scheme, two latest conventional CADx algorithms are implemented for comparison. 10 times of 10-fold cross-validations are conducted to illustrate the efficacy of the SDAE-based CADx algorithm. The experimental results show the significant performance boost by the SDAE-based CADx algorithm over the two conventional methods, suggesting that deep learning techniques can potentially change the design paradigm of the CADx systems without the need of explicit design and selection of problem-oriented features.

An integrated index for automated detection of infarcted myocardium from cross-sectional echocardiograms using texton-based features (Part 1)

Cross-sectional view echocardiography is an efficient non-invasive diagnostic tool for characterizing Myocardial Infarction (MI) and stages of expansion leading to heart failure. An automated computer-aided technique of cross-sectional echocardiography feature assessment can aid clinicians in early and more reliable detection of MI patients before subsequent catastrophic post-MI medical conditions. Therefore, this paper proposes a novel Myocardial Infarction Index (MII) to discriminate infarcted and normal myocardium using features extracted from apical cross-sectional views of echocardiograms. The cross-sectional view of normal and MI echocardiography images are represented as textons using Maximum Responses (MR8) filter banks. Fractal Dimension (FD), Higher-Order Statistics (HOS), Hu's moments, Gabor Transform features, Fuzzy Entropy (FEnt), Energy, Local binary Pattern (LBP), Renyi's Entropy (REnt), Shannon's Entropy (ShEnt), and Kapur's Entropy (KEnt) features are extracted from textons. These features are ranked using t-test and fuzzy Max-Relevancy and Min-Redundancy (mRMR) ranking methods. Then, combinations of highly ranked features are used in the formulation and development of an integrated MII. This calculated novel MII is used to accurately and quickly detect infarcted myocardium by using one numerical value. Also, the highly ranked features are subjected to classification using different classifiers for the characterization of normal and MI LV ultrasound images using a minimum number of features. Our current technique is able to characterize MI with an average accuracy of 94.37%, sensitivity of 91.25% and specificity of 97.50% with 8 apical four chambers view features extracted from only single frame per patient making this a more reliable and accurate classification.

Automatic 3D pulmonary nodule detection in CT images: A survey

This work presents a systematic review of techniques for the 3D automatic detection of pulmonary nodules in computerized-tomography (CT) images. Its main goals are to analyze the latest technology being used for the development of computational diagnostic tools to assist in the acquisition, storage and, mainly, processing and analysis of the biomedical data. Also, this work identifies the progress made, so far, evaluates the challenges to be overcome and provides an analysis of future prospects. As far as the authors know, this is the first time that a review is devoted exclusively to automated 3D techniques for the detection of pulmonary nodules from lung CT images, which makes this work of noteworthy value. The research covered the published works in the Web of Science, PubMed, Science Direct and IEEEXplore up to December 2014. Each work found that referred to automated 3D segmentation of the lungs was individually analyzed to identify its objective, methodology and results. Based on the analysis of the selected works, several studies were seen to be useful for the construction of medical diagnostic aid tools. However, there are certain aspects that still require attention such as increasing algorithm sensitivity, reducing the number of false positives, improving and optimizing the algorithm detection of different kinds of nodules with different sizes and shapes and, finally, the ability to integrate with the Electronic Medical Record Systems and Picture Archiving and Communication Systems. Based on this analysis, we can say that further research is needed to develop current techniques and that new algorithms are needed to overcome the identified drawbacks.

Computer-aided classification of lung nodules on computed tomography images via deep learning technique

Lung cancer has a poor prognosis when not diagnosed early and unresectable lesions are present. The management of small lung nodules noted on computed tomography scan is controversial due to uncertain tumor characteristics. A conventional computer-aided diagnosis (CAD) scheme requires several image processing and pattern recognition steps to accomplish a quantitative tumor differentiation result. In such an ad hoc image analysis pipeline, every step depends heavily on the performance of the previous step. Accordingly, tuning of classification performance in a conventional CAD scheme is very complicated and arduous. Deep learning techniques, on the other hand, have the intrinsic advantage of an automatic exploitation feature and tuning of performance in a seamless fashion. In this study, we attempted to simplify the image analysis pipeline of conventional CAD with deep learning techniques. Specifically, we introduced models of a deep belief network and a convolutional neural network in the context of nodule classification in computed tomography images. Two baseline methods with feature computing steps were implemented for comparison. The experimental results suggest that deep learning methods could achieve better discriminative results and hold promise in the CAD application domain.

Statistical shape model reconstruction with sparse anomalous deformations: Application to intervertebral disc herniation

Many medical image processing techniques rely on accurate shape modeling of anatomical features. The presence of shape abnormalities challenges traditional processing algorithms based on strong morphological priors. In this work, a sparse shape reconstruction from a statistical shape model is presented. It combines the advantages of traditional statistical shape models (defining a 'normal' shape space) and previously presented sparse shape composition (providing localized descriptors of anomalies). The algorithm was incorporated into our image segmentation and classification software. Evaluation was performed on simulated and clinical MRI data from 22 sciatica patients with intervertebral disc herniation, containing 35 herniated and 97 normal discs. Moderate to high correlation (R=0.73) was achieved between simulated and detected herniations. The sparse reconstruction provided novel quantitative features describing the herniation morphology and MRI signal appearance in three dimensions (3D). The proposed descriptors of local disc morphology resulted to the 3D segmentation accuracy of 1.07±1.00mm (mean absolute vertex-to-vertex mesh distance over the posterior disc region), and improved the intervertebral disc classification from 0.888 to 0.931 (area under receiver operating curve). The results show that the sparse shape reconstruction may improve computer-aided diagnosis of pathological conditions presenting local morphological alterations, as seen in intervertebral disc herniation.

Automated system for lung nodules classification based on wavelet feature descriptor and support vector machine

Background

Lung cancer is a leading cause of death worldwide; it refers to the uncontrolled growth of abnormal cells in the lung. A computed tomography (CT) scan of the thorax is the most sensitive method for detecting cancerous lung nodules. A lung nodule is a round lesion which can be either non-cancerous or cancerous. In the CT, the lung cancer is observed as round white shadow nodules. The possibility to obtain a manually accurate interpretation from CT scans demands a big effort by the radiologist and might be a fatiguing process. Therefore, the design of a computer-aided diagnosis (CADx) system would be helpful as a second opinion tool.

Methods

The stages of the proposed CADx are: a supervised extraction of the region of interest to eliminate the shape differences among CT images. The Daubechies db1, db2, and db4 wavelet transforms are computed with one and two levels of decomposition. After that, 19 features are computed from each wavelet sub-band. Then, the sub-band and attribute selection is performed. As a result, 11 features are selected and combined in pairs as inputs to the support vector machine (SVM), which is used to distinguish CT images containing cancerous nodules from those not containing nodules.

Results

The clinical data set used for experiments consists of 45 CT scans from ELCAP and LIDC. For the training stage 61 CT images were used (36 with cancerous lung nodules and 25 without lung nodules). The system performance was tested with 45 CT scans (23 CT scans with lung nodules and 22 without nodules), different from that used for training. The results obtained show that the methodology successfully classifies cancerous nodules with a diameter from 2 mm to 30 mm. The total preciseness obtained was 82%; the sensitivity was 90.90%, whereas the specificity was 73.91%.

Conclusions

The CADx system presented is competitive with other literature systems in terms of sensitivity. The system reduces the complexity of classification by not performing the typical segmentation stage of most CADx systems. Additionally, the novelty of the algorithm is the use of a wavelet feature descriptor.

[Lung cancer screening. What have we learnt for the practice so far?]

CLINICAL/METHODICAL ISSUE

Lung cancer is the most frequent cause of tumor-associated death and only has a good prognosis if detected at a very early tumor stage.

METHODICAL INNOVATIONS

For the first time the American National Lung Screening Trial (NLST) could prove that low-dose computed tomography (CT) screening is able to reduce lung cancer mortality by 20 %.

PERFORMANCE

To date, however, three much smaller and therefore statistically underpowered European trials could not confirm the positive results of the NLST. The results of the largest European trial NELSON are expected within the next 2 years. In addition, there are a number of open or not yet satisfactorily answered questions, such as the definition of the appropriate screening population, the management of nodules detected by screening, the effects of over-diagnosis and the risk of cumulative radiation exposure.

PRACTICAL RECOMMENDATIONS

The success of the NLST prompted several predominantly American professional societies to issue a positive recommendation about the implementation of lung cancer screening in a population at risk. However, potentially conflicting results of European studies and a number of not yet optimized issues justify caution and call for a pooled analysis of European studies in order to provide statistically sound results and to ensure a high efficiency of screening with respect to the radiation applied, mental and physical patient burden and, last but not least, the financial efforts.

Diagnostic accuracy of computer-aided detection of pulmonary tuberculosis in chest radiographs: a validation study from sub-Saharan Africa

Background

Chest radiography to diagnose and screen for pulmonary tuberculosis has limitations, especially due to inter-reader variability. Automating the interpretation has the potential to overcome this drawback and to deliver objective and reproducible results. The CAD4TB software is a computer-aided detection system that has shown promising preliminary findings. Evaluation studies in different settings are needed to assess diagnostic accuracy and practicability of use.

Methods

CAD4TB was evaluated on chest radiographs of patients with symptoms suggestive of pulmonary tuberculosis enrolled in two cohort studies in Tanzania. All patients were characterized by sputum smear microscopy and culture including subsequent antigen or molecular confirmation of Mycobacterium tuberculosis (M.tb) to determine the reference standard. Chest radiographs were read by the software and two human readers, one expert reader and one clinical officer. The sensitivity and specificity of CAD4TB was depicted using receiver operating characteristic (ROC) curves, the area under the curve calculated and the performance of the software compared to the results of human readers.

Results

Of 861 study participants, 194 (23%) were culture-positive for M.tb. The area under the ROC curve of CAD4TB for the detection of culture-positive pulmonary tuberculosis was 0.84 (95% CI 0.80–0.88). CAD4TB was significantly more accurate for the discrimination of smear-positive cases against non TB patients than for smear-negative cases (p-value<0.01). It differentiated better between TB cases and non TB patients among HIV-negative compared to HIV-positive individuals (p<0.01). CAD4TB significantly outperformed the clinical officer, but did not reach the accuracy of the expert reader (p = 0.02), for a tuberculosis specific reading threshold.

Conclusion

CAD4TB accurately distinguished between the chest radiographs of culture-positive TB cases and controls. Further studies on cost-effectiveness, operational and ethical aspects should determine its place in diagnostic and screening algorithms.

Predicting visual semantic descriptive terms from radiological image data: preliminary results with liver lesions in CT

We describe a framework to model visual semantics of liver lesions in CT images in order to predict the visual semantic terms (VST) reported by radiologists in describing these lesions. Computational models of VST are learned from image data using linear combinations of high-order steerable Riesz wavelets and support vector machines (SVM). In a first step, these models are used to predict the presence of each semantic term that describes liver lesions. In a second step, the distances between all VST models are calculated to establish a nonhierarchical computationally-derived ontology of VST containing inter-term synonymy and complementarity. A preliminary evaluation of the proposed framework was carried out using 74 liver lesions annotated with a set of 18 VSTs from the RadLex ontology. A leave-one-patient-out cross-validation resulted in an average area under the ROC curve of 0.853 for predicting the presence of each VST. The proposed framework is expected to foster human-computer synergies for the interpretation of radiological images while using rotation-covariant computational models of VSTs to 1) quantify their local likelihood and 2) explicitly link them with pixel-based image content in the context of a given imaging domain.

Clinical evaluation of semi-automatic landmark-based lesion tracking software for CT-scans

Background

To evaluate a semi-automatic landmark-based lesion tracking software enabling navigation between RECIST lesions in baseline and follow-up CT-scans.

Methods

The software automatically detects 44 stable anatomical landmarks in each thoraco/abdominal/pelvic CT-scan, sets up a patient specific coordinate-system and cross-links the coordinate-systems of consecutive CT-scans. Accuracy of the software was evaluated on 96 RECIST lesions (target- and non-target lesions) in baseline and follow-up CT-scans of 32 oncologic patients (64 CT-scans). Patients had to present at least one thoracic, one abdominal and one pelvic RECIST lesion. Three radiologists determined the deviation between lesions’ centre and the software’s navigation result in consensus.

Results

The initial mean runtime of the system to synchronize baseline and follow-up examinations was 19.4 ± 1.2 seconds, with subsequent navigation to corresponding RECIST lesions facilitating in real-time. Mean vector length of the deviations between lesions’ centre and the semi-automatic navigation result was 10.2 ± 5.1 mm without a substantial systematic error in any direction. Mean deviation in the cranio-caudal dimension was 5.4 ± 4.0 mm, in the lateral dimension 5.2 ± 3.9 mm and in the ventro-dorsal dimension 5.3 ± 4.0 mm.

Conclusion

The investigated software accurately and reliably navigates between lesions in consecutive CT-scans in real-time, potentially accelerating and facilitating cancer staging.

Computer-aided detection system for lung cancer in computed tomography scans: review and future prospects

Introduction

The goal of this paper is to present a critical review of major Computer-Aided Detection systems (CADe) for lung cancer in order to identify challenges for future research. CADe systems must meet the following requirements: improve the performance of radiologists providing high sensitivity in the diagnosis, a low number of false positives (FP), have high processing speed, present high level of automation, low cost (of implementation, training, support and maintenance), the ability to detect different types and shapes of nodules, and software security assurance.

Methods

The relevant literature related to “CADe for lung cancer” was obtained from PubMed, IEEEXplore and Science Direct database. Articles published from 2009 to 2013, and some articles previously published, were used. A systemic analysis was made on these articles and the results were summarized.

Discussion

Based on literature search, it was observed that many if not all systems described in this survey have the potential to be important in clinical practice. However, no significant improvement was observed in sensitivity, number of false positives, level of automation and ability to detect different types and shapes of nodules in the studied period. Challenges were presented for future research.

Conclusions

Further research is needed to improve existing systems and propose new solutions. For this, we believe that collaborative efforts through the creation of open source software communities are necessary to develop a CADe system with all the requirements mentioned and with a short development cycle. In addition, future CADe systems should improve the level of automation, through integration with picture archiving and communication systems (PACS) and the electronic record of the patient, decrease the number of false positives, measure the evolution of tumors, evaluate the evolution of the oncological treatment, and its possible prognosis.

[Future of mammography-based imaging]

Mammography is the central diagnostic method for clinical diagnostics of breast cancer and the breast cancer screening program. In the clinical routine complementary methods, such as ultrasound, tomosynthesis and optional magnetic resonance imaging (MRI) are already combined for the diagnostic procedure. Future developments will utilize investigative procedures either as a hybrid (combination of several different imaging modalities in one instrument) or as a fusion method (the technical fusion of two or more of these methods) to implement fusion imaging into diagnostic algorithms. For screening there are reasonable hypotheses to aim for studies that individualize the diagnostic process within the screening procedure. Individual breast cancer risk prediction and individualized knowledge about sensitivity and specificity for certain diagnostic methods could be tested. The clinical implementation of these algorithms is not yet in sight.

A retrieval-based computer-aided diagnosis system for the characterization of liver lesions in CT scans

RATIONALE AND OBJECTIVES

To evaluate a computer-aided diagnosis (CADx) system for the characterization of liver lesions in computed tomography (CT) scans. The stand-alone predictive performance of the CADx system was assessed and compared to that of three radiologists who were provided with the same amount of image information to which the CADx system had access.

MATERIALS AND METHODS

The CADx system operates as an image search engine exploiting texture analysis of liver lesion image data for the lesion in question and lesions from a database. A region of interest drawn around an indeterminate liver lesion is used as input query. The CADx system retrieves lesions of similar histology (benign/malignant), density (hypodense/hyperdense), or type (cyst/hemangioma/metastasis). The system's performance was evaluated with leave-one-patient-out receiver operating characteristic area under the curve on 685 CT scans from 372 patients that contained 2325 liver lesions (193 <1 cm(3)). Sensitivity, specificity, and positive and negative predictive values were evaluated separately for subcentimeter lesions. Results were compared to those of three radiologists who rated 83 liver lesions (20 hemangiomas, 20 metastases, 20 cysts, 20 hepatocellular carcinomas, and 3 focal nodular hyperplasias) displaying only the liver.

RESULTS

The CADx system's leave-one-patient-out receiver operating characteristic area under the curve was 97.1% for density, 91.4% for histology, and 95.5% for lesion type. For subcentimeter lesions, input of additional semantic information improved the system's performance. The CADx system has been proved to significantly outperform radiologists in discriminating lesion histology and type, provided the radiologists have no access to information other than the image. The radiologists were most reliable in diagnosing hemangioma given the limited image data.

CONCLUSIONS

The CADx system under study discriminated reliably between various liver lesions, even outperforming radiologists when accessing the same image information and demonstrated promising performance in classifying subcentimeter lesions in particular.

Bone age assessment: automated techniques coming of age?

Bone age determination from hand radiographs is one of the oldest radiographic procedures. The first atlas was published by Poland in 1898, and to date the Greulich Pyle atlas, although it dates from 1959, is still the most commonly used method. Bone age rating is time-consuming, suffers from an unsatisfactorily high rater variability, and therefore already 25 years ago it was proposed to replace the manual rating by an automated, computerized method, a field nowadays referred to as computer-aided diagnosis (CAD). The pursuit of this goal reached a first stage of accomplishment in 1992-1996 with the presentation of several systems. However, they had limited clinical value, and efforts in CAD research were increasingly focused on lesion detection for cancer screening. It was only in 2008 that a fully-automated bone age method was presented, which appears to be clinically acceptable. In this paper we consider the requirements that should be met by an automated bone age method and review the state of the art. Integration in PACS and saving time are important factors for radiologists. But it is the validation of the methods which poses the greatest challenge, because there is no gold standard for bone age rating, and the direct comparison to manual rating is therefore not sufficient for demonstrating that manual rating can be replaced by automated rating. One needs additional studies assessing the precision of a method and its accuracy when used for adult height prediction, which serves as an objective.