End-to-End Adversarial Retinal Image Synthesis.

In medical image analysis applications, the availability of the large amounts of annotated data is becoming increasingly critical. However, annotated medical data is often scarce and costly to obtain. In this paper, we address the problem of synthesizing retinal color images by applying recent techniques based on adversarial learning. In this setting, a generative model is trained to maximize a loss function provided by a second model attempting to classify its output into real or synthetic. In particular, we propose to implement an adversarial autoencoder for the task of retinal vessel network synthesis. We use the generated vessel trees as an intermediate stage for the generation of color retinal images, which is accomplished with a generative adversarial network. Both models require the optimization of almost everywhere differentiable loss functions, which allows us to train them jointly. The resulting model offers an end-to-end retinal image synthesis system capable of generating as many retinal images as the user requires, with their corresponding vessel networks, by sampling from a simple probability distribution that we impose to the associated latent space. We show that the learned latent space contains a well-defined semantic structure, implying that we can perform calculations in the space of retinal images, e.g., smoothly interpolating new data points between two retinal images. Visual and quantitative results demonstrate that the synthesized images are substantially different from those in the training set, while being also anatomically consistent and displaying a reasonable visual quality.

Improved Automated Detection of Diabetic Retinopathy on a Publicly Available Dataset Through Integration of Deep Learning.

PURPOSE: To compare performance of a deep-learning enhanced algorithm for automated detection of diabetic retinopathy (DR), to the previously published performance of that algorithm, the Iowa Detection Program (IDP)-without deep learning components-on the same publicly available set of fundus images and previously reported consensus reference standard set, by three US Board certified retinal specialists. METHODS: We used the previously reported consensus reference standard of referable DR (rDR), defined as International Clinical Classification of Diabetic Retinopathy moderate, severe nonproliferative (NPDR), proliferative DR, and/or macular edema (ME). Neither Messidor-2 images, nor the three retinal specialists setting the Messidor-2 reference standard were used for training IDx-DR version X2.1. Sensitivity, specificity, negative predictive value, area under the curve (AUC), and their confidence intervals (CIs) were calculated. RESULTS: Sensitivity was 96.8% (95% CI: 93.3%-98.8%), specificity was 87.0% (95% CI: 84.2%-89.4%), with 6/874 false negatives, resulting in a negative predictive value of 99.0% (95% CI: 97.8%-99.6%). No cases of severe NPDR, PDR, or ME were missed. The AUC was 0.980 (95% CI: 0.968-0.992). Sensitivity was not statistically different from published IDP sensitivity, which had a CI of 94.4% to 99.3%, but specificity was significantly better than the published IDP specificity CI of 55.7% to 63.0%. CONCLUSIONS: A deep-learning enhanced algorithm for the automated detection of DR, achieves significantly better performance than a previously reported, otherwise essentially identical, algorithm that does not employ deep learning. Deep learning enhanced algorithms have the potential to improve the efficiency of DR screening, and thereby to prevent visual loss and blindness from this devastating disease.

Multimodal Segmentation of Optic Disc and Cup From SD-OCT and Color Fundus Photographs Using a Machine-Learning Graph-Based Approach.

In this work, a multimodal approach is proposed to use the complementary information from fundus photographs and spectral domain optical coherence tomography (SD-OCT) volumes in order to segment the optic disc and cup boundaries. The problem is formulated as an optimization problem where the optimal solution is obtained using a machine-learning theoretical graph-based method. In particular, first the fundus photograph is registered to the 2D projection of the SD-OCT volume. Three in-region cost functions are designed using a random forest classifier corresponding to three regions of cup, rim, and background. Next, the volumes are resampled to create radial scans in which the Bruch's Membrane Opening (BMO) endpoints are easier to detect. Similar to in-region cost function design, the disc-boundary cost function is designed using a random forest classifier for which the features are created by applying the Haar Stationary Wavelet Transform (SWT) to the radial projection image. A multisurface graph-based approach utilizes the in-region and disc-boundary cost images to segment the boundaries of optic disc and cup under feasibility constraints. The approach is evaluated on 25 multimodal image pairs from 25 subjects in a leave-one-out fashion (by subject). The performances of the graph-theoretic approach using three sets of cost functions are compared: 1) using unimodal (OCT only) in-region costs, 2) using multimodal in-region costs, and 3) using multimodal in-region and disc-boundary costs. Results show that the multimodal approaches outperform the unimodal approach in segmenting the optic disc and cup.

Reproducibility of diabetic macular edema estimates from SD-OCT is affected by the choice of image analysis algorithm.

PURPOSE: To evaluate the intersession repeatability of retinal thickness measurements in patients with diabetic macular edema (DME) using the Heidelberg Spectralis optical coherence tomography (OCT) algorithm and a publicly available, three-dimensional graph search-based multilayer OCT segmentation algorithm, the Iowa Reference Algorithm. METHODS: Thirty eyes from 21 patients diagnosed with clinically significant DME were included and underwent consecutive, registered macula-centered spectral-domain optical coherence scans (Heidelberg Spectralis). The OCT scans were segmented into separate surfaces, and the average thickness between internal limiting membrane and outer retinal pigment epithelium complex surfaces was determined using the Iowa Reference Algorithm. Variability between paired scans was analyzed and compared with the retinal thickness obtained from the manufacturer-supplied Spectralis software. RESULTS: The coefficient of repeatability (variation) for central macular thickness using the Iowa Reference Algorithm was 5.26 µm (0.62% [95% confidence interval (CI), 0.43-0.71]), while for the Spectralis algorithm this was 6.84 µm (0.81% [95% CI, 0.55-0.92]). When the central 3 mm was analyzed, the coefficient of repeatability (variation) was 2.46 µm (0.31% [95% CI, 0.23-0.38]) for the Iowa Reference Algorithm and 4.23 µm (0.53% [95% CI, 0.39-0.65]) for the Spectralis software. CONCLUSIONS: The Iowa Reference Algorithm and the Spectralis software provide excellent reproducibility between serial scans in patients with clinically significant DME. The publicly available Iowa Reference Algorithm may have lower between-measurement variation than the manufacturer-supplied Spectralis software for the central 3 mm subfield. These findings have significant implications for the management of patients with DME.

Automated analysis of retinal images for detection of referable diabetic retinopathy.

IMPORTANCE: The diagnostic accuracy of computer detection programs has been reported to be comparable to that of specialists and expert readers, but no computer detection programs have been validated in an independent cohort using an internationally recognized diabetic retinopathy (DR) standard. OBJECTIVE: To determine the sensitivity and specificity of the Iowa Detection Program (IDP) to detect referable diabetic retinopathy (RDR). DESIGN AND SETTING: In primary care DR clinics in France, from January 1, 2005, through December 31, 2010, patients were photographed consecutively, and retinal color images were graded for retinopathy severity according to the International Clinical Diabetic Retinopathy scale and macular edema by 3 masked independent retinal specialists and regraded with adjudication until consensus. The IDP analyzed the same images at a predetermined and fixed set point. We defined RDR as more than mild nonproliferative retinopathy and/or macular edema. PARTICIPANTS: A total of 874 people with diabetes at risk for DR. MAIN OUTCOME MEASURES: Sensitivity and specificity of the IDP to detect RDR, area under the receiver operating characteristic curve, sensitivity and specificity of the retinal specialists' readings, and mean interobserver difference (κ). RESULTS: The RDR prevalence was 21.7% (95% CI, 19.0%-24.5%). The IDP sensitivity was 96.8% (95% CI, 94.4%-99.3%) and specificity was 59.4% (95% CI, 55.7%-63.0%), corresponding to 6 of 874 false-negative results (none met treatment criteria). The area under the receiver operating characteristic curve was 0.937 (95% CI, 0.916-0.959). Before adjudication and consensus, the sensitivity/specificity of the retinal specialists were 0.80/0.98, 0.71/1.00, and 0.91/0.95, and the mean intergrader κ was 0.822. CONCLUSIONS: The IDP has high sensitivity and specificity to detect RDR. Computer analysis of retinal photographs for DR and automated detection of RDR can be implemented safely into the DR screening pipeline, potentially improving access to screening and health care productivity and reducing visual loss through early treatment.

Splat feature classification with application to retinal hemorrhage detection in fundus images.

A novel splat feature classification method is presented with application to retinal hemorrhage detection in fundus images. Reliable detection of retinal hemorrhages is important in the development of automated screening systems which can be translated into practice. Under our supervised approach, retinal color images are partitioned into nonoverlapping segments covering the entire image. Each segment, i.e., splat, contains pixels with similar color and spatial location. A set of features is extracted from each splat to describe its characteristics relative to its surroundings, employing responses from a variety of filter bank, interactions with neighboring splats, and shape and texture information. An optimal subset of splat features is selected by a filter approach followed by a wrapper approach. A classifier is trained with splat-based expert annotations and evaluated on the publicly available Messidor dataset. An area under the receiver operating characteristic curve of 0.96 is achieved at the splat level and 0.87 at the image level. While we are focused on retinal hemorrhage detection, our approach has potential to be applied to other object detection tasks.

Quantification of external limiting membrane disruption caused by diabetic macular edema from SD-OCT.

PURPOSE: Disruption of external limiting membrane (ELM) integrity on spectral-domain optical coherence tomography (SD-OCT) is associated with lower visual acuity outcomes in patients suffering from diabetic macular edema (DME). However, no automated methods to detect ELM and/or determine its integrity from SD-OCT exist. METHODS: Sixteen subjects diagnosed with clinically significant DME (CSME) were included and underwent macula-centered SD-OCT (512 × 19 × 496 voxels). Sixteen subjects without retinal thickening and normal acuity were also scanned (200 × 200 × 1024 voxels). Automated quantification of ELM disruption was achieved as follows. First, 11 surfaces were automatically segmented using our standard 3-D graph-search approach, and the subvolume between surface 6 and 11 containing the ELM region was flattened based on the segmented retinal pigment epithelium (RPE) layer. A second, edge-based graph-search surface-detection method segmented the ELM region in close proximity "above" the RPE, and each ELM A-scan was classified as disrupted or nondisrupted based on six texture features in the vicinity of the ELM surface. The vessel silhouettes were considered in the disruption classification process to avoid false detections of ELM disruption. RESULTS: In subjects with CSME, large areas of disrupted ELM were present. In normal subjects, ELM was largely intact. The mean and 95% confidence interval (CI) of the detected disruption area volume for normal and CSME subjects were mean(normal) = 0.00087 mm(3) and CI(normal) = (0.00074, 0.00100), and mean(CSME) = 0.00461 mm(3) and CI(CSME) = (0.00347, 0.00576) mm(3), respectively. CONCLUSIONS: In this preliminary study, we were able to show that automated quantification of ELM disruption is feasible and can differentiate continuous ELM in normal subjects from disrupted ELM in subjects with CSME. We have started determining the relationships of quantitative ELM disruption markers to visual outcome in patients undergoing treatment for CSME.

Automated segmentation of the choroid from clinical SD-OCT.

PURPOSE: We developed and evaluated a fully automated 3-dimensional (3D) method for segmentation of the choroidal vessels, and quantification of choroidal vasculature thickness and choriocapillaris-equivalent thickness of the macula, and evaluated repeat variability in normal subjects using standard clinically available spectral domain optical coherence tomography (SD-OCT). METHODS: A total of 24 normal subjects was imaged twice, using clinically available, 3D SD-OCT. A novel, fully-automated 3D method was used to segment and visualize the choroidal vasculature in macular scans. Local choroidal vasculature and choriocapillaris-equivalent thicknesses were determined. Reproducibility on repeat imaging was analyzed using overlapping rates, Dice coefficient, and root mean square coefficient of variation (CV) of choroidal vasculature and choriocapillaris-equivalent thicknesses. RESULTS: For the 6 × 6 mm(2) macula-centered region as depicted by the SD-OCT, average choroidal vasculature thickness in normal subjects was 172.1 µm (95% confidence interval [CI] 163.7-180.5 µm) and average choriocapillaris-equivalent thickness was 23.1 µm (95% CI 20.0-26.2 µm). Overlapping rates were 0.79 ± 0.07 and 0.75 ± 0.06, Dice coefficient was 0.78 ± 0.08, CV of choroidal vasculature thickness was 8.0% (95% CI 6.3%-9.4%), and of choriocapillaris-equivalent thickness was 27.9% (95% CI 21.0%-33.3%). CONCLUSIONS: Fully automated 3D segmentation and quantitative analysis of the choroidal vasculature and choriocapillaris-equivalent thickness demonstrated excellent reproducibility in repeat scans (CV 8.0%) and good reproducibility of choriocapillaris-equivalent thickness (CV 27.9%). Our method has the potential to improve the diagnosis and management of patients with eye diseases in which the choroid is affected.

Automatic coronary calcium scoring in low-dose chest computed tomography.

The calcium burden as estimated from non-ECG-synchronized computed tomography (CT) exams acquired in screening of heavy smokers has been shown to be a strong predictor of cardiovascular events. We present a method for automatic coronary calcium scoring with low-dose, non-contrast-enhanced, non-ECG-synchronized chest CT. First, a probabilistic coronary calcium map was created using multi-atlas segmentation. This map assigned an a priori probability for the presence of coronary calcifications at every location in a scan. Subsequently, a statistical pattern recognition system was designed to identify coronary calcifications by texture, size, and spatial features; the spatial features were computed using the coronary calcium map. The detected calcifications were quantified in terms of volume and Agatston score. The best results were obtained by merging the results of three different supervised classification systems, namely direct classification with a nearest neighbor classifier, and two-stage classification with nearest neighbor and support vector machine classifiers.We used a total of 231 test scans containing 45,674 mm³ of coronary calcifications. The presented method detected on average 157/198 mm³ (sensitivity 79.2%) of coronary calcium volume with on average 4 mm false positive volume. Calcium scoring can be performed automatically in low-dose, non-contrast enhanced, non-ECG-synchronized chest CT in screening of heavy smokers to identify subjects who might benefit from preventive treatment.

Distribution of damage to the entire retinal ganglion cell pathway: quantified using spectral-domain optical coherence tomography analysis in patients with glaucoma.

OBJECTIVES To test the hypothesis that the amount and distribution of glaucomatous damage along the entire retinal ganglion cell-axonal complex (RGC-AC) can be quantified and to map the RGC-AC connectivity in early glaucoma using automated image analysis of standard spectral-domain optical coherence tomography. METHODS Spectral-domain optical coherence tomography volumes were obtained from 116 eyes in 58 consecutive patients with glaucoma or suspected glaucoma. Layer and optic nerve head (ONH) analysis was performed; the mean regional retinal ganglion cell layer thickness (68 regions), nerve fiber layer (NFL) thickness (120 regions), and ONH rim area (12 wedge-shaped regions) were determined. Maps of RGC-AC connectivity were created using maximum correlation between regions' ganglion cell layer thickness, NFL thickness, and ONH rim area; for retinal nerve fiber bundle regions, the maximum "thickness correlation paths" were determined. RESULTS The mean (SD) NFL thickness and ganglion cell layer thickness across all macular regions were 22.5 (7.5) µm and 33.9 (8.4) µm, respectively. The mean (SD) rim area across all ONH wedge regions was 0.038 (0.004) mm2. Connectivity maps were obtained successfully and showed typical nerve fiber bundle connectivity of the RGC-AC cell body segment to the initial NFL axonal segment, of the initial to the final RGC-AC NFL axonal segments, of the final RGC-AC NFL axonal to the ONH axonal segment, and of the RGC-AC cell body segment to the ONH axonal segment. CONCLUSIONS In early glaucoma, the amount and distribution of glaucomatous damage along the entire RGC-AC can be quantified and mapped using automated image analysis of standard spectral-domain optical coherence tomography. Our findings should contribute to better detection and improved management of glaucoma.

Multimodal retinal vessel segmentation from spectral-domain optical coherence tomography and fundus photography.

Segmenting retinal vessels in optic nerve head (ONH) centered spectral-domain optical coherence tomography (SD-OCT) volumes is particularly challenging due to the projected neural canal opening (NCO) and relatively low visibility in the ONH center. Color fundus photographs provide a relatively high vessel contrast in the region inside the NCO, but have not been previously used to aid the SD-OCT vessel segmentation process. Thus, in this paper, we present two approaches for the segmentation of retinal vessels in SD-OCT volumes that each take advantage of complimentary information from fundus photographs. In the first approach (referred to as the registered-fundus vessel segmentation approach), vessels are first segmented on the fundus photograph directly (using a k-NN pixel classifier) and this vessel segmentation result is mapped to the SD-OCT volume through the registration of the fundus photograph to the SD-OCT volume. In the second approach (referred to as the multimodal vessel segmentation approach), after fundus-to-SD-OCT registration, vessels are simultaneously segmented with a k -NN classifier using features from both modalities. Three-dimensional structural information from the intraretinal layers and neural canal opening obtained through graph-theoretic segmentation approaches of the SD-OCT volume are used in combination with Gaussian filter banks and Gabor wavelets to generate the features. The approach is trained on 15 and tested on 19 randomly chosen independent image pairs of SD-OCT volumes and fundus images from 34 subjects with glaucoma. Based on a receiver operating characteristic (ROC) curve analysis, the present registered-fundus and multimodal vessel segmentation approaches [area under the curve (AUC) of 0.85 and 0.89, respectively] both perform significantly better than the two previous OCT-based approaches (AUC of 0.78 and 0.83, p < 0.05). The multimodal approach overall performs significantly better than the other three approaches (p < 0.05).

Three-dimensional segmentation of fluid-associated abnormalities in retinal OCT: probability constrained graph-search-graph-cut.

An automated method is reported for segmenting 3-D fluid-associated abnormalities in the retina, so-called symptomatic exudate-associated derangements (SEAD), from 3-D OCT retinal images of subjects suffering from exudative age-related macular degeneration. In the first stage of a two-stage approach, retinal layers are segmented, candidate SEAD regions identified, and the retinal OCT image is flattened using a candidate-SEAD aware approach. In the second stage, a probability constrained combined graph search-graph cut method refines the candidate SEADs by integrating the candidate volumes into the graph cut cost function as probability constraints. The proposed method was evaluated on 15 spectral domain OCT images from 15 subjects undergoing intravitreal anti-VEGF injection treatment. Leave-one-out evaluation resulted in a true positive volume fraction (TPVF), false positive volume fraction (FPVF) and relative volume difference ratio (RVDR) of 86.5%, 1.7%, and 12.8%, respectively. The new graph cut-graph search method significantly outperformed both the traditional graph cut and traditional graph search approaches (p < 0.01, p < 0.04) and has the potential to improve clinical management of patients with choroidal neovascularization due to exudative age-related macular degeneration.

2-D pattern of nerve fiber bundles in glaucoma emerging from spectral-domain optical coherence tomography.

PURPOSE: To correlate the thicknesses of focal regions of the macular ganglion cell layer with those of the peripapillary nerve fiber layer using spectral-domain optical coherence tomography (SD-OCT) in glaucoma subjects. METHODS: Macula and optic nerve head SD-OCT volumes were obtained in 57 eyes of 57 subjects with open-angle glaucoma or glaucoma suspicion. Using a custom automated computer algorithm, the thickness of 66 macular ganglion cell layer regions and the thickness of 12 peripapillary nerve fiber layer regions were measured from registered SD-OCT volumes. The mean thickness of each ganglion cell layer region was correlated to the mean thickness of each peripapillary nerve fiber layer region across subjects. Each ganglion cell layer region was labeled with the peripapillary nerve fiber layer region with the highest correlation using a color-coded map. RESULTS: The resulting color-coded correlation map closely resembled the nerve fiber bundle (NFB) pattern of retinal ganglion cells. The mean r(2) value across all local macular-peripapillary correlations was 0.49 (± 0.11). When separately analyzing the 30 glaucoma subjects from the 27 glaucoma-suspect subjects, the mean r(2) value across all local macular-peripapillary correlations was significantly larger in the glaucoma group (0.56 ± 0.13 vs. 0.37 ± 0.11; P < 0.001). CONCLUSIONS: A two-dimensional (2-D) spatial NFB map of the retina can be developed using structure-structure relationships from SD-OCT. Such SD-OCT-based NFB maps may enhance glaucoma detection and contribute to monitoring change in the future.

Contextual computer-aided detection: improving bright lesion detection in retinal images and coronary calcification identification in CT scans.

Contextual information plays an important role in medical image understanding. Medical experts make use of context to detect and differentiate pathologies in medical images, especially when interpreting difficult cases. The majority of computer-aided diagnosis (CAD) systems, however, employ only local information to classify candidates, without taking into account global image information or the relation of a candidate with neighboring structures. In this paper, we present a generic system for including contextual information in a CAD system. Context is described by means of high-level features based on the spatial relation between lesion candidates and surrounding anatomical landmarks and lesions of different classes (static contextual features) and lesions of the same type (dynamic contextual features). We demonstrate the added value of contextual CAD for two real-world CAD tasks: the identification of exudates and drusen in 2D retinal images and coronary calcifications in 3D computed tomography scans. Results show that in both applications contextual CAD is superior to a local CAD approach with a significant increase of the figure of merit of the Free Receiver Operating Characteristic curve from 0.84 to 0.92 and from 0.88 to 0.98 for exudates and drusen, respectively, and from 0.87 to 0.93 for coronary calcifications.

Automated measurement of the arteriolar-to-venular width ratio in digital color fundus photographs.

A decreased ratio of the width of retinal arteries to veins [arteriolar-to-venular diameter ratio (AVR)], is well established as predictive of cerebral atrophy, stroke and other cardiovascular events in adults. Tortuous and dilated arteries and veins, as well as decreased AVR are also markers for plus disease in retinopathy of prematurity. This work presents an automated method to estimate the AVR in retinal color images by detecting the location of the optic disc, determining an appropriate region of interest (ROI), classifying vessels as arteries or veins, estimating vessel widths, and calculating the AVR. After vessel segmentation and vessel width determination, the optic disc is located and the system eliminates all vessels outside the AVR measurement ROI. A skeletonization operation is applied to the remaining vessels after which vessel crossings and bifurcation points are removed, leaving a set of vessel segments consisting of only vessel centerline pixels. Features are extracted from each centerline pixel in order to assign these a soft label indicating the likelihood that the pixel is part of a vein. As all centerline pixels in a connected vessel segment should be the same type, the median soft label is assigned to each centerline pixel in the segment. Next, artery vein pairs are matched using an iterative algorithm, and the widths of the vessels are used to calculate the AVR. We trained and tested the algorithm on a set of 65 high resolution digital color fundus photographs using a reference standard that indicates for each major vessel in the image whether it is an artery or vein. We compared the AVR values produced by our system with those determined by a semi-automated reference system. We obtained a mean unsigned error of 0.06 (SD 0.04) in 40 images with a mean AVR of 0.67. A second observer using the semi-automated system obtained the same mean unsigned error of 0.06 (SD 0.05) on the set of images with a mean AVR of 0.66. The testing data and reference standard used in this study has been made publicly available.

Evaluation of a computer-aided diagnosis system for diabetic retinopathy screening on public data.

PURPOSE: To evaluate the performance of a comprehensive computer-aided diagnosis (CAD) system for diabetic retinopathy (DR) screening, using a publicly available database of retinal images, and to compare its performance with that of human experts. METHODS: A previously developed, comprehensive DR CAD system was applied to 1200 digital color fundus photographs (nonmydriatic camera, single field) of 1200 eyes in the publicly available Messidor dataset (Methods to Evaluate Segmentation and Indexing Techniques in the Field of Retinal Ophthalmology (http://messidor.crihan.fr). The ability of the system to distinguish normal images from those with DR was determined by using receiver operator characteristic (ROC) analysis. Two experts also determined the presence of DR in each of the images. RESULTS: The system achieved an area under the ROC curve of 0.876 for successfully distinguishing normal images from those with DR with a sensitivity of 92.2% at a specificity of 50%. These compare favorably with the two experts, who achieved sensitivities of 94.5% and 91.2% at a specificity of 50%. CONCLUSIONS: This study shows, for the first time, the performance of a comprehensive DR screening system on an independent, publicly available dataset. The performance of the system on this dataset is comparable with that of human experts.

Vessel boundary delineation on fundus images using graph-based approach.

This paper proposes an algorithm to measure the width of retinal vessels in fundus photographs using graph-based algorithm to segment both vessel edges simultaneously. First, the simultaneous two-boundary segmentation problem is modeled as a two-slice, 3-D surface segmentation problem, which is further converted into the problem of computing a minimum closed set in a node-weighted graph. An initial segmentation is generated from a vessel probability image. We use the REVIEW database to evaluate diameter measurement performance. The algorithm is robust and estimates the vessel width with subpixel accuracy. The method is used to explore the relationship between the average vessel width and the distance from the optic disc in 600 subjects.

On combining computer-aided detection systems.

Computer-aided detection (CAD) is increasingly used in clinical practice and for many applications a multitude of CAD systems have been developed. In practice, CAD systems have different strengths and weaknesses and it is therefore interesting to consider their combination. In this paper, we present generic methods to combine multiple CAD systems and investigate what kind of performance increase can be expected. Experimental results are presented using data from the ANODE09 and ROC09 online CAD challenges for the detection of pulmonary nodules in computed tomography scans and red lesions in retinal images, respectively. For both applications, combination results in a large and significant increase in performance when compared to the best individual CAD system.

Automated segmentation of 3-D spectral OCT retinal blood vessels by neural canal opening false positive suppression.

We present a method for automatically segmenting the blood vessels in optic nerve head (ONH) centered spectral-domain optical coherence tomography (SD-OCT) volumes, with a focus on the ability to segment the vessels in the region near the neural canal opening (NCO). The algorithm first pre-segments the NCO using a graph-theoretic approach. Oriented Gabor wavelets rotated around the center of the NCO are applied to extract features in a 2-D vessel-aimed projection image. Corresponding oriented NCO-based templates are utilized to help suppress the false positive tendency near the NCO boundary. The vessels are identified in a vessel-aimed projection image using a pixel classification algorithm. Based on the 2-D vessel profiles, 3-D vessel segmentation is performed by a triangular-mesh-based graph search approach in the SD-OCT volume. The segmentation method is trained on 5 and is tested on 10 randomly chosen independent ONH-centered SD-OCT volumes from 15 subjects with glaucoma. Using ROC analysis, for the 2-D vessel segmentation, we demonstrate an improvement over the closest previous work with an area under the curve (AUC) of 0.81 (0.72 for previously reported approach) for the region around the NCO and 0.84 for the region outside the NCO (0.81 for previously reported approach).

Comparing and combining algorithms for computer-aided detection of pulmonary nodules in computed tomography scans: The ANODE09 study.

Numerous publications and commercial systems are available that deal with automatic detection of pulmonary nodules in thoracic computed tomography scans, but a comparative study where many systems are applied to the same data set has not yet been performed. This paper introduces ANODE09 ( http://anode09.isi.uu.nl), a database of 55 scans from a lung cancer screening program and a web-based framework for objective evaluation of nodule detection algorithms. Any team can upload results to facilitate benchmarking. The performance of six algorithms for which results are available are compared; five from academic groups and one commercially available system. A method to combine the output of multiple systems is proposed. Results show a substantial performance difference between algorithms, and demonstrate that combining the output of algorithms leads to marked performance improvements.