Jost Lennart Lauermann

Automated detection of exudative age-related macular degeneration in spectral domain optical coherence tomography using deep learning

PurposeOur purpose was to use deep learning for the automated detection of age-related macular degeneration (AMD) in spectral domain optical coherence tomography (SD-OCT).MethodsA total of 1112 cross-section SD-OCT images of patients with exudative AMD and a healthy control group were used for this study. In the first step, an open-source multi-layer deep convolutional neural network (DCNN), which was pretrained with 1.2 million images from ImageNet, was trained and validated with 1012 cross-section SD-OCT scans (AMD: 701; healthy: 311). During this procedure training accuracy, validation accuracy and cross-entropy were computed. The open-source deep learning framework TensorFlow™ (Google Inc., Mountain View, CA, USA) was used to accelerate the deep learning process. In the last step, a created DCNN classifier, using the information of the above mentioned deep learning process, was tested in detecting 100 untrained cross-section SD-OCT images (AMD: 50; healthy: 50). Therefore, an AMD testing score was computed: 0.98 or higher was presumed for AMD.ResultsAfter an iteration of 500 training steps, the training accuracy and validation accuracies were 100%, and the cross-entropy was 0.005. The average AMD scores were 0.997 ± 0.003 in the AMD testing group and 0.9203 ± 0.085 in the healthy comparison group. The difference between the two groups was highly significant (p < 0.001).ConclusionsWith a deep learning-based approach using TensorFlow™, it is possible to detect AMD in SD-OCT with high sensitivity and specificity. With more image data, an expansion of this classifier for other macular diseases or further details in AMD is possible, suggesting an application for this model as a support in clinical decisions. Another possible future application would involve the individual prediction of the progress and success of therapy for different diseases by automatically detecting hidden image information.

Optical Coherence Tomography Angiography Offers New Insights into Choriocapillaris Perfusion

The choriocapillaris (CC) represents a fundamentally important vascular layer that is subject to physiologic changes with increasing age and that is also associated with a wide range of chorioretinal diseases. So far, information on blood flow in this specific layer has remained limited. With the advent of optical coherence tomography angiography (OCTA), new perspectives and possibilities of CC imaging have begun to evolve. This article shall review the opportunities and challenges of applying OCTA technology to the CC layer and summarize the current clinical efforts in OCTA CC imaging exemplarily in dry age-related macular degeneration and central serous chorioretinopathy.

Comparison of Choriocapillaris Flow Measurements between Two Optical Coherence Tomography Angiography Devices

Purpose: To evaluate choriocapillaris (CC) perfusion in healthy subjects using 2 different optical coherence tomography angiography (OCT-A) devices. Procedures: Macular OCT-A imaging (36 eyes of 36 subjects) was performed using Optovue AngioVue and Zeiss AngioPlex devices. CC decorrelation signal index was assessed, and CC data were analyzed regarding intra-device variability, inter-device correlation, age, signal strength, and fields of view. Results: The intra-device variability of CC measurements in the 3 × 3 mm2 field was 5.3 and 2.6% (Angiovue and Angioplex, coefficients of variation; 6 × 6 mm2: 8.0 and 2.8%, respectively). Mean CC decorrelation signal index in 3 × 3 mm2 was 104.3 ± 6.7 (Angiovue) and 81.3 ± 9.2 (Angioplex) (6 × 6 mm2: 95.6 ± 8.1, 81.1 ± 6.5) with high correlation between both devices (3 × 3 mm2: p = 0.0053; 6 × 6 mm2: p = 0.0139). CC decorrelation signal index in 3 × 3 mm2 was significantly higher in subjects aged ≤58 years compared to subjects aged ≥59 years (Angiovue: 107.3 ± 3.6, 101.3 ± 7.7, p = 0.0156; Angioplex: 84.6 ± 7.6, 78.0 ± 9.5, p = 0.0371). Signal strength was 64.6 ± 8.9 (Angiovue) and 9.5 ± 0.8 (Angioplex). Conclusion: Both devices showed low intra-device variability and a high inter-device correlation. CC decorrelation signal index was negatively correlated with advancing age.

Deep learning-based detection and classification of geographic atrophy using a deep convolutional neural network classifier

PurposeTo automatically detect and classify geographic atrophy (GA) in fundus autofluorescence (FAF) images using a deep learning algorithm.MethodsIn this study, FAF images of patients with GA, a healthy comparable group and a comparable group with other retinal diseases (ORDs) were used to train a multi-layer deep convolutional neural network (DCNN) (1) to detect GA and (2) to differentiate in GA between a diffuse-trickling pattern (dt-GA) and other GA FAF patterns (ndt-GA) in FAF images.1.For the automated detection of GA in FAF images, two classifiers were built (GA vs. healthy/GA vs. ORD). The DCNN was trained and validated with 400 FAF images in each case (GA 200, healthy 200, or ORD 200). For the subsequent testing, the built classifiers were then tested with 60 untrained FAF images in each case (AMD 30, healthy 30, or ORD 30). Hereby, both classifiers automatically determined a GA probability score and a normal FAF probability score or an ORD probability score.2.To automatically differentiate between dt-GA and ndt-GA, the DCNN was trained and validated with 200 FAF images (dt-GA 72; ndt-GA 138). Afterwards, the built classifier was tested with 20 untrained FAF images (dt-GA 10; ndt-GA 10) and a dt-GA probability score and an ndt-GA probability score was calculated.For both classifiers, the performance of the training and validation procedure after 500 training steps was measured by determining training accuracy, validation accuracy, and cross entropy.ResultsFor the GA classifiers (GA vs. healthy/GA vs. ORD), the achieved training accuracy was 99/98%, the validation accuracy 96/91%, and the cross entropy 0.062/0.100. For the dt-GA classifier, the training accuracy was 99%, the validation accuracy 77%, and the cross entropy 0.166.The mean GA probability score was 0.981 ± 0.048 (GA vs. healthy)/0.972 ± 0.439 (GA vs. ORD) in the GA image group and 0.01 ± 0.016 (healthy)/0.061 ± 0.072 (ORD) in the comparison groups (p < 0.001). The mean dt-GA probability score was 0.807 ± 0.116 in the dt-GA image group and 0.180 ± 0.100 in the ndt-GA image group (p < 0.001).ConclusionFor the first time, this study describes the use of a deep learning-based algorithm to automatically detect and classify GA in FAF. Hereby, the created classifiers showed excellent results. With further developments, this model may be a tool to predict the individual progression risk of GA and give relevant information for future therapeutic approaches.