Effect of patch size and network architecture on a convolutional neural network approach for automatic segmentation of OCT retinal layers

Abstract

Deep learning strategies, particularly convolutional neural networks (CNNs), are especially suited to finding patterns in images and using those patterns for image classification. The method is normally applied to an image patch and assigns a class weight to the patch; this method has recently been used to detect the probability of retinal boundary locations in OCT images, which is subsequently used to segment the OCT image using a graph-search approach. This paper examines the effects of a number of modifications to the CNN architecture with the aim of optimizing retinal layer segmentation, specifically the effect of patch size as well as the network architecture design on CNN performance and subsequent layer segmentation. The results demonstrate that increasing patch size can improve the performance of the classification and provides a more reliable segmentation in the analysis of retinal layer characteristics in OCT imaging. Similarly, this work shows that changing aspects of the CNN network design can also significantly improve the segmentation results. This work also demonstrates that the performance of the method can change depending on the number of classes (i.e. boundaries) used to train the CNN, with fewer classes showing an inferior performance due to the presence of similar image features between classes that can trigger false positives. Changes in the network (patch size and or architecture) can be applied to provide a superior segmentation performance, which is robust to the class effect. The findings from this work may inform future CNN development in OCT retinal image analysis.

Effect of Altered OCT Image Quality on Deep Learning Boundary Segmentation

“…The baseline performance (evaluation of images with nonaltered image quality) demonstrated a high level of consistency across the four networks. Here, a median absolute error below 0.70 pixels for the retinal boundaries and 2.30 pixels for the choroid/scleral interface was obtained, which matches well with previously published results [24]- [26], [28].…”

“…However, deep learning with neural networks, in particular, is frequently used for OCT image segmentation. A number of different methods have been utilized including patch-based classification [24]- [27], semantic segmentation [28]- [34], adversarial learning [35], and transfer learning [36]. Additionally, some methods [37], [38] have used volumetric input data, consisting of multiple image slices instead of a standard single image.…”

Choroidal changes in human myopia: insights from optical coherence tomography imaging

“…Although manual validation and correction of automated segmentation performance is still required, these methods demonstrate the potential for reduced reliance on manual analysis to derive choroidal characteristics. Developments in image processing methods, such as improvements in deep learning methods and their application to image segmentation, hold the promise for more reliable methods of automated OCT choroidal segmentation in the future.…”

A convolutional neural network‐based model observer for breast CT images

“…Note that zero-padding was excluded from the convolutional layers to maximize GPU memory utilization. 38 Because the signal to be detected resides near the central region in our tasks, the zero-padding is not necessary. The fully connected layer was used to process the feature map of the last convolution layer, resulting in the output of the model in likelihood form, which indicates the possibility that the signal is present in each input image.…”