Improving Automated Retinal Layer Segmentation from OCT Images: Addressing Limitations of CNN-STSF

Hassan et al [10] introduced the CNN-STSF framework, combining a deep convolutional neural network and a structural tensor-based segmentation framework, to automatically segment eight retinal layers from both normal and diseased Optical Coherence Tomography (OCT) images. While promising, this approach presents certain limitations. The use of overly complex backbones can increase inference time, potentially hindering real-time applications. Additionally, the framework's disregard for semantic information may impact its ability to accurately delineate subtle layer boundaries. Furthermore, CNN-STSF may be susceptible to variations in image quality and the presence of artifacts, ultimately affecting the precision of segmentation results. Its scalability to larger datasets and adaptability to other imaging modalities also require further investigation. Future research should focus on addressing these limitations to improve the accuracy, efficiency, and generalizability of automated retinal layer segmentation methods.