An interpretable transformer network for the retinal disease classification using optical coherence tomography

Abstract

Retinal illnesses such as age-related macular degeneration and diabetic macular edema will lead to irreversible blindness. With optical coherence tomography (OCT), doctors are able to see cross-sections of the retinal layers and provide patients with a diagnosis. Manual reading of OCT images is time-consuming, labor-intensive and even error-prone. Computer-aided diagnosis algorithms improve efficiency by automatically analyzing and diagnosing retinal OCT images. However, the accuracy and interpretability of these algorithms can be further improved through effective feature extraction, loss optimization and visualization analysis. In this paper, we propose an interpretable Swin-Poly Transformer network for performing automatically retinal OCT image classification. By shifting the window partition, the Swin-Poly Transformer constructs connections between neighboring non-overlapping windows in the previous layer and thus has the flexibility to model multi-scale features. Besides, the Swin-Poly Transformer modifies the importance of polynomial bases to refine cross entropy for better retinal OCT image classification. In addition, the proposed method also provides confidence score maps, assisting medical practitioners to understand the models’ decision-making process. Experiments in OCT2017 and OCT-C8 reveal that the proposed method outperforms both the convolutional neural network approach and ViT, with an accuracy of 99.80% and an AUC of 99.99%.

TransU2Net: A Hybrid Network for Retinal Layer Segmentation in OCT

“…This approach combines cross convolution with a transformer structure that includes boundary regression and feature polarization, complemented by a CNN and a modified lightweight transformer structure, enhancing the performance of retinal layer segmentation. Jingzhen He et al 28 proposed the SwinPoly Transformer network, which builds connections between adjacent windows in the previous layer, boosting information exchange and allowing segmentation with multi-scale features.…”

MacD‐Net: An automatic guided‐ensemble approach for macular pathology detection using optical coherence tomography images

“…On the APTOS2019 dataset, their suggested technique produced a classification accuracy of 87.43%. A Swin Transformer has been suggested by He et al 21 for the classification of retinal OCT images. The Swin transformer identifies the connection between adjacent, non‐overlapping image patches.…”

AlterNet-K: a small and compact model for the detection of glaucoma

“…This shifted window method results in better efficiency by focusing self-attention on small, local groups of patches. He et al [ 10 ] proposed a network that combined Swin-Transformer and PolyLoss [ 24 ] to diagnose several retinal diseases in OCT images automatically. To visually understand the decision-making ability of the model, they create Class Activation Mappings (CAMs) using the score-CAM [ 25 ] method.…”

“…However, these transformer models must be pre-trained on enormous amounts of data to achieve competitive results. Several recent works [ 9 , 10 ] explore the effectiveness of modern transformer architectures based on ViT in diagnosing retinal diseases.…”