[Paper Review] Explainable Convolutional Neural Networks for Retinal Fundus Classification and Cutting-Edge Segmentation Models for Retinal Blood Vessels from Fundus Images

Fatema Tuj Johora Faria, Mukaffi Bin Moin+3 more|arXiv (Cornell University)|May 12, 2024

Retinal Imaging and Analysis5 citations

TL;DR

The paper compares eight pre-trained CNNs for fundus image classification with XAI methods, and evaluates several advanced segmentation models for retinal vessels, using two datasets (FIVES and DRIVE). It reports high classification accuracy (e.g., ResNet101 94.17%) and strong segmentation performance (e.g., Swin-Unet 86.19% Mean Pixel Accuracy).

ABSTRACT

Our research focuses on the critical field of early diagnosis of disease by examining retinal blood vessels in fundus images. While automatic segmentation of retinal blood vessels holds promise for early detection, accurate analysis remains challenging due to the limitations of existing methods, which often lack discrimination power and are susceptible to influences from pathological regions. Our research in fundus image analysis advances deep learning-based classification using eight pre-trained CNN models. To enhance interpretability, we utilize Explainable AI techniques such as Grad-CAM, Grad-CAM++, Score-CAM, Faster Score-CAM, and Layer CAM. These techniques illuminate the decision-making processes of the models, fostering transparency and trust in their predictions. Expanding our exploration, we investigate ten models, including TransUNet with ResNet backbones, Attention U-Net with DenseNet and ResNet backbones, and Swin-UNET. Incorporating diverse architectures such as ResNet50V2, ResNet101V2, ResNet152V2, and DenseNet121 among others, this comprehensive study deepens our insights into attention mechanisms for enhanced fundus image analysis. Among the evaluated models for fundus image classification, ResNet101 emerged with the highest accuracy, achieving an impressive 94.17%. On the other end of the spectrum, EfficientNetB0 exhibited the lowest accuracy among the models, achieving a score of 88.33%. Furthermore, in the domain of fundus image segmentation, Swin-Unet demonstrated a Mean Pixel Accuracy of 86.19%, showcasing its effectiveness in accurately delineating regions of interest within fundus images. Conversely, Attention U-Net with DenseNet201 backbone exhibited the lowest Mean Pixel Accuracy among the evaluated models, achieving a score of 75.87%.

Motivation & Objective

• Assess eight pre-trained CNN models for fundus image classification with multiple XAI techniques to enhance interpretability.
• Evaluate advanced segmentation architectures (Attention U-Net variants, TransUNet, Swin-Unet) for retinal vessel delineation.
• Compare performance across classification and segmentation tasks using standard metrics on public datasets.
• Identify the most effective models and explainability methods for trust-worthy fundus image analysis.

Proposed method

• Benchmark eight pre-trained CNNs: ResNet101, DenseNet169, Xception, InceptionV3, DenseNet121, InceptionResNetV2, ResNet50, EfficientNetB0.
• Apply XAI techniques: Grad-CAM, Grad-CAM++, Score-CAM, Faster Score-CAM, Layer CAM to interpret predictions.
• Evaluate segmentation with Attention U-Net variants, TransUNet, Swin-Unet using backbones like DenseNet and ResNet series.
• Datasets: classification on FIVES; segmentation on DRIVE and FIVES; report metrics such as Accuracy, Precision, Recall, F1, Jaccard, and Log Loss (classification) and IoU, Dice, Mean Pixel Accuracy, Mean Modified Hausdorff Distance, and Mean Surface Dice (segmentation).
• Compare performance: report top accuracy (ResNet101 94.17% for classification) and segmentation (Swin-Unet 86.19% Mean Pixel Accuracy).
• Discuss explainability outcomes and model behavior across architectures.

(a) AMD

Experimental results

Research questions

• RQ1 Which pre-trained CNNs provide the highest accuracy in fundus image classification when paired with diverse XAI explanations?
• RQ2 How do advanced segmentation models (Attention U-Net variants, TransUNet, Swin-Unet) perform on retinal vessel segmentation compared to baselines?
• RQ3 What is the impact of different XAI techniques on the interpretability of fundus classification models?
• RQ4 Which backbone choices yield the best segmentation metrics on DRIVE and FIVES datasets?

Key findings

• ResNet101 achieved the highest classification accuracy at 94.17%.
• EfficientNetB0 had the lowest classification accuracy at 88.33%.
• Swin-Unet achieved a Mean Pixel Accuracy of 86.19% in segmentation.
• Attention U-Net with DenseNet201 backbone had the lowest segmentation Mean Pixel Accuracy at 75.87%.
• The study employs Grad-CAM, Grad-CAM++, Score-CAM, Faster Score-CAM, and Layer CAM to illuminate model decisions.

(b) DR