[论文解读] 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 Analysis被引用 5

一句话总结

该论文比较了八种用于眼底图像分类的预训练CNN，并结合XAI方法评估多种先进的血管分割模型，在两个数据集（FIVES 和 DRIVE）上进行评估。它报告了较高的分类准确率（例如 ResNet101 94.17%）和强的分割性能（例如 Swin-Unet 86.19% 均值像素准确度）。

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%.

研究动机与目标

• 评估八个用于眼底图像分类的预训练模型，并结合多种XAI技术以提高可解释性。
• 评估先进的分割架构（注意力U-Net变体、TransUNet、Swin-Unet）在视网膜血管描绘中的表现。
• 在公共数据集上使用标准指标比较分类与分割任务的性能。
• 识别最有效的模型与可解释性方法，以实现对眼底图像分析的可信度提升。

提出的方法

• 基准测试八个预训练CNN：ResNet101、DenseNet169、Xception、InceptionV3、DenseNet121、InceptionResNetV2、ResNet50、EfficientNetB0。
• 应用XAI技术：Grad-CAM、Grad-CAM++、Score-CAM、Faster Score-CAM、Layer CAM来解释预测。
• 使用Attention U-Net变体、TransUNet、Swin-Unet等分割模型，并以DenseNet和ResNet系列等骨干网络为 backbone 进行评估。
• 数据集：在FIVES上进行分类，在DRIVE和FIVES上进行分割；报告指标如分类的准确率、精确率、召回率、F1、Jaccard、以及对数损失等；分割指标如IoU、Dice、均值像素准确度、均值修正Hausdorff距离、均值表面Dice等。
• 比较性能：报告最高准确率（分类中ResNet101 94.17%）以及分割中Swin-Unet 86.19% 的均值像素准确度。
• 讨论可解释性结果以及不同架构下的模型行为。

(a) AMD

实验结果

研究问题

• RQ1 哪些预训练CNN在结合多种XAI解释时，在眼底图像分类任务中提供最高准确率？
• RQ2 相比基线模型，先进的分割模型（注意力U-Net变体、TransUNet、Swin-Unet）在视网膜血管分割上的表现如何？
• RQ3 不同的XAI技术对眼底分类模型的可解释性有何影响？
• RQ4 哪些骨干网络在DRIVE和FIVES数据集上实现最佳的分割指标？

主要发现

• ResNet101在分类中达到最高准确率94.17%。
• EfficientNetB0在分类中的准确率最低，为88.33%。
• Swin-Unet在分割中的均值像素准确度达到86.19%。
• 使用DenseNet201骨干的Attention U-Net分割均值像素准确度最低，为75.87%。
• 研究使用Grad-CAM、Grad-CAM++、Score-CAM、Faster Score-CAM和Layer CAM来揭示模型决策。

(b) DR