Multi-Scale Classification of Rare Diseases Using ResNet

Abstract

This study proposed a multi-scale classification algorithm based on ResNet for the automated diagnosis of rare diseases. Due to the low incidence and complex pathological characteristics of rare diseases, traditional classification models have difficulty in effectively identifying these diseases. To address this challenge, we used the residual structure and multi-scale feature extraction capabilities of ResNet to accurately classify rare disease images. The experiment compared the performance of five deep learning models, ResNet, VGG, Inception, DenseNet, and EfficientNet. The results showed that ResNet outperformed other models in terms of accuracy, recall, precision, and F1 score, proving its superiority in multi-scale lesion recognition. In addition, this study further improved the generalization ability of the model through data augmentation and transfer learning strategies, providing a more robust and efficient solution for rare disease classification. This method brings new possibilities for the automated diagnosis of rare diseases and is expected to promote the development of medical image analysis technology in clinical applications.