HSCFNet: Lightweight Convolutional Neural Network for the Classification of Infectious and Non-Infectious Skin Diseases

Abstract

Accurate differentiation between infectious and non-infectious skin diseases is crucial in the field of dermatological diagnosis, and although deep learning techniques have achieved remarkable results in the classification of a wide range of dermatological diseases, there is still a lack of effective unified frameworks to achieve this goal. To this end, this paper proposes a lightweight convolutional neural network, HSCFNet, for classifying 9 mainstream infectious skin diseases and 10 non-infectious skin diseases. HSCFNet consists of two core modules, the multi-gate hybrid convolution module (MGHC) and triple residual fusion module (TRF). MGHC integrates standard convolution and improved deformable convolution to form two branches and selects different branches for feature extraction through parameter control, while introducing a gating mechanism for feature selection of the extracted features to strengthen the ability of extracting important features. The TRF module facilitates the information interaction between features by fusing three different resolutions of features, which further improves the classification performance of the model. The experimental results show that the accuracy, precision, recall, specificity, and F1 score of HSCFNet reach 97.87%, 97.76%, 97.26%, 99.88%, and 97.43%, respectively, and the size of the model is only 26.1 MB, which is lightweight while maintaining high performance. Compared with 10 existing mainstream classification models, HSCFNet demonstrates the best classification performance. This study provides an efficient and lightweight solution for clinical skin disease diagnosis, which is important for accurately distinguishing mainstream infectious and non-infectious skin diseases.