Transfer LearningDriven Detection of Autoimmune Pathologies from HighDimensional Diagnostic Modalities

Autoimmune skin diseases, which encompass a variety of disorders where the body's immune system attacks healthy skin cells, can present significant diagnostic challenges due to their complex and diverse manifestations. Early detection is crucial for effective treatment and management. This study proposes a transfer learning-based approach to detect autoimmune skin diseases, leveraging high-dimensional diagnostic modalities, including skin images. The dataset utilized in this research is sourced from Kaggle, consisting of images representing multiple autoimmune skin conditions, such as Psoriasis, Dermatomyositis, Morphea, Pityriasis Alba, and Vitiligo, alongside normal skin images. The detection framework integrates advanced deep learning architectures, including ConvNeXt V2, MobileNet V3, Swin Transformer V2, and MaxViT, all of which are pre-trained on large-scale datasets and fine-tuned to the specific characteristics of autoimmune skin diseases. By employing transfer learning, the models are able to benefit from the feature-rich representations learned from general image datasets, improving the accuracy and robustness of disease classification. The proposed method demonstrates superior performance in distinguishing between various autoimmune skin conditions and healthy skin, facilitating more accurate and efficient diagnosis. The findings highlight the potential of using transfer learning techniques for medical image analysis, particularly in dermatology, where visual appearance is a critical factor. This research offers a promising solution for automated detection, potentially aiding healthcare professionals in early disease identification and treatment planning.
Keywords: Autoimmune skin diseases, transfer learning, ConvNeXt V2, MobileNet V3, Swin Transformer V2, MaxViT, image classification, dermatology, Psoriasis, Vitiligo, Dermatomyositis, Morphea, Pityriasis Alba.