Automated Diagnosis of Psoriasis in Whole Slide Images by Using Deep Learning

Psoriasis is a chronic inflammatory autoimmune skin disease that significantly impacts patients' quality of life and presents notable psychosocial challenges. The "gold standard" for diagnosing psoriasis in clinical settings is through pathologic biopsy, which reveals distinct features of skin lesions. With the rapid advancements in deep learning(DL) algorithms, computer-aided diagnostic(CAD) systems have emerged as a potential solution to identify psoriatic cells by leveraging morphological and cellular features present in Whole Slide Images (WSI). This paper proposes a DL-based CAD system for the automatic diagnosis of psoriasis. Our method extracts features from WSI without relying on precise annotations to achieve the diagnosis of psoriasis. Additionally, we introduce a novel convolutional neural network(CNN) architecture that balances classification accuracy with model complexity, making it suitable for clinical deployment. Our methodology involves performing patch-level classification and generating a heatmap that represents the WSI. Subsequently, we employ machine learning classifiers to extract feature information from the heatmap and perform WSI-level classification between psoriasis and normal tissue sections. Experiments are conducted using a psoriasis dataset obtained from Hainan General Hospital. The results demonstrate that our proposed CNN model achieves an accuracy of 92.90% at the patch level, with an AUC of 0.9289. At the WSI level, our method achieves a 98.18% accuracy, showcasing competitive performance in diagnosing psoriasis based on unannotated pathology images.