Image Processing for Diagnosing Psoriasis: A Machine Learning Approach to Classify Skin Lesions Into Psoriasis Subtypes Publisher



Masoorian H ; Gholamzadeh M ; Firooz A ; Safdari R
Source: Medical Journal of the Islamic Republic of Iran Published:2026

Abstract

Background: Psoriasis is a chronic autoimmune skin condition that affects 2-3% of the global population and manifests in various subtypes, including plaque, guttate, inverse, pustular, and erythrodermic psoriasis. Accurate subtype differentiation is crucial for effective treatment, but traditional diagnostic methods are time-consuming and prone to observer variability. This study aims to develop a machine learning model that classifies psoriasis lesions into the five primary subtypes using convolutional neural networks (CNNs) and transfer learning, offering a scalable tool to assist clinicians in diagnosing psoriasis and making informed treatment decisions. Methods: This is a methodological-developmental study that develops and evaluates a deep learning model for psoriasis subtype classification. The dataset was obtained from Kaggle, applying image augmentation techniques (rotation, translation, shearing, flipping, zoom) to enhance dataset diversity. A pre-trained Visual Geometry Group 16-layer architecture (VGG16) model was used for feature extraction, with a custom classification head added, incorporating ReLU-activated dense layers and dropout regularization to mitigate overfitting. The model was trained and evaluated using accuracy and loss metrics, with early stopping and model checkpointing for optimization. Results: The model achieved 96% accuracy on the training dataset and 90% on the test dataset, demonstrating strong generalization. A confusion matrix analysis confirmed accurate differentiation between the five subtypes. Conclusion: This study developed a deep learning model that accurately classifies psoriasis subtypes, utilizing CNNs and transfer learning. The model was integrated into a web-based tool, providing real-time diagnostic assistance for clinicians. This AI-driven system has the potential to enhance diagnostic accuracy, improve clinical workflows, and offer scalable solutions for psoriasis management, particularly in areas with limited access to dermatologists. Copyright © Iran University of Medical Sciences. This work has been published under CC BY-NC-SA 4.0 license.