PVEMLPTS: design of an efficient psoriasis and vitiligo detection model through enhanced machine learning and personalized treatment strategies

The current dermatology methodologies to handle Psoriasis and Vitiligo has limitations in terms of accuracy, specificity, and timely diagnosis. In addition, limited dataset becomes the challenge in skin disease detection due to the privacy of patient data. Traditional methods rely heavily on manual diagnostics, leading to potential delays and less precise treatment plans. This paper introduces an innovative approach that integrates enhanced Generative Adversarial Networks (GANs), transfer learning, finetuning, and multimodal learning strategies to address these shortcomings. The proposed model overcomes these barriers by employing Progressive Growing of GANs (ProGAN) combined with StyleGAN, chosen for their proficiency in generating high-resolution, detailed images of skin lesions. We used pre-trained models—ResNet, InceptionV3, and EfficientNet, which capitalizes their proven success in image classification tasks. We reduced the training time and enhance model efficiency, by fine-tuning these models with a specialized dataset of skin diseases. Additionally, we implemented personalized treatment recommendation algorithms using a blend of Collaborative Filtering and Content-Based Filtering, ensuring tailored treatment plans based on patient-specific data and histories. The synergy of these methodologies in proposed model not only elevates disease detection capabilities but also pre-emptively identifies potential skin conditions. Clinical testing on dataset revealed that proposed model outperforms existing methods, reporting increase in precision (3.9% for detection, 1.5% for pre-emption), accuracy (4.5% and 1.9%), recall (3.5% and 2.4%), specificity (4.3% and 1.4%), and a significant reduction in diagnostic delays.