Unifying Segmentation and Classification in Skin Disease Diagnosis: U-Net & CNN Hybrid Approach

Skin-related conditions are a common global health concern. Depression and physical health issues are both brought on by the hidden risks of infections. Furthermore, in extreme circumstances, Skin cancer could arise from it. Thus, one of the most difficult challenges in medical image processing is identifying skin disorders from clinical photographs. Furthermore, Medical practitioners' physical diagnosis of dermatological problems is laborious and subjective, Because of this, Patients and dermatologists alike need autonomous skin disease prognosis since it expedites the treatment approach. We describe here a morphological filtering-based digitized hair removal technique that makes use of the Black-Hat transformation and the inpainting algorithm. After that, the images are processed using a denoising convolutional neural network in order to remove noise. Furthermore, To improve skin lesion segmentation accuracy, we utilize the U-Net model. Then, using the attributes that were collected, we apply the computationally efficient deep learning technique Convolutional Neural Network (CNN) to effectively identify the skin photos as Actinic keratosis, Benign keratosis, Eczema, and Melonocytic Nevi. Our U-Net model gives a accuracy of 92% and CNN gives a accuracy of 70%.