Utilizing Deep Neural Networks for Enhanced Diagnosis of Dermatological Conditions

Automated Computer-Aided Diagnosis (CAD) is essential for accurate and timely diagnosis due to the variety of skin issues and the shortage of dermatologists. By using sickness taxonomy, improving classification performance, and exploring the potential of Deep Learning to classify different skin problems, this work aims to build on previous research on CAD for dermatology. The state-of-the-art DNNs (Deep Neural Networks) are trained using two publicly available datasets of skin images: DermNet and ISIC Archive. This study also improved the proposed model's classification accuracy by using disease taxonomy. With an Area Under the Curve (AUC) of 98% and an accuracy of 80%, 23 diseases are classified on DermNet with unprecedented results. Furthermore, an accuracy of 67% and an AUC of 98% is achieved, setting a record for identifying 622 distinct sub-classes in this dataset. With an AUC of 99% on ISIC Archive and an average accuracy of 93%, were all seven illnesses are identified. This research demonstrates the accurate classification of deep learning algorithm for a variety of skin disorders with almost human precision and far better repeatability. It has the potential to assist doctors in the wide-scale, real-time diagnosis of skin conditions via the use of clinical or dermoscopic images.