Automating Dermatology through CNN-based Comparison Models for Improved Efficiency
Rohit Kumar Singh, Swapnil Srivastava, Swati Sharma, Ayush Dullat, Taru Malik, Harshit Goel
Abstract
Skin diseases are common, affecting the health along with the quality of life of millions worldwide. Effective treatment and prevention of all complications depend on detection that is both early and accurate. Typical diagnosis approaches depend on dermatologists' manual inspection and can be quite slow, fairly subjective, and somewhat inaccessible in many faraway places. This research employs a deep learning-based method to fully automate skin disease detection, delivering a considerably faster and more dependable solution. For image analysis, we thoroughly evaluated four different deep learning models: Custom CNN, ResNet50, VGG16, and EfficientNetB0, based on their individual abilities to classify multiple skin conditions. Accuracy, precision, recall, and F1-score were key metrics used for identifying the most effective model. Out of the few models that have been evaluated, Custom CNN has proven to add scope concerning accuracy, precision, and recall. Thus, it acts as an effective classifier of skin diseases. Even a VGG16 architecture to a certain degree, has managed to yield promising results. The findings show deep learning is a strong way to find skin diseases early and right, increasing patient outcomes and increasing access to skin care.