DERMONET: LIGHTWEIGHT DIAGNOSTIC SYSTEM FOR DERMATOLOGICAL CONDITIONS USING SQUEEZENET FRAMEWORK

Source abstract

Skin malignancies are regarded as the most dangerous disease. Skin cancer has recently received much attention among people worldwide. An earlier diagnosis of skin cancer can lower the mortality rate. Skin cancer can be found and identified via dermoscopy. Automated tools using computer-aided diagnosis models become necessary because visually evaluating dermoscopic images is tedious and time-consuming. The healthcare industry has greatly benefited from recent machine learning advancements like deep learning. Modern technical designs and methodologies make detecting this type of cancer possible; however, automated classification in earlier phases is challenging due to the lack of contrast. As a result, a squeeze net algorithm-based automated computer system is developed for diagnosing skin illnesses. The HAM10000 dataset is gathered for skin lesions. Images of the four skin cancer conditions BCC, DF, MEL, BKL, and NV are included in the dataset. With a 92.25% overall accuracy, 85% precision, 84% recall, and 83% F1 score, the proposed dermonet model did well in classifying skin cancer conditions from the image samples.