Recognition of Skin Cancer from Dermoscopic Images via DNN
Automatic melanoma detection from dermoscopic skin samples is one of the most challenging tasks in medical image analysis. Deep Convolutional Networks (DCN) have emerged as practical tools in addressing these challenges, particularly in skin cancer diagnosis. The research aims to explore the performance of various neural networks in skin cancer recognition, assessing metrics such as accuracy (ACC), Receiver Operating Characteristic (ROC), and Area Under the Curve (AUC). The study uses modified DCN classifiers, namely ResNet50, InceptionV3, and DenseNet121, on the HAM10000 dataset. The results demonstrate that the modified InceptionV3 exhibited high performance compared to other networks, achieving a weighted accuracy of 86.58% and a weighted ROC of 0.96. This model's advantage lies in its efficient feature extraction capability and acceptable inference time per epoch. The study concludes that DCN provides a reliable framework for automating melanoma diagnosis and expediting identification, potentially saving lives. The paper also delves into the dataset preprocessing, model optimization strategies, and detailed performance analysis of the modified models. Key findings include the superior performance of the modified InceptionV3 in skin cancer classification and its potential to enhance early skin cancer detection, aligning with developing an efficient melanoma detection method.