Convolutional Neural Network-Based Skin Lesion Classification With Variable Nonlinear Activation Functions

One of the worst forms skin cancer is melanoma which can be curable if it diagnosed at an early stage. The earlier diagnosed, better outcome. risk death from directly related to delay in identifying a lesion. A Deep Learning-based computer diagnosing system automated solution clinical assessments overcome this problem. Convolutional Neural Network (CNN) help improve classification rate lesions dermoscopic images without need for any human assistance. linear and nonlinear activation functions act as node placed hidden layers or output play role deciding whether that should pass information following layer not. This critical mathematical mechanism influences accuracy CNN. To obtain acceptable performance, CNN requires large amount training data. research shows how fast effective different types work on with limited image datasets. Experimental analysis reveals proposed model parameterized Leaky ReLU function outperforms (97.50% accuracy, 98.00% precision, sensitivity) same network distinct problem recognition by classifying lesion into three classes. All experimental studies are carried out using PH2 (a database obtained Hospital Pedro Hispano Dermatology Service Matosinhos, Portugal) International Skin Imaging Collaboration (ISIC) archive