FUSION OF DEEP LEARNING AND HANDCRAFTED FEATURES FOR MELANOMA DETECTION

Abstract

Melanoma is resulting in high numbers of fatalities in humans, and is regarded as being one of the most aggressive and dangerous forms of skin cancer. Early diagnosis must be made to maximize the chances of survival, but scalability, precision, and interpretability of the machine and human diagnostic approaches currently being utilized are constrained. This paper offers a hybrid method that merges deep learning features derived using the InceptionV3 convolutional neural network with handcrafted ones like "Color Histograms" and "Histogram of Oriented Gradients (HOG)". A "Support Vector Machine (SVM)" is employed for classification once a fusion technique unifies the complementary strengths of the two types of features. The model is evaluated using the "ISIC2019 Skin Cancer HAM10000" dataset. The 98% accuracy of the proposed method demonstrates the efficacy of concatenating handcrafted features with deep learning for strong melanoma diagnosis. With its solution to key issues in automated skin lesion analysis with a sound and interpretable framework, this paper showcases a set of opportunities for effective clinical application.