ARTIFICIAL INTELLIGENCE FRAMEWORK FOR SKIN CANCER DETECTION AND CLASSIFICATION USING RECURRENT NEURAL NETWORK
Abstract
Millions of people every year lose their lives to skin cancer, which is by far the most frequent kind of the illness. Preventing deaths and increasing survival rates from skin cancer requires early diagnosis of potentially dangerous instances and the provision of appropriate therapies. Most of the research that are relevant focus on machine learning- based algorithms, however these algorithms fall short when it comes to providing the best possible accuracy and specificity. To this end, the research project has deployed a classification mechanism predicated on state-of-the-art deep learning and recurrent neural networks (RNN). As a first step in accurately pinpointing the location of skin cancer, we use a segmentation technique based on k-means clustering (KMC). As a means of classifying skin cancers, a RNN was created by fusing texture characteristics from the gray level co-occurrence matrix (GLCM), low level data from the discrete wavelet transform (DWT), and statistical color features. This action was taken to preserve the system's highest possible level of efficiency. This suggests that the study's results might be applied correctly to the classification of benign and malignant skin cancers. Results from simulation studies demonstrate that the suggested method excels in both qualitative and quantitative analysis over the state-of-the-art methods.
Keywords
recurrent neural networks, discrete wavelet transform, gray level co-occurrence matrix, colour features, k-means clustering