Machine Learning Neural Network Classifier Interfaced Skin Cancer Identification for Medical Diagnosis System

• Published in: 2024 International Conference on Advancement in Renewable Energy and Intelligent Systems (AREIS) pp. 1 - 6
• Main Authors: B, Chempavathy, Veerappan, Kavitha, P, Jose, Theodore, Shaji. K. A., Rajasree, R., Prasanna, S. Gnana
• Format: Conference Proceeding
• Language: English
• Published: IEEE 05.12.2024
• Subjects: Accuracy; Bilateral filter; DCNN; DNA; Feature extraction; GLCM; Image color analysis; Lesions; Machine learning; Medical diagnostic imaging; Skin; Skin cancer; Training; U-NET
• DOI: 10.1109/AREIS62559.2024.10893615

Skin cancer, which lethal, is one of the top three tumours caused by DNA damage. This damaged DNA causes cells to grow uncontrolled, and they are currently growing swiftly. Numerous trainings performed on the automatic diagnosis of cancer in images of skin lesions. Analysis of these images is rather challenging, though, because of some disruptive factor like light observations from the skin's surface, differences in color enlightenment, and different forms and dimensions of the lesions. The accuracy and skill of analyzers in the early stages improved by machine learning (ML) based autonomous skin cancer diagnosis. A Deep Convolutional Neural Network (DCNN) model for identifying malignant and benign skin lesions is presented in this research. Applying a bilateral filter as the initial step in preprocessing eliminates noise and artefacts. The second phase involves utilizing U-Net to segment the input images and GLCM to extract features that assist with correct categorization. Data classification, the third phase, increases the quantity of images and improves classification precision. This work implements a skin cancer detection model using the ISIC dataset, which contains a large collection of medical images for training and evaluating ML algorithms. Proposed DCNN model is more dependable and resilient, according to the results. The training accuracy is 95% and the training loss is 0.01 after 35 epochs.