AI-Driven Non-Invasive Skin Lesion Classification

Abstract: This project aims to use the HAM-10,000 dataset, along with other collected data, to design a skin-lesion classifier using Convolutional Neural Networks (CNN), and other various algorithms, including Random Forest, Logistic Regression, Support Vector Machine, and more. An app will be designed with this algorithm at its back end. This project will differentiate between melanocytic nevuses (moles), vascular lesions (birthmarks), benign keratosis (common non-cancerous growth), dermatofibroma (lesion caused by puncture or bug bite), actinic keratosis (precancerous lesion caused by UV exposure), basal cell carcinoma (cancer of the stratum basale, most common skin cancer), and malignant melanoma (deadliest skin cancer). Images on other skin cancer, gathered from other image directories, such as squamous cell carcinoma and merkel cell carcinoma, will also be used. K-Fold Cross Validation will be used to determine the best algorithm. This project will have many community benefits, including saving lives due to early detection of metastatic melanoma, which claims up to 6,850 lives every year (Cancer.Net). It will differentiate between cancerous and benign skin lesions. To account for the data imbalance, more data will be gathered through image directories and possibly crowdsourcing. An app will be made to implement this algorithm.