Rajat Rawat

Assessing the performance of machine learning in melanoma detection

Started Apr. 13, 2023

Abstract or project description

I can not include my project for publication anywhere else, as I am submitting to a journal, so here is just a small part of my project's summary: Melanoma is one of the forms of skin cancer and is dangerous as it easily spreads to other organs if malignant. Around 8000 people lost their lives to melanoma in 2023. However, diagnosing melanoma early on can significantly improve chances of survival for a patient. This paper tests Machine Learning algorithms to classify whether a skin lesion is malignant melanoma or non-malignant. We hypothesize that employing machine learning methodologies accurately classify malignant melanoma skin lesions. In 2020, the Society for Imaging Informatics in Medicine (SIIM) hosted a competition providing a dataset of skin lesion images with associated meta-data of the lesion. Data Scientist Chris Deotte combined the thirty thousand image dataset provided by SIIM with 40k external images of skin lesions diagnosed either with or without malignant melanoma. The meta-data associated with each image includes the patient's sex, age, and location of lesion.