Melanoma detection through AI models

This project explores the use of AI models to detect skin cancer from images of skin lesions, achieving 80-90% accuracy in predicting malignancy and revealing key physical attributes that could aid early detection.

Skin cancer is one of the most prominent forms of cancer, and early detection can significantly improve the recovery process for those affected. Through this project, I aimed to understand the ability to detect cancer from images of skin lesions using artificial intelligence (AI) models. I attempted to build software which would allow users to predict the possibility of cancerous lesions by processing images through these models. The approach involved the use of different AI algorithms which were trained on a dataset of thousands of individual images, and identifying which models yielded the best results. Just from graphing the data, it was shown that the malignant images were darker on average, with lower average color values for each pixel. The test accuracy of most models ranged between 80 - 90%, proving that the AI models were able to correctly predict malignant v/s non-malignant 80 - 90% of the time. Through my research, I discovered that AI models have the ability to correctly predict melanoma by processing images using a variety of algorithms through which the data is trained. Additionally, I learned of the numerous physical attributes of an image which could help people predict the malignancy of skin lesions before using an AI model for prediction.

Explore More!

Predicting Dementia Risk with Machine Learning

This research addresses the urgent need for early dementia detection by using machine learning to analyze patient health factors, such as heart rate and diabetes, for dementia prediction, achieving a 52% accuracy with a random forest model and paving the way for future model improvements and deeper insights into influential risk factors.

Min W.

Mentored by John Basbagill

workspace_premium

Evaluating the Efficacy of the 3D U-Net Architecture For Glioblastoma Multiforme Tumor Segmentation

This research evaluates the performance of the 3D U-Net model for automated glioblastoma tumor segmentation from MRIs, achieving 98.6% accuracy and significantly faster processing times than human oncologists, crucial for effective radiation therapy.

Arnav J.