AI-driven identification of melanoma risk factors using choroidal nevi retinal images

This thesis explores the potential of using choroidal nevi retinal images to identify melanoma risk factors, employing advanced machine learning techniques. Utilizing a comprehensive dataset annotated by ocular oncology specialists, the study develops and validates models capable of distinguishing benign nevi from those at risk of transforming into melanoma. Our models achieve a peak Area Under the Curve (AUC) of 0.93 for identifying significant risk factors, outperforming baseline models such as ResNet-50. A significant focus of this research is on enhancing the interpretability of these AI models, ensuring that the diagnostic predictions are transparent and can be understood by clinicians. This approach not only improves trust in AI-driven diagnostics but also facilitates deeper insights into the decision-making process of the models. Moreover, the models demonstrate robust performance under various imaging conditions, including a maximum performance drop of only 5.28% at 40% zoom out, highlighting their utility in diverse clinical settings. The results demonstrate the efficacy of the models in identifying key risk factors and predicting nevi transformation, which could lead to earlier interventions and potentially improved patient outcomes in ophthalmology. This thesis sets the groundwork for future research aimed at integrating AI with traditional imaging techniques to create more robust, interpretable, and clinically applicable diagnostic tools.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2024

Author dc:creator

• Suri, Muhammad Huzaifa Khan

Contributors dc:contributor

• Varatharajah, Yogatheesan

• Ai
• Machine Learning
• Ophthalmology
• Choroidal Nevus
• Choroidal Melanoma
• Fundus Images
• Risk Factors
• Interpretability

Statement dc:rights

• Copyright 2024 Muhammad Huzaifa Suri

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/125648

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F125648