Online PhD defence Ballambat Suraj Pai

Keywords: Self Supervised Learning, Deep Learning, Radiology AI, Cancer Imaging

"Representation Learning in Radiology and Cancer Imaging"

This thesis explored representation learning approaches for advancing artificial intelligence applications in radiology and cancer imaging. The research developed foundation models capable of learning meaningful representations from medical images to improve cancer diagnosis, treatment planning, and outcome prediction. A key contribution was the creation of a novel biomarker measuring thymic health from CT scans, which demonstrated significant associations with immunotherapy treatment outcomes across multiple cancer types. The work extended beyond oncology to develop models for general radiological understanding, enabling applications across diverse anatomical regions. Additionally, the thesis addressed critical challenges in AI research reproducibility by developing cloud-based workflows and configuration-driven frameworks that make advanced AI techniques more accessible to the broader medical research community. Through these methodological innovations and clinical applications, this work demonstrated how representation learning can transform radiological analysis and enhance personalized patient care while promoting open science practices

Click here for the live stream.