PhD defence Shruti Atul Mali

Keywords: Radiomics, Harmonization, Robustness, Reproducibility

"Advancing Reliable Radiomics: Harmonization, Foundation Models, and Robustness in Cancer Imaging"

This thesis investigated how medical images can be used more reliably to support cancer care. Modern image analysis methods can detect patterns in scans that are not visible to the human eye, but their results are often affected by differences between hospitals, scanners, and imaging settings. This makes it difficult to apply such methods safely in everyday clinical practice. The research therefore focused on improving the reliability of these image-based models through harmonization methods, which aim to reduce unwanted technical differences in imaging data. The thesis also studied the value of foundation models, which are large pre-trained artificial intelligence models, for tasks such as cancer classification and risk prediction. In addition, it introduced new frameworks to evaluate the quality, maturity, and clinical readiness of radiomics research. Overall, the thesis shows that improving robustness, reproducibility, and evaluation standards is essential before AI-based imaging tools can be translated into real clinical benefit for patients.

Click here for the live stream.