PhD defence Teresa Miguel Tareco Bucho

Keywords: Radiological response assessment, Variability, Total tumor burden, Artificial intelligence

"A Map is Not the Territory: on the Problem of Variability in Radiological Response Criteria"

When a new cancer treatment enters a clinical trial, a fundamental question is whether it truly benefits patients. The most direct answer lies in overall survival, but this takes years to measure and can be influenced by factors beyond the treatment itself. To make earlier decisions, trials rely on imaging-based endpoints.

Repeated imaging scans allow clinicians to track changes in tumor size over time. These changes are translated into response categories using standardized guidelines known as RECIST. Under RECIST, a limited number of lesions are selected and measured, while the remaining disease is assessed qualitatively. This structure enables consistent reporting across studies, but it also introduces variability. Differences in lesion selection, measurement and interpretation can alter how treatment efficacy is classified.

This thesis examines where this variability arises, how well RECIST reflects the underlying tumor burden and whether AI-based methods can support more comprehensive response assessment

Click here for the live stream.