Signals in the Spread: Quantitative Imaging and Tumor Heterogeneity in Metastatic Cancer

Radiomics is an emerging field that transforms standard medical imaging into quantitative data, offering the potential to non-invasively characterize tumor biology and predict clinical outcomes. However, most radiomic studies adhere to a “one patient, one lesion” paradigm, which overlooks the biological and spatial heterogeneity that defines multi-metastatic disease. This thesis proposes a new framework for radiomic analysis that explicitly embraces lesion-level variability to improve patient assessment in advanced cancers. The work begins with a systematic benchmarking of ten multi-lesion feature aggregation strategies across three datasets representing regional, organ-confined, and widespread metastatic disease. Results demonstrate that no single method performs optimally across all contexts, underscoring the need for task-specific modeling approaches. Building on this, a novel imaging-derived metric—Measured Intrapatient Radiomic Variability (MIRV)—is introduced to quantify intertumor heterogeneity. In a cohort of soft-tissue sarcoma patients, MIRV was associated with volumetric response variability, ctDNA positivity, and survival outcomes, suggesting its potential as a non-invasive biomarker of treatment heterogeneity. The thesis then shifts toward predictive modeling, developing lesion-specific classifiers to forecast treatment response at the level of individual pulmonary metastases. Using radiomic features from leiomyosarcoma patients, these models outperformed volume-based predictors and revealed the peritumoral region as a key source of biological signal. In a secondary analysis, the same modeling framework was repurposed to explore radiomic signatures associated with tumor hypoxia. By leveraging the design of the SARC021 trial and validating against an independent dataset, preliminary evidence is presented suggesting that radiomics may capture microenvironmental features relevant to hypoxia-driven treatment response. Collectively, these studies contribute new tools, biomarkers, and modeling strategies that reposition radiomics as a system-level approach capable of capturing the spatial and biological complexity of multi-metastatic disease. The findings support a more nuanced, lesion-informed vision for precision oncology and offer a foundation for both clinical application and biological discovery.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.