Integrating radiomics and immunology: non-invasive assessment of CD8+ T cell levels

Radiomic features, commonly used in machine learning models, have demonstrated efficacy in predicting tumor-related outcomes. Nonetheless, the clinical application of these features is impeded by their inherent non-biological interpretability. It is, therefore, critically important to unravel the biological meaning of these radiomic features. In this paper, we modulate the tumor microenvironment of a murine cancer model through the administration of monoclonal antibodies, subsequently imaging and isolating the tumors for CD8+ T cell quantification. Through the extraction and analysis of radiomic features, we seek to bridge the gap between these quantitative variables and their underlying biological meaning. An exploratory factor analysis leads to the extraction of two factors. Factor 2 emerges as a potential indicator of CD8+ T cell levels. These encouraging results not only contribute to advancing our understanding of the synergistic potential of immunology and radiomics but also establish the foundation to address the explainability issue of radiomics-based prediction models.