ABSTRACT: With nearly 45,000 rectal cancers to be diagnosed in 2023, the major clinical question of interest
is the identification of those patients who will achieve sustained clinical complete response (cCR, no detectable
macroscopic tumor) or true pCR (pathologic complete response, with very few/no tumor cells). With the recent
incorporation of options such as total neoadjuvant therapy (neoadjuvant chemotherapy and chemoradiation),
these complete responder patients can comprise up to ~50% of the population. If these “true” complete response
patients could be reliably identified, they could avoid the major complications of rectal resection (anastomotic
leakage, pelvic sepsis, ureteric injury) and quality-of-life issues (permanent ostomy, fecal incontinence), with no
impact on their overall survival. However, there are no validated objective diagnostic criteria (using endoscopy,
MRI, and digital rectal exams) or serum/blood-based markers (e.g., carcinoembryonic antigen, CEA) that can
definitively identify true complete responders for organ-preserving strategies and non-operative management in
rectal cancers. Quantitative non-invasive imaging markers that accurately characterize pathologic tumor
regression in rectal cancers after neoadjuvant therapy could be hugely impactful in selecting complete response
patients who can be safely recommended organ-preservation vs partial/non-responders who require surgery.
To specifically capture subtle pathologic cues on MRI resulting from the desmoplastic fibrosis and tumor
response after chemoradiation/TNT, we have identified unique computer-extracted image features of gradient
response and morphological disruptions within the lesion (wall, tumor) and peri-lesion environment (lumen,
peritumoral fat). This R01 will build on these exciting findings to develop and validate a Radiomic Tumor
Response (RadTR) classifier to accurately identify complete responders after neoadjuvant therapy in rectal
cancers, via routine MRI. The RadTR classifier will additionally incorporate a novel “spatial diversity” descriptor
that captures macro-scale tissue organization and stromal composition that are biological hallmarks of complete
response. Using a multi-institutional cohort of ~900 patients, our validation of the RadTR classifier will focus on
three key criteria. (i) To enable clinical interpretability, we will conduct detailed histomorphological evaluation of
RadTR features via spatially mapped MRI and whole-mount rectal pathology specimens. (ii) For confirming
diagnostic reliability, the RadTR classifier will be comprehensively compared against clinical evaluation criteria
(endoscopy, MRI, digital rectal exam) as well as within a multi-reader study; to confirm performance
improvements in identifying CR via MRI. (iii) Towards ensuring equity in population performance for future clinical
deployment, the RadTR classifier will be validated across multiple institutions and populations using a national
clinical trial cohort of TNT in rectal cancers (NRG GI-002). To be clinically actionable for recommending organ-
preserving treatment in rectal cancers, RadTR model will target a 50% improvement in sensitivity for identifying
complete response as well as ensuring 95% specificity in identifying partial/non-pCR patients needing surgery.