Mayo Clinic Pairs QuantCRC Deep-Learning Pathology with ctDNA to Refine Stage III Colon Cancer Risk

Mayo Clinic researchers led by gastroenterologist Frank A. Sinicrope used a deep-learning pathology algorithm to quantify tumor microenvironment features in stage III colon cancer and showed that combining them with postoperative circulating tumor DNA (ctDNA) status improves prognostic discrimination, in a retrospective analysis of the phase III NCCTG N0147 (Alliance) trial presented at the 2026 ASCO Annual Meeting. Adding ctDNA status to clinical variables raised Harrell's C-index by 0.062 for disease-free survival and 0.044 for overall survival; adding the AI-derived features on top raised it by a further 0.018 and 0.022. "ctDNA positivity was associated with a more aggressive and stromal-rich TME, whereas ctDNA negativity is associated with more immune-infiltrated tumor," Sinicrope and coauthors wrote. Postoperative ctDNA is a strong marker of residual disease, but outcomes within ctDNA-positive and ctDNA-negative groups still vary, and the tumor microenvironment may influence both ctDNA detection and prognosis. The team applied QuantCRC, a deep-learning histopathology algorithm from Aiforia Technologies, to digitized hematoxylin and eosin whole-slide images of resected stage III colon adenocarcinomas from the FOLFOX-based adjuvant trial. Of the trial cohort, 1,817 cases met data quality requirements, and 15 pathologist-defined quantitative features were extracted. ctDNA status had been measured earlier with the tissue-free Guardant Reveal assay. The researchers related features to time to recurrence, disease-free survival and overall survival using univariate and multivariable Cox models adjusted for clinicopathologic and molecular factors including KRAS, BRAF V600E and mismatch repair status, with interaction tests by ctDNA status. ctDNA-positive tumors showed more lymphovascular invasion, high-grade histology, tumor budding, necrosis and stroma, and less inflammatory stroma. Regardless of ctDNA status, tumor budding, mucin, signet ring cells and inflammatory features were linked to outcomes. Among ctDNA-negative patients, lower tumor-infiltrating lymphocyte density and less necrosis identified a high-risk subgroup with poorer outcomes, which the authors say may have clinical utility. Analysis of ctDNA tumor fraction and tumor genotyping is ongoing.