Use of Deep Learning to Evaluate Tumor Microenvironmental Features for Prediction of Colon Cancer Recurrence

• DOI: 10.1158/2767-9764.crc-24-0031
• Publication Date: 2024-05-06T12:54:03Z
• Authors: Frank A. Sinicrope; Garth D. Nelson; Bahar Saberzadeh-...; Diana I. Segovia; Rondell P. Graham; Christina Wu; Catherine E. Hagen; Sameer Shivji; Paul Savage; Dan D. Buchanan; Mark A. Jenkins; Amanda I. Phipps; Carol Swallow; Loic LeMarchand; Steven Gallinger; Robert C. Grant; Reetesh K. Pai; Stephen N. Sinicrope; Dongyao Yan; Kandavel Shanmugam; James Conner; David P. Cyr; Richard Kirsch; Imon Banerjee; Steve R. Alberts; Qian Shi; Rish K. Pai
• Keywords: Male; 0301 basic medicine; Organoplatinum Compounds; Leucovorin; 610; Middle Aged; Prognosis; DNA Mismatch Repair; 03 medical and health sciences; Deep Learning; Chemotherapy, Adjuvant; Colonic Neoplasms; Antineoplastic Combined Chemotherapy Protocols; Tumor Microenvironment; Humans; Female; Fluorouracil; Neoplasm Recurrence, Local; Research Article; Aged

Abstract Deep learning may detect biologically important signals embedded in tumor morphologic features that confer distinct prognoses. Tumor were quantified to enhance patient risk stratification within DNA mismatch repair (MMR) groups using deep learning. Using a quantitative segmentation algorithm (QuantCRC) identifies 15 features, we analyzed 402 resected stage III colon carcinomas [191 deficient (d)-MMR; 189 proficient (p)-MMR] from participants phase trial of FOLFOX-based adjuvant chemotherapy. Results validated an independent cohort (176 d-MMR; 1,094 p-MMR). Association with clinicopathologic variables, MMR, KRAS, BRAFV600E, and time-to-recurrence (TTR) was determined. Multivariable Cox proportional hazards models developed predict TTR. differed significantly by MMR status. Cancers p-MMR had more immature desmoplastic stroma. Tumors d-MMR increased inflammatory stroma, epithelial tumor-infiltrating lymphocytes (TIL), high-grade histology, mucin, signet ring cells. Stromal subtype did not differ BRAFV600E or KRAS In tumors, multivariable analysis identified tumor-stroma ratio (TSR) as the strongest feature associated TTR [HRadj 2.02; 95% confidence interval (CI), 1.14–3.57; P = 0.018; 3-year recurrence: 40.2% vs. 20.4%; Q1 Q2–4]. Among extent stroma (continuous HRadj 0.98; CI, 0.96–0.99; 0.028; 13.3% 33.4%, Q4 Q1) N most robust prognostically. TSR independently validated. conclusion, QuantCRC can quantify differences routine sections determine their relative contributions prognosis, elucidate relevant pathophysiologic mechanisms driving prognosis. Significance: A reflect underlying prognosis groups. cancers. Extent prognostic tumors. TIL density either group.