Artificial intelligence (AI) -powered spatial analysis of tumor-infiltrating lymphocytes (TIL) for prediction of prognosis in stage II-III resectable colon cancer.

Abstract

211 Background: T-cell infiltration in tumors and the surrounding stroma has been suggested as an important prognostic marker in colorectal cancer, but assessment usually requires additional tissue processing and interpretational efforts. The aim of this study is to assess the clinical significance of AI-powered spatial TIL analysis using only a hematoxylin and eosin (H&E)-stained whole-slide image (WSI) for the prediction of prognosis in stage II-III colon cancer patients treated with surgery and adjuvant chemotherapy. Methods: H&E-stained WSI of primary tumors and clinical data were collected retrospectively from stage II or III colon cancer patients treated with curative surgery and adjuvant chemotherapy at Seoul National University Bundang Hospital (Seongnam, Korea) between 2009 and 2012. For spatial TIL analysis, we used Lunit SCOPE IO, an AI-powered H&E WSI analyzer, which identifies and quantifies TIL within cancer or stroma area from a representative H&E-stained WSI. Intratumoral (iTIL) and stromal TIL (sTIL) densities were defined as the number of TILs in the intratumoral area (/mm2) and the surrounding stroma (/mm2), respectively. Results: A total of 289 patients were included in this analysis. During the follow-up period (median 99.6 months, 95% confidence interval [CI] 95.8 - 103.5), 28 (9.7%) recurrences were observed. The median iTIL and sTIL densities in all patients were 44.4 (/mm2, interquartile range [IQR] 28.4 – 71.7) and 878.0 (/mm2, IQR 554.9 – 1209.6), and showed a modest positive correlation (Pearson’s r = 0.349, p < 0.001). Patients having tumors with high microsatellite instability (MSI-H) status were more likely to have higher iTIL densities (mean iTIL 161.5/mm2 in MSI-H vs. 58.9/mm2 in MSI-L/MSS, p = 0.024), and patients with stage II disease (vs. stage III, iTIL and sTIL, p = 0.046 and 0.049, respectively) or without perineural invasion (vs. invasion (+), iTIL, p = 0.001) were also more likely to have higher mean iTIL and/or sTIL densities. Analyzing the iTIL, sTIL densities in association to clinical outcomes, the patients with confirmed recurrences had significantly lower mean sTIL density compared to those without recurrences (630.2/mm2 in recurrence (+) vs. 1021.3/mm2 in recurrence (-), p < 0.001). Comparing the risk of recurrence using the quartile cutoffs, the patients having either sTIL or iTIL densities in the lowest quartile were more likely to recur (unadjusted hazard ratio[HR]s of lowest quartile vs. others in sTIL, 2.77 [95% CI 1.32 - 5.82], p = 0.009; iTIL, 2.68 [95% CI 1.28 – 5.64], p = 0.009). Patients having both sTIL and iTIL in the lowest quartile (N=31) had the highest risk of recurrence, exhibiting a recurrence rate at 3 years of 26.1% (HR 4.30, 95% CI 1.94-9.51, p <0.001). Conclusions: AI-powered TIL analysis can provide prognostic information for stage II-III colon cancer in a practical manner.