Deep learning (DL) can accelerate the prediction of prognostic biomarkers from routine pathology slides in colorectal cancer (CRC). However, current approaches rely on convolutional neural networks (CNNs) and have mostly been validated on small patient cohorts. Here, we develop a new transformer-based pipeline for end-to-end biomarker prediction from pathology slides by combining a pre-trained transformer encoder with a transformer network for patch aggregation. Our transformer-based approach substantially improves the performance, generalizability, data efficiency, and interpretability as compared with current state-of-the-art algorithms. After training and evaluating on a large multicenter cohort of over 13,000 patients from 16 colorectal cancer cohorts, we achieve a sensitivity of 0.99 with a negative predictive value of over 0.99 for prediction of microsatellite instability (MSI) on surgical resection specimens. We demonstrate that resection specimen-only training reaches clinical-grade performance on endoscopic biopsy tissue, solving a long-standing diagnostic problem.

An increasing number of patients with locally advanced gastric carcinomas (GC) are being treated with preoperative chemotherapy before surgery. Histopathological tumor regression may have an important prognostic impact in addition to the UICC-TNM classification system. We evaluated the histopathological tumor regression in 480 surgical resection specimens of GC after neoadjuvant cisplatin-based chemotherapy, using an established system encompassing three tumor regression grades based on the estimation of the percentage of residual tumor tissue at the primary tumor site in relation to the macroscopically identifiable former tumor bed. Tumor regression was correlated to clinicopathological characteristics and patient survival. Of the patients in this study, 102 (21.2%) had complete or subtotal tumor regression (50% residual tumor). Tumor regression was significantly associated with posttreatment tumor category (pT), lymph node status (pN), lymphatic invasion status (pL), and resection status (P < 0.001). Major histopathological regression was less frequent in tumors of the distal stomach and tumors of nonintestinal type (P = 0.003). Tumor regression (P = 0.009) and postoperative Lymph node status (P < 0.001) were independent prognostic factors for survival in a multivariate analysis of tumor regression, ypT/N/L category, resection status, grading and Lauren´s classification. Assessment of histological tumor regression after preoperative chemotherapy in GC provides objective and highly valuable prognostic information in addition to posttherapeutic lymph node status. A standardized tumor regression grading system should be implemented in pathological reports of these tumors.