TY - JOUR
T1 - Transformer-based biomarker prediction from colorectal cancer histology
T2 - A large-scale multicentric study
AU - TransSCOT consortium
AU - Wagner, Sophia J.
AU - Reisenbüchler, Daniel
AU - West, Nicholas P.
AU - Niehues, Jan Moritz
AU - Zhu, Jiefu
AU - Foersch, Sebastian
AU - Veldhuizen, Gregory Patrick
AU - Quirke, Philip
AU - Grabsch, Heike I.
AU - van den Brandt, Piet A.
AU - Hutchins, Gordon G.A.
AU - Richman, Susan D.
AU - Yuan, Tanwei
AU - Langer, Rupert
AU - Jenniskens, Josien C.A.
AU - Offermans, Kelly
AU - Mueller, Wolfram
AU - Gray, Richard
AU - Gruber, Stephen B.
AU - Greenson, Joel K.
AU - Rennert, Gad
AU - Bonner, Joseph D.
AU - Schmolze, Daniel
AU - Jonnagaddala, Jitendra
AU - Hawkins, Nicholas J.
AU - Ward, Robyn L.
AU - Morton, Dion
AU - Seymour, Matthew
AU - Magill, Laura
AU - Nowak, Marta
AU - Hay, Jennifer
AU - Koelzer, Viktor H.
AU - Church, David N.
AU - Church, David
AU - Domingo, Enric
AU - Edwards, Joanne
AU - Glimelius, Bengt
AU - Gogenur, Ismail
AU - Harkin, Andrea
AU - Hay, Jen
AU - Iveson, Timothy
AU - Jaeger, Emma
AU - Kelly, Caroline
AU - Kerr, Rachel
AU - Maka, Noori
AU - Morgan, Hannah
AU - Oien, Karin
AU - Orange, Clare
AU - Tomlinson, Ian
AU - Schnabel, Julia A.
N1 - Publisher Copyright:
© 2023 The Author(s)
PY - 2023/9/11
Y1 - 2023/9/11
N2 - 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.
AB - 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.
KW - artificial intelligence
KW - biomarker
KW - colorectal cancer
KW - deep learning
KW - microsatellite instability
KW - multiple instance learning
KW - transformer
UR - http://www.scopus.com/inward/record.url?scp=85169513346&partnerID=8YFLogxK
U2 - 10.1016/j.ccell.2023.08.002
DO - 10.1016/j.ccell.2023.08.002
M3 - Article
C2 - 37652006
AN - SCOPUS:85169513346
SN - 1535-6108
VL - 41
SP - 1650-1661.e4
JO - CANCER CELL
JF - CANCER CELL
IS - 9
ER -
