Prediction of homologous recombination deficiency from routine histology with attention-based multiple instance learning in nine different tumor types
Chiara M. L. Loeffler, Omar S. M. El Nahhas, Hannah S. Muti, Zunamys I. Carrero, Tobias Seibel, Marko Treeck, Didem Cifci, Marco Gustav, Kevin Bretz, Nadine T. Gaisa, Kjong-Van Lehmann, Alexandra Leary, Pier Selenica, Jorge S. Reis-Filho, Nadina Ortiz-Bruechle, and Jakob Nikolas Kather
Homologous recombination deficiency (HRD) is a pan-cancer predictive biomarker for response to PARP inhibitors, but standard HRD testing is complex. This study develops a deep-learning pipeline using attention-based multiple instance learning (attMIL) to predict HRD status directly from routine H&E histology across nine tumor types, showing that a breast-cancer-trained classifier transfers to endometrial, prostate, and pancreatic cancer.
@article{loeffler2024hrd,title={Prediction of homologous recombination deficiency from routine histology with attention-based multiple instance learning in nine different tumor types},author={Loeffler, Chiara M. L. and El Nahhas, Omar S. M. and Muti, Hannah S. and Carrero, Zunamys I. and Seibel, Tobias and van Treeck, Marko and Cifci, Didem and Gustav, Marco and Bretz, Kevin and Gaisa, Nadine T. and Lehmann, Kjong-Van and Leary, Alexandra and Selenica, Pier and Reis-Filho, Jorge S. and Ortiz-Bruechle, Nadina and Kather, Jakob Nikolas},journal={BMC Biology},volume={22},number={225},year={2024},publisher={Springer Nature},doi={10.1186/s12915-024-02022-9},}