Digital pathology in combination with deep learning is revolutionizing cancer care and oncology
One key application is the prediction of patient outcomes directly from digitized tumor slides as deep learning-based algorithms can extract morphological information with prognostic value from the image.
The research group at the Leiden University Medical Center (LUMC) has developed a similar deep learning model for predicting distant recurrence of endometrial cancer patients – named HECTOR (Histopathology-based Endometrial Cancer Tailored Outcome Risk). HECTOR was developed using a total cohort 2,751 patients with digitized H&E-stained tumor slides, outcome and molecular data, including the PORTEC randomized trials. Specifically, HECTOR uses two input data broadly available, one digitized H&E-stained tumor slide of the hysterectomy section and the information about the tumor spread beyond the uterus.
Just published in Nature Medicine, HECTOR shows its accuracy in predicting distant recurrence outcomes when tested in three unseen cohorts with C-indices of 78.9%, 82.8% and 81.5%. For instance, in the first test set with 353 unseen patients, 10-year distant recurrence-free probability by predicted HECTOR low, intermediate, or high risk was 97.0%, 77.7% and 58.1%, exemplifying the capacity of HECTOR to be used as a prognostic tool. Importantly, HECTOR outperforms the prognostic variables used in a diagnostic workflow including histopathological data and the molecular classification. In the PORTEC-3 randomized trial, the subset of patients predicted high-risk of distant recurrence by HECTOR benefit the most from adjuvant chemotherapy, outperforming current biomarkers.
Given prospective validation, HECTOR has the potential to provide more accurate, quicker, and more personalized prediction of distant recurrence risk for any patient with endometrial cancer. This eventually may support clinicians for treatment-decision making and improve care management of patients with endometrial cancer.
An intuitive video is also provided here.
