Photo Credit: Mohammed Haneefa Nizamudeen
A deep learning model called HECTOR stratified patients by determining the probability of endometrial cancer recurrence, which could have clinical implications.
A multimodal, deep learning prognostic model enhanced the prediction of distant recurrence in patients with endometrial cancer (EC), according to findings published in Nature Medicine.
“Although most women with localized disease are cured by surgery, 10% to 20% develop distant recurrence, which is typically incurable,” wrote Sarah Volinsky-Fremond, PhD candidate, and colleagues. “There remains a pressing unmet need for a method that can predict EC distant recurrence from input data generated as part of routine clinical diagnostics.”
To address this need, the researchers developed the deep learning model HECTOR, which stands for ‘histopathology-based endometrial cancer tailored outcome risk.’ HECTOR functions by learning from tumor images and then estimating distant tumor recurrence.
The researchers trained HECTOR using data from patients with stages 1 to 3 EC. The data inputs included hematoxylin and eosin-stained, whole-slide images of patients’ hysterectomy specimens, clinicopathological datasets, molecular data, and clinical distant recurrence data.
“Altogether, including the two training steps and the downstream analyses, the present study comprised tumor data from 2,751 patients,” Dr. Volinsky-Fremond and colleagues noted.
HECTOR yielded respective C-indices of 0.789, 0.828, and 0.815 in one internal (n=353) and two external (n=160 and n=151) test sets. In addition, HECTOR successfully stratified patients by recurrence outcomes. Patients in the HECTOR low-, intermediate-, and high-risk groups had 97%, 77.7%, and 58.1% respective probabilities for 10-year distant recurrence-free survival.
“Notably, HECTOR outperformed the current diagnostic gold standard of combined pathological and molecular analysis for distant recurrence risk prediction and was also found to be predictive of adjuvant chemotherapy benefit in the PORTEC-3 randomized trial,” Dr. Volinsky-Fremond and colleagues explained.
The authors concluded that HECTOR enhanced prognostication but must be validated in prospective trials with more diverse cohorts.
“Pending prospective validation, our results suggest that HECTOR may have the potential to be a highly effective tool for individualized prognostication of women with EC while delivering shorter turnaround times and reducing testing costs. HECTOR may also enable biomarker discoveries for improving targeted treatment decision-making,” the authors wrote.
