Advancing Personalized Treatment in Endometrial Cancer With the HECTOR Deep Learning Model
According to a study published in Nature, endometrial cancer (EC) cases are on the rise, and 10% to 20% of patients with EC experience distant recurrence. In this study, in researchers aimed to predict distant recurrence of EC to improve personalized adjuvant treatment by called HECTOR.
HECTOR is a deep learning model that uses only 2 inputs—hematoxylin and eosin-stained whole-slide images as well as tumor stage—to predict distant recurrence. It combines self-supervised learning on tumor images and anatomical staging information to capture morphological and molecular details without extensive molecular profiling. The current gold standard for prediction is costly and combines pathological and molecular profiling.
HECTOR was tested on 2072 patients with EC across 8 cohorts. According to the study, HECTOR outperformed traditional methods of predicting distant recurrence, particularly in identifying high-risk patients who would benefit most from chemotherapy, as demonstrated in the PORTEC-3 trial. It showed C-indices of up to 0.828 and stratified patients into low-, intermediate-, and high-risk groups for 10-year recurrence probabilities of 97%, 77.7%, and 58.1%. The risk scores it provides correlate strongly with known high-risk EC features and can identify markers with therapeutic potential.
The tool is not only promising for clinical use but is also a cost-effective way to personalize adjuvant treatment by refining patient risk stratification. Researchers hope that this will help to improve outcomes while minimizing unnecessary chemotherapy, pending further validation in clinical trials. Prospective trials, such as the ongoing PORTEC-4a study, will be crucial for confirming clinical utility.
“In summary, validation and extension of HECTOR could help delivery of precision medicine to advance prognostication of women with stage I-III EC who underwent primary surgery, with improvement worldwide on both systemic therapy recommendation and treatment de-escalation,” researchers concluded.
Reference
Volinsky-Fremond S, Horeweg N, Andani S, et al. Prediction of recurrence risk in endometrial cancer with multimodal deep learning. Nat Med. 2024;30(7):1962-1973. doi:10.1038/s41591-024-02993-w
