MRI Habitat Radiomics and AI Predict p53 Status in Endometrial Cancer

A retrospective, multicenter study has developed an imaging model that combines multiparametric MRI-based habitat radiomics with a Vision Transformer deep learning architecture to predict the p53 immunophenotype of endometrial cancer before surgery. The work, published open access in Cancer Imaging on 1 October 2026, reports that a fusion model integrating habitat radiomics features, deep learning features and clinical predictors achieved areas under the curve of 0.881, 0.869 and 0.904 in internal and two external validation cohorts.

The habitat model attained areas under the curve of 0.770 to 0.802 in the validation cohorts, outperforming the conventional radiomics model. The fusion model, which additionally incorporated clinical predictors, delivered the best performance across all validation sets.

Among clinical variables, maximum tumor diameter and histologic grade were selected as predictors of p53-abnormal status. According to the reported results, this analysis indicated that habitat and deep learning features made significant contributions to identifying p53-abnormal endometrial cancer.

Beyond diagnosis, the authors explored prognostic relevance using progression-free survival. Their exploratory survival analysis showed that the p53 status predicted by the fusion model was independently associated with progression-free survival. The authors characterize this as exploratory, and the finding would require confirmation in dedicated prognostic studies before informing clinical risk stratification.

Several limitations deserve emphasis. The article was also shared early as a peer-reviewed, accepted version that is subject to further edits before the final Version of Record. As with any predictive model, reported areas under the curve reflect performance on the studied cohorts and do not guarantee equivalent accuracy in other populations or scanner settings.

Nevertheless, the study adds to a growing body of work applying habitat-based imaging analysis and transformer architectures to molecular profiling of tumors, including recent studies targeting p53-abnormal endometrial cancer, prostate cancer risk stratification, and Ki-67 expression in breast cancer. If prospectively validated, such fusion models could offer a noninvasive preoperative estimate of tumor biology, helping clinicians and patients prepare for surgery with more information in hand.

Subject of Research: Preoperative prediction of p53 status in endometrial cancer using multiparametric MRI habitat radiomics and a Vision Transformer fusion model

Article Title: Multiparametric MRI-based habitat radiomics integrated with a vision transformer for preoperative prediction of p53 status in endometrial cancer

Article References: Lin, G., Chen, W., Yang, D., Xu, L., Cao, J., Xia, S., Yang, W., Xu, M., Chen, M., Lu, C., & Ji, J. (2026). Multiparametric MRI-based habitat radiomics integrated with a vision transformer for preoperative prediction of p53 status in endometrial cancer. Cancer Imaging. https://doi.org/10.1186/s40644-026-01133-6

Image Credits: AI Generated

DOI: 10.1186/s40644-026-01133-6

Keywords: endometrial cancer, p53 status, multiparametric MRI, habitat radiomics, Vision Transformer, deep learning, fusion model, preoperative prediction, progression-free survival, SHAP interpretability, multicenter retrospective study, Cancer Imaging