Deep-feature MRI model distinguishes stage I endometrial cancer from atypical hyperplasia

A DenseNet121 deep-feature model using multiparametric MRI achieved AUC 0.854 (95% CI 0.743–0.965) for distinguishing stage I endometrial carcinoma from atypical hyperplasia in an external cohort, but differences versus radiomics and clinical models were not significant.

• In a two-center retrospective study of 297 patients, the DenseNet121-derived deep-feature model achieved an AUC of 0.854 in external validation for differentiating stage I endometrial carcinoma from atypical hyperplasia, though not statistically superior to radiomics (AUC 0.803) or clinical (AUC 0.773) models.
• SHAP analysis indicated that features derived from ADC maps contributed substantially to the deep-feature model's predictions.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.