MRI- and report-based multimodal model with SHAP-based explanation for preoperative prediction of deep stromal invasion in early-stage cervical cancer
Combining MRI scans with AI text analysis predicts cervical cancer severity with high accuracy, potentially avoiding unnecessary surgery in some patients.
An explainable multimodal model combining T2W-MRI radiomics, NLP-extracted report features (BERT), and clinical variables achieved AUC 0.912 for predicting deep stromal invasion in early cervical cancer, significantly outperforming unimodal approaches. External validation (n=20) was limited, requiring independent multi-center confirmation before clinical use.
What the study was
- Study design
- Retrospective diagnostic study, multi-cohort ML validation
- Population
- Early-stage cervical cancer patients; n not specified in abstract but AUC validated on internal and external cohorts (external n=20)
- Category
- Diagnostics
- Maturity
- Exploratory
- Journal
- Insights Imaging
Why it surfaced
Well-designed multimodal ML model for cervical cancer staging with good internal AUC. Limited by very small external cohort (n=20). SHAP explainability is methodological strength.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.