Automated AI Tool Enables Uncertainty-Aware Meningioma Volume Tracking on MRI

These methods can miss true tumor burden in slowly growing or irregular lesions such as meningiomas, which arise from the meninges and frequently challenge boundary estimation. Uncertainty in automated three-dimensional segmentation further limits clinical adoption of artificial intelligence. To help address this challenge, researchers have developed an uncertainty-aware deep learning framework for MRI-based tumor volumetrics.

The Evidential Deep Learning (EDL) framework from the University of California, San Francisco (UCSF) generates calibrated volumetric measurements together with interpretable voxel-level uncertainty maps for meningioma segmentation on clinical brain MRI. The approach uses ensembles of EDL models to quantify case- and region-specific uncertainty rather than assuming a single, definitive output. By coupling segmentation with uncertainty estimation, the system is designed to support safer deployment in clinical workflows that depend on precise volume tracking.

The framework was trained on 1,655 MRIs from 788 patients, including postoperative studies in which treatment-related changes can resemble tumor tissue and elevate uncertainty. Performance was assessed on an independent test set of 68 MRIs from 43 patients by measuring spatial agreement between the model’s uncertainty maps and neuroradiologist-identified ambiguous regions. The model achieved high accuracy, with uncertainty maps aligning to ambiguous areas and volume estimates that were well calibrated. External validation in 353 patients supported cross-institutional generalizability.

According to the investigators, these calibrated uncertainty outputs may be applicable to lesion segmentation beyond meningiomas and could increase clinician confidence in AI-generated brain tumor volumetrics. The work advances transparent, trustworthy AI by enabling uncertainty-aware, quantitative monitoring of tumor dynamics in routine care. The study, published in npj Digital Medicine, also notes that future evaluations should include multicenter datasets and multiple-rater annotations to benchmark model uncertainty against human interobserver variability.

"The purpose of this study was to develop and validate a fully automated framework for tumor segmentation on real clinical brain MRIs, built on ensembles of EDL models, capable of producing clinically meaningful volumetric measurements and interpretable uncertainty maps," said Andreas Rauschecker, M.D., Ph.D., UCSF assistant professor of radiology and co-chief of Intelligent Imaging Research.

"Even with a good segmentation algorithm, there is always going to be some level of uncertainty about the volume of the tumor because AI can make mistakes in segmentation. Our objective was to capture a mathematical, quantifiable way of assessing the amount of uncertainty," said Rauschecker.

Related Links
UCSF