Brain tumor scans get more reliable even when some parts are missing
Missing Modality-Aware Calibration for Trustworthy Brain Tumor Segmentation
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Brain tumor scans often use several types of MRI images to find and measure tumors, but sometimes some MRI types aren’t available due to technical problems or different hospital protocols. The authors found that while current methods can still find tumors well when some images are missing, they aren’t good at telling how confident those findings are. Their method improves the way the system estimates confidence in its tumor maps, by considering which MRI types are missing and how hard it is to make predictions in different brain areas. This makes the tumor segmentation results more trustworthy and better suited for real medical use.
What this means in practice
- •For clinical radiology teams: Improve the trustworthiness of tumor detection from incomplete MRI scans by providing more accurate confidence measures for segmentations.
- •For medical image software developers: Integrate voxel-level confidence calibration methods that adapt to missing MRI types to enhance segmentation outputs in brain tumor imaging tools.$Commercial implications: Enables vendors to market more reliable brain tumor analysis software that supports variable MRI inputs, improving clinical adoption.
Authors
Sol Lee, Hyunji Kim, Sungrae Hong, Donghee Han, Mun Yi
Abstract
Multimodal brain tumor segmentation typically leverages multiple MRI modalities, yet incomplete modality acquisition is common in clinical practice due to protocol heterogeneity and scan failures. Although recent methods maintain segmentation accuracy under missing modality conditions, they frequently overlook prediction reliability, leading to miscalibrated confidence estimates that hinder clinical adoption. Existing calibration techniques are largely modality-agnostic or assume that prediction difficulty decreases monotonically as additional modalities become available. However, in brain tumor segmentation, prediction difficulty depends primarily on which modalities are absent rather than how many, leading to combination-specific and spatially heterogeneous calibration errors. To address this, we propose Missing Modality-Aware Local Temperature Scaling (MMA-LTS), a post-hoc voxel-wise confidence calibration method. It estimates a spatially adaptive temperature field conditioned on a modality-availability learnable token and a voxel-wise difficulty score. Experiments on BraTS 2020 and FeTS 2024 show that MMA-LTS improves calibration while preserving the segmentation accuracy of state-of-the-art models across diverse missing-modality scenarios, thereby enhancing trustworthiness toward clinical deployment.