paper
Confidence is Not Reliability: Rethinking MC Dropout in Brain Tumour Segmentation
paperactiveprovisional
confidence-is-not-reliability-rethinking-mc-dropout-in-brain-tumour-segmentation-7ff5ba89·1 events·first seen 2d agoAliases: Confidence is Not Reliability: Rethinking MC Dropout in Brain Tumour Segmentation
Co-occurring entities
More like this (12)
Monte Carlo DropoutCompositional Reasoning Depth Predicts Clinical AI FailureUnstable Features, Reproducible Subspaces: Understanding Seed Dependence in Sparse AutoencodersDynamic-Probabilistic Consistency GapNeural Surrogate ModelsQuantifying Faithful Confidence Expression in Large Reasoning Models3D Medical Image ClassificationTowards a Science of AI Agent ReliabilityAnatomy of Post-Training: Using Interpretability to Characterize Data and Shape the Learning SignalUncertainty-aware Multi-Granularity RAGSparse Subspace-to-Expert Sharing for Task-Agnostic Continual LearningBeyond Fully Random Masking: Attention-Guided Denoising and Optimization for Diffusion Language Models
Recent events (1)
MC Dropout uncertainty estimation masks sub-region calibration failures in brain tumour segmentation
A preprint from arXiv evaluates Monte Carlo Dropout for voxel-level uncertainty estimation in glioma segmentation on 126 BraTS21 patients, comparing a pretrained SegResNet and a locally trained UNet-Res. While global uncertainty-error alignment is strong (AUROC ~0.97), the study finds that UNet-Res exhibits near-zero entropy and an ECE of 0.915 on the enhancing tumour sub-region despite a Dice of only 0.714, a severe miscalibration invisible to standard Dice and AUROC metrics. The paper argues that sub-region-specific calibration assessment is necessary for clinical safety and cannot be replaced by aggregate metrics alone.