Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
The paper introduces CATMIL, a novel unified objective function for brain MRI segmentation that combines voxel-level segmentation with component-adaptive and lesion-level supervision. CATMIL reweights voxel contributions based on connected components and uses multiple instance learning to encourage lesion detection. Evaluated on the MSLesSeg dataset, CATMIL improves Dice score, reduces boundary error, increases small lesion recall, and minimizes false positives compared to standard losses within an nnU-Net framework.
Stop missing small but critical lesions in brain MRIs: CATMIL's component-adaptive and lesion-level supervision boosts small lesion recall while minimizing false positives.
We propose a unified objective function, termed CATMIL, that augments the base segmentation loss with two auxiliary supervision terms operating at different levels. The first term, Component-Adaptive Tversky, reweights voxel contributions based on connected components to balance the influence of lesions of different sizes. The second term, based on Multiple Instance Learning, introduces lesion-level supervision by encouraging the detection of each lesion instance. These terms are combined with the standard nnU-Net loss to jointly optimize voxel-level segmentation accuracy and lesion-level detection. We evaluate the proposed objective on the MSLesSeg dataset using a consistent nnU-Net framework and 5-fold cross-validation. The results show that CATMIL achieves the most balanced performance across segmentation accuracy, lesion detection, and error control. It improves Dice score (0.7834) and reduces boundary error compared to standard losses. More importantly, it substantially increases small lesion recall and reduces false negatives, while maintaining the lowest false positive volume among compared methods. These findings demonstrate that integrating component-level and lesion-level supervision within a unified objective provides an effective and practical approach for improving small lesion segmentation in highly imbalanced settings. All code and pretrained models are available at \href{https://github.com/luumsk/SmallLesionMRI}{this url}.