🤖 AI Summary
This work addresses the challenge of segmenting small lesions in brain MRI, which is hindered by their minuscule scale and extreme class imbalance, often resulting in low recall and high false-negative rates with conventional methods. To overcome this, the authors propose a unified objective function, CATMIL, integrated within the nnU-Net framework, which jointly leverages voxel-level, connected-component-level, and lesion-level supervision. The approach innovatively combines adaptive weighting of connected components with multiple instance learning (MIL), employing a Component-Adaptive Tversky loss to balance the influence of lesions across varying sizes and incorporating lesion-level detection constraints to co-optimize segmentation and detection performance. Evaluated on the MSLesSeg dataset, the method achieves a Dice score of 0.7834, significantly improves recall for small lesions, reduces false negatives, and yields the lowest false-positive volume among existing approaches.
📝 Abstract
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}.