dice coefficient

This work addresses the performance bottlenecks in medical image segmentation caused by data imbalance and the difficulty of detecting small lesions. To this end, the authors propose a novel polynomial Dice loss function based on Taylor expansion. This approach reformulates the conventional Dice loss into a polynomial form with adjustable higher-order terms, enabling flexible control over each term’s contribution to the overall loss and thereby refining the geometric properties and optimization dynamics of the loss function. Experimental results across multiple medical image segmentation tasks demonstrate that the proposed method significantly outperforms standard Dice loss and its Tversky variant, achieving superior segmentation accuracy and robustness.

Interactive medical image segmentation lacks a unified, clinically credible evaluation standard, leading to distorted algorithm comparisons and inaccurate performance assessment. This paper proposes a clinical-need-driven standardized evaluation framework that defines reproducible task paradigms and metrics. It systematically identifies— for the first time—the critical roles of information preservation, adaptive scaling, and training-validation prompt consistency in model robustness. The framework enables cross-domain comparative evaluation of both 2D and 3D models on multimodal data, slab-like structures, and irregular targets, while explicitly modeling user interaction behavior. Experiments demonstrate that 3D contextual modeling significantly improves segmentation accuracy for large-scale and irregular lesions; conversely, non-medical pre-trained models exhibit sharp performance degradation under low-contrast conditions and complex morphologies. This work establishes the first clinically grounded, fair benchmark for interactive segmentation evaluation.

Small cerebellar lesions are challenging to segment due to their minute volume, leading to suboptimal optimization and frequent false negatives when using conventional voxel-level loss functions (e.g., Dice). To address this, we propose CC-DiceCE, a novel instance-level loss function that integrates CC-Metrics—measuring lesion-level correspondence via connected components—into segmentation loss design for the first time. Implemented and validated within the nnU-Net framework, CC-DiceCE models lesions as connected-component instances, thereby enhancing sensitivity to small pathological structures. Experiments on a multi-center dataset demonstrate that CC-DiceCE significantly improves recall for small lesions (+8.2%–12.7%) over state-of-the-art blob-based losses, while maintaining stable overall Dice scores and incurring only a marginal increase in false positives. This confirms the efficacy and generalizability of instance-level supervision for segmenting tiny anatomical abnormalities.

In biomedical image segmentation validation, metrics such as the Hausdorff distance suffer from implementation inconsistencies across open-source toolkits, compromising benchmark reliability, introducing biomarker bias, and posing clinical deployment risks. To address this, we systematically evaluate 11 widely used toolkits and introduce, for the first time, a reference implementation based on high-fidelity 3D surface meshes. Our framework integrates real-world clinical data and a cross-platform consistency analysis. Statistical analysis reveals significant inter-tool variation in Hausdorff distance computations (p < 0.001), with interpolation strategy, boundary handling, and sampling density identified as primary sources of discrepancy. Based on these findings, we propose a reproducible and verifiable paradigm for distance-based evaluation, accompanied by standardized computational guidelines. This work substantially enhances the reliability, comparability, and clinical translatability of segmentation assessment.

To address imprecise mask representation and weak uncertainty estimation in medical image segmentation, this paper proposes the first image-guided conditional flow matching framework based on signed distance functions (SDFs). The method models segmentation masks as implicit probability distributions, leveraging the continuous geometric prior of SDFs to enhance deformation modeling naturalness and prediction robustness. By directly regressing vector fields from noisy SDFs to target SDFs, it natively generates pixel-wise variance-based uncertainty maps, enabling high-fidelity sampling and statistical analysis. Evaluated on public nuclear and gland segmentation benchmarks, our approach significantly outperforms existing state-of-the-art methods. Comprehensive qualitative and quantitative experiments demonstrate superior segmentation accuracy, robustness to anatomical variability and noise, and effective uncertainty quantification—particularly critical for clinical decision support and model calibration.

This work addresses the impractical computational cost of conventional KernelSHAP in 3D medical image segmentation, stemming from expensive coalition evaluations and sliding-window inference. Focusing on whole-body CT segmentation, the authors propose an efficient KernelSHAP framework that restricts explanations to regions of interest and their receptive fields, incorporates a patch logit caching mechanism, and leverages organ-aware supervoxels—including face-centered cubic (FCC) and hybrid strategies—alongside multiple value functions (True Positive, Dice, and Soft Dice). While preserving nnU-Net’s ensemble strategy, this approach substantially improves explanation efficiency. Experiments demonstrate a 15%–30% reduction in computational overhead for whole-body CT segmentation. Organ-aware units more effectively uncover false-positive causes under normalized metrics, whereas regular supervoxels, despite improving perturbation-based scores, lack anatomical consistency.

This study investigates whether commonly used evaluation metrics in multilevel image thresholding—namely Structural Similarity Index (SSIM) and Peak Signal-to-Noise Ratio (PSNR)—exhibit implicit preferences toward classical objective functions such as Otsu’s method and Kapur’s entropy. By exhaustively enumerating the full threshold space on the BSDS500 dataset, the work systematically analyzes the correlation between objective functions and evaluation metrics. Empirical results reveal, for the first time, that SSIM and PSNR demonstrate strong and consistent positive correlations with Otsu’s criterion, whereas their correlations with Kapur’s entropy are weak and unstable. Notably, Otsu outperforms Kapur in PSNR correlation across all images and achieves superior SSIM correlation in over 91% of cases. These findings challenge the widely held assumption of metric neutrality, exposing a systematic bias in current evaluation practices.

This work addresses the limited robustness of 3D brain glioma segmentation models under cross-institutional, scanner, or protocol-induced image context shifts by proposing a counterfactual context auditing framework. It introduces counterfactual representation learning into 3D medical image segmentation for the first time, generating anatomically plausible perturbed samples that preserve lesion structure while systematically altering imaging context. This enables controlled evaluation of model prediction stability under distributional shifts. The framework explicitly uncovers model reliance on non-lesion contextual cues—overcoming the limitations of conventional Dice and HD95 metrics—and reveals context-sensitive failure modes on BraTS and UTSW datasets that standard evaluations miss. Moreover, it achieves strong performance both within and across domains.

Clinical CT images often suffer from degradations such as noise, low resolution, and contrast variations, which significantly impair segmentation performance. To address this challenge, this work proposes RAMP, a novel framework that systematically integrates clinically motivated multi-degradation augmentation strategies. Specifically, RAMP enhances model robustness by combining anatomically constrained spatial perturbations, CT-specific intensity transformations, and stochastic combinations of multiple degradations, building upon the nnU-Net architecture. Evaluated on both a five-organ dataset and Abdomen1K, RAMP achieves average Dice scores of 0.753 and 0.789, respectively, on degraded images, reducing the robustness gap to only 0.064 and 0.070. These results demonstrate that RAMP substantially mitigates segmentation collapse under severe image degradation.

Existing instance segmentation evaluation metrics—such as AP@50, PQ, SEG, and AJI—are limited in biological imaging due to their reliance on hard IoU thresholds, object-level normalization biases, and greedy matching strategies, which lead to discontinuous scores, low sensitivity, and unreliable ranking. This work proposes Maximum Matching Accuracy (MMA), a threshold-free and continuous metric that establishes a globally optimal one-to-one correspondence between predicted and ground-truth instances via the Hungarian algorithm and quantifies total overlap using pixel-level normalization. Experimental results demonstrate that MMA substantially outperforms existing metrics across synthetic failure cases, progressive perturbations, and model ranking tasks, exhibiting superior stability, sensitivity, and interpretability.