An Evidence-based Analysis of Mask Selection in Zero-Shot Medical Image Segmentation

MRD-UNet: a compute-efficient medical image segmentation architecture with multi-attention and parallel dilated fusion

MRD-UNet provides a practical balance between segmentation accuracy and computational efficiency and outperforms baseline CNNs and performs comparably to heavier transformer-based models while using significantly fewer parameters.

Distribution-Constrained Semi-Supervised Learning for Medical Organ Segmentation

BoxSegUS: A Spatial Consistency Box Supervised Multi-Class Segmentation with Prior and Boundary Constraint for Ultrasound Images.

Accurate segmentation of anatomical structures in ultrasound images is essential for quantitative assessment, fetal development monitoring, and disease diagnosis. While fully supervised models have achieved strong performance, their reliance on costly pixel-level annotations has driven interest in box-based weakly supervised methods. However, existing medical box-supervised approaches are predominantly designed for single-class segmentation and remain less effective for complex multi-structure ultrasound images. These methods struggle due to inherent characteristics such as spatial variations, unreliable local appearance cues, and fuzzy boundaries. Moreover, the dense spatial distribution of anatomical structures leads to heavily overlapping boxes and class-assignment ambiguity. In this study, we propose BoxSegUS, a box-supervised framework that exploits bounding-box annotations for accurate ultrasound segmentation. Specifically, we enforce weak-strong spatial consistency to improve robustness against spatial variations and employ a detection-prior global context modeling mechanism to reduce the influence of unreliable local appearance cues. To mitigate class-assignment ambiguity caused by overlapping anatomical boxes, we incorporate inter-/intra-class anatomical priors to guide mask evolution with coarse structural knowledge. Finally, a Soft Projection loss and sparse boundary regularization are used to handle imperfect box annotations and refine boundaries. Experiments on four ultrasound datasets, including three multi-structure datasets and one external thyroid nodule dataset, show that BoxSegUS consistently outperforms representative box-supervised baselines and achieves competitive performance compared with fully supervised models.

An Efficient and Automated Brain Tumor Segmentation in MRI Images Using Non-Supervised Model

Brain neoplasm is life-threatening abnormalities that require accurate segmentation for effective diagnosis along with treatment planning. Existing segmentation methods often depend on manual annotations or supervised learning, which are time-consuming, subjective, and not scalable. To overcome these challenges, this study proposes a fully unsupervised brain tumor segmentation framework using multimodal MRI data. The 3D MRI volumes are first sliced into 2D axial views and preprocessed to enhance quality. K-means clustering is then applied to generate pseudo-labels that distinguish tumor from healthy regions. These labels guide M-Net CNN training process, which produces coarse tumor localization. A U-Net model further refines the segmentation by fusing the original sliced MRI inputs with M-Net CNN outputs to enhance boundary precision and anatomical alignment. The proposed pipeline achieved a high Dice coefficient of 84.2%, demonstrating its accuracy and robustness. This multi-stage, annotation-free approach enables precise tumor delineation and is suitable for large-scale clinical integration.

GroundMed-SAM: Prompt-based Zero-shot Medical Image Segmentation

Medical image segmentation is a key component of computer-aided diagnosis and treatment planning. Despite substantial progress in deep learning–based models, most existing approaches depend heavily on large annotated datasets and often fail to generalize across heterogeneous clinical environments, limiting their deployment in real-world settings characterized by domain shifts and scarce expert annotations. This paper presents a zero-shot learning framework named GroundMed-SAM for medical image segmentation. The framework integrates GroundingDINO for prompt-based region localization and MedSAM for mask generation. To address the weak alignment between visual features and medical semantics in GroundingDINO, which is pretrained on general domain image-text pairs, we introduce learnable medical text embeddings that explicitly parameterize domain-specific terminology in a continuous semantic space. These embeddings are optimized during training to better align medical concepts with visual representations, thereby strengthening text-image correspondence and improving detection-guided segmentation. The proposed framework preserves true zero-shot capability, enabling segmentation of previously unseen anatomical structures without task-specific labels. Extensive experiments on multiple public datasets across diverse modalities and clinical contexts demonstrate that our method achieves competitive segmentation performance in-domain while exhibiting superior robustness under cross-domain evaluation. Although supervised baselines outperform the proposed framework by only 3–5% on in-domain datasets, they experience substantial performance degradation when evaluated on unseen domains. Additionally, the framework achieves an AUC of 98.9 in endoscopic polyp detection, highlighting the effectiveness of the proposed medical-aware textual embeddings in guiding region localization. These results demonstrate the effectiveness of the proposed framework in improving cross-domain generalization for medical image segmentation with limited annotations.

Conditional flow matching with coordinate-aware patch training for efficient 3D medical image segmentation

Medical image segmentation plays a crucial role in clinical diagnosis and treatment planning. In recent years, generative methods such as diffusion models have shown strong potential for medical image segmentation because they can model complex data distributions and capture aleatoric uncertainty. However, these methods usually require hundreds of iterative sampling steps during inference, which results in high computational cost and limits their practical use in clinical environments. In addition, when generative segmentation models are applied to three dimensional medical image segmentation tasks, they often face severe GPU memory limitations. To address these issues, we propose a conditional Flow Matching (CFM) based framework for efficient 3D medical image segmentation. By adopting a straight probability path derived from optimal transport, the proposed method can generate high quality segmentation masks with only a single ODE integration step during inference. To alleviate the memory limitation caused by volumetric data and the loss of spatial location information in conventional patch training, we introduce a coordinate-aware patch training strategy, in which normalized 3D Cartesian coordinates are explicitly concatenated with local image patches, enabling the model to retain global anatomical positional awareness under constrained GPU memory. Experiments on the BraTS2020 brain tumor dataset and a private prostate MRI dataset demonstrate that the proposed method maintains competitive segmentation accuracy while achieving an inference time of only 1.14 seconds per volume, offering a clear efficiency advantage over existing generative segmentation methods and providing a viable solution for efficient 3D medical image segmentation.