Accurate and efficient brain tumor segmentation remains a critical challenge in neuroimaging due to the heterogeneous nature of tumor subregions and the high computational cost of volumetric inference. In this paper, we propose EfficientGFormer, a novel architecture that integrates pretrained foundation models with graph-based reasoning and lightweight efficiency mechanisms for robust 3D brain tumor segmentation. Our framework leverages nnFormer as a modality-aware encoder, transforming multi-modal MRI volumes into patch-level embeddings. These features are structured into a dual-edge graph that captures both spatial adjacency and semantic similarity. A pruned, edge-type-aware Graph Attention Network (GAT) enables efficient relational reasoning across tumor subregions, while a distillation module transfers knowledge from a full-capacity teacher to a compact student model for real-time deployment. Experiments on the MSD Task01 and BraTS 2021 datasets demonstrate that EfficientGFormer achieves state-of-the-art accuracy with significantly reduced memory and inference time, outperforming recent transformer-based and graph-based baselines. This work offers a clinically viable solution for fast and accurate volumetric tumor delineation, combining scalability, interpretability, and generalization.

Accurate segmentation of renal arteries from X-ray angiography videos is crucial for evaluating renal sympathetic denervation (RDN) procedures but remains challenging due to dynamic changes in contrast concentration and vessel morphology across frames. The purpose of this study is to propose TCA-Net, a deep learning model that improves segmentation consistency by leveraging local and global contextual information in angiography videos. Our approach utilizes a novel deep learning framework that incorporates two key modules: a local temporal window vessel enhancement module and a global vessel refinement module (GVR). The local module fuses multi-scale temporal-spatial features to improve the semantic representation of vessels in the current frame, while the GVR module integrates decoupled attention strategies (video-level and object-level attention) and gating mechanisms to refine global vessel information and eliminate redundancy. To further improve segmentation consistency, a temporal perception consistency loss function is introduced during training. We evaluated our model using 195 renal artery angiography sequences for development and tested it on an external dataset from 44 patients. The results demonstrate that TCA-Net achieves an F1-score of 0.8678 for segmenting renal arteries, outperforming existing state-of-the-art segmentation methods. We present TCA-Net, a deep learning-based model that significantly improves segmentation consistency for renal artery angiography videos. By effectively leveraging both local and global temporal contextual information, TCA-Net outperforms current methods and provides a reliable tool for assessing RDN procedures.

The lumbar spine and pelvis are critical stress-bearing structures of the human body, and their rapid and accurate segmentation plays a vital role in clinical diagnosis and intervention. However, conventional CT imaging poses significant challenges due to the low contrast of sacral and bilateral hip tissues and the complex and highly similar intervertebral space structures within the lumbar spine. To address these challenges, we propose a general-purpose segmentation network that integrates a cross-scale feature fusion strategy with a linear self-attention mechanism. The proposed network effectively extracts multi-scale features and fuses them along the channel dimension, enabling both structural and boundary information of lumbar and pelvic regions to be captured within the encoder-decoder architecture.Furthermore, we introduce a linear mapping strategy to approximate the traditional attention matrix with a low-rank representation, allowing the linear attention mechanism to significantly reduce computational complexity while maintaining segmentation accuracy for vertebrae and pelvic bones. Comparative and ablation experiments conducted on the CTSpine1K and CTPelvic1K datasets demonstrate that our method achieves improvements of 1.5% in Dice Similarity Coefficient (DSC) and 2.6% in Hausdorff Distance (HD) over state-of-the-art models, validating the effectiveness of our approach in enhancing boundary segmentation quality and segmentation accuracy in homogeneous anatomical regions.

Intracranial aneurysms (IAs) are abnormal dilations of cerebral blood vessels that, if ruptured, can lead to life-threatening consequences. However, their small size and soft contrast in radiological scans often make it difficult to perform accurate and efficient detection and morphological analyses, which are critical in the clinical care of the disorder. Furthermore, the lack of large public datasets with voxel-wise expert annotations pose challenges for developing deep learning algorithms to address the issues. Therefore, we proposed a novel weakly supervised 3D multi-task UNet that integrates vesselness priors to jointly perform aneurysm detection and segmentation in time-of-flight MR angiography (TOF-MRA). Specifically, to robustly guide IA detection and segmentation, we employ the popular Frangi's vesselness filter to derive soft cerebrovascular priors for both network input and an attention block to conduct segmentation from the decoder and detection from an auxiliary branch. We train our model on the Lausanne dataset with coarse ground truth segmentation, and evaluate it on the test set with refined labels from the same database. To further assess our model's generalizability, we also validate it externally on the ADAM dataset. Our results demonstrate the superior performance of the proposed technique over the SOTA techniques for aneurysm segmentation (Dice = 0.614, 95%HD =1.38mm) and detection (false positive rate = 1.47, sensitivity = 92.9%).

Deep learning can automate nerve identification by learning from expert-labeled examples to detect and highlight nerves in ultrasound images. This study aims to evaluate the performance of deep-learning models in identifying nerves for ultrasound-guided nerve blocks. A total of 3594 raw ultrasound images were collected from public sources-an open GitHub repository and publicly available YouTube videos-covering 9 nerve block regions: Transversus Abdominis Plane (TAP), Femoral Nerve, Posterior Rectus Sheath, Median and Ulnar Nerves, Pectoralis Plane, Sciatic Nerve, Infraclavicular Brachial Plexus, Supraclavicular Brachial Plexus, and Interscalene Brachial Plexus. Of these, 10 images per nerve region were kept for testing, with each image labeled by 10 expert anesthesiologists. The remaining 3504 were labeled by a medical anesthesia resident and augmented to create a diverse training dataset of 25,000 images per nerve region. Additionally, 908 negative ultrasound images, which do not contain the targeted nerve structures, were included to improve model robustness. Ten convolutional neural network-based deep-learning architectures were selected to identify nerve structures. Models were trained using a 5-fold cross-validation approach on an Extended Video Graphics Array (EVGA) GeForce RTX 3090 GPU, with batch size, number of epochs, and the Adam optimizer adjusted to enhance the models' effectiveness. Posttraining, models were evaluated on a set of 10 images per nerve region, using the Dice score (range: 0 to 1, where 1 indicates perfect agreement and 0 indicates no overlap) to compare model predictions with expert-labeled images. Further validation was conducted by 10 medical experts who assessed whether they would insert a needle into the model's predictions. Statistical analyses were performed to explore the relationship between Dice scores and expert responses. The R2U-Net model achieved the highest average Dice score (0.7619) across all nerve regions, outperforming other models (0.7123-0.7619). However, statistically significant differences in model performance were observed only for the TAP nerve region (χ² = 26.4, df = 9, P = .002, ε² = 0.267). Expert evaluations indicated high accuracy in the model predictions, particularly for the Popliteal nerve region, where experts agreed to insert a needle based on all 100 model-generated predictions. Logistic modeling suggested that higher Dice overlap might increase the odds of expert acceptance in the Supraclavicular region (odds ratio [OR] = 8.59 × 10⁴, 95% confidence interval [CI], 0.33-2.25 × 10¹⁰; P = .073). The findings demonstrate the potential of deep-learning models, such as R2U-Net, to deliver consistent segmentation results in ultrasound-guided nerve block procedures.

Accurate segmentation of small lesions in Breast Dynamic Contrast-Enhanced MRI (DCE-MRI) is critical for early cancer detection, especially in high-risk patients. While recent deep learning methods have advanced lesion segmentation, they primarily target large lesions and neglect valuable longitudinal and clinical information routinely used by radiologists. In real-world screening, detecting subtle or emerging lesions requires radiologists to compare across timepoints and consider previous radiology assessments, such as the BI-RADS score. We propose LesiOnTime, a novel 3D segmentation approach that mimics clinical diagnostic workflows by jointly leveraging longitudinal imaging and BIRADS scores. The key components are: (1) a Temporal Prior Attention (TPA) block that dynamically integrates information from previous and current scans; and (2) a BI-RADS Consistency Regularization (BCR) loss that enforces latent space alignment for scans with similar radiological assessments, thus embedding domain knowledge into the training process. Evaluated on a curated in-house longitudinal dataset of high-risk patients with DCE-MRI, our approach outperforms state-of-the-art single-timepoint and longitudinal baselines by 5% in terms of Dice. Ablation studies demonstrate that both TPA and BCR contribute complementary performance gains. These results highlight the importance of incorporating temporal and clinical context for reliable early lesion segmentation in real-world breast cancer screening. Our code is publicly available at https://github.com/cirmuw/LesiOnTime

Classification models that provide human-interpretable explanations enhance clinicians' trust and usability in medical image diagnosis. One research focus is the integration and prediction of pathology-related visual attributes used by radiologists alongside the diagnosis, aligning AI decision-making with clinical reasoning. Radiologists use attributes like shape and texture as established diagnostic criteria and mirroring these in AI decision-making both enhances transparency and enables explicit validation of model outputs. However, the adoption of such models is limited by the scarcity of large-scale medical image datasets annotated with these attributes. To address this challenge, we propose synthesizing attribute-annotated data using a generative model. We enhance the Diffusion Model with attribute conditioning and train it using only 20 attribute-labeled lung nodule samples from the LIDC-IDRI dataset. Incorporating its generated images into the training of an explainable model boosts performance, increasing attribute prediction accuracy by 13.4% and target prediction accuracy by 1.8% compared to training with only the small real attribute-annotated dataset. This work highlights the potential of synthetic data to overcome dataset limitations, enhancing the applicability of explainable models in medical image analysis.

HCC recurrence frequently occurs after curative surgery. Histological microvascular invasion (MVI) predicts recurrence but cannot provide preoperative prognostication, whereas clinical prediction scores have variable performances. Recurr-NET, a multimodal multiphasic residual-network random survival forest deep-learning model incorporating preoperative CT and clinical parameters, was developed to predict HCC recurrence. Preoperative triphasic CT scans were retrieved from patients with resected histology-confirmed HCC from 4 centers in Hong Kong (internal cohort). The internal cohort was randomly divided in an 8:2 ratio into training and internal validation. External testing was performed in an independent cohort from Taiwan.Among 1231 patients (age 62.4y, 83.1% male, 86.8% viral hepatitis, and median follow-up 65.1mo), cumulative HCC recurrence rates at years 2 and 5 were 41.8% and 56.4%, respectively. Recurr-NET achieved excellent accuracy in predicting recurrence from years 1 to 5 (internal cohort AUROC 0.770-0.857; external AUROC 0.758-0.798), significantly outperforming MVI (internal AUROC 0.518-0.590; external AUROC 0.557-0.615) and multiple clinical risk scores (ERASL-PRE, ERASL-POST, DFT, and Shim scores) (internal AUROC 0.523-0.587, external AUROC: 0.524-0.620), respectively (all p < 0.001). Recurr-NET was superior to MVI in stratifying recurrence risks at year 2 (internal: 72.5% vs. 50.0% in MVI; external: 65.3% vs. 46.6% in MVI) and year 5 (internal: 86.4% vs. 62.5% in MVI; external: 81.4% vs. 63.8% in MVI) (all p < 0.001). Recurr-NET was also superior to MVI in stratifying liver-related and all-cause mortality (all p < 0.001). The performance of Recurr-NET remained robust in subgroup analyses. Recurr-NET accurately predicted HCC recurrence, outperforming MVI and clinical prediction scores, highlighting its potential in preoperative prognostication.

Clinical decision-making relies heavily on understanding relative positions of anatomical structures and anomalies. Therefore, for Vision-Language Models (VLMs) to be applicable in clinical practice, the ability to accurately determine relative positions on medical images is a fundamental prerequisite. Despite its importance, this capability remains highly underexplored. To address this gap, we evaluate the ability of state-of-the-art VLMs, GPT-4o, Llama3.2, Pixtral, and JanusPro, and find that all models fail at this fundamental task. Inspired by successful approaches in computer vision, we investigate whether visual prompts, such as alphanumeric or colored markers placed on anatomical structures, can enhance performance. While these markers provide moderate improvements, results remain significantly lower on medical images compared to observations made on natural images. Our evaluations suggest that, in medical imaging, VLMs rely more on prior anatomical knowledge than on actual image content for answering relative position questions, often leading to incorrect conclusions. To facilitate further research in this area, we introduce the MIRP , Medical Imaging Relative Positioning, benchmark dataset, designed to systematically evaluate the capability to identify relative positions in medical images.