Introduction: The classification of lesion types in Digital Breast Tomosynthesis (DBT) images is crucial for the early diagnosis of breast cancer. However, the task remains challenging due to the complexity of breast tissue and the subtle nature of lesions. To alleviate radiologists’ workload, computer-aided diagnosis (CAD) systems have been developed. The breast lesion regions vary in size and complexity, which leads to performance degradation. Methods: To tackle this problem, we propose a novel DBT Dual-Net architecture comprising two complementary neural network branches that extract both low-level and high-level features. By fusing different-level feature representations, the model can better capture subtle structure. Furthermore, we introduced a pseudo-color enhancement procedure to improve the visibility of lesions on DBT. Moreover, most existing DBT classification studies rely on two-dimensional (2D) slice-level analysis, neglecting the rich three-dimensional (3D) spatial context within DBT volumes. To address this limitation, we used majority voting for image-level classification from predictions across slices. Results: We evaluated our method on a public DBT dataset and compared its performance with several existing classification approaches. The results showed that our method outperforms baseline models. Discussion: The use of pseudo-color enhancement, extracting high and low-level features and inter-slice majority voting proposed method is effective for lesion classification in DBT. The code is available at https://github.com/xiaoerlaigeid/DBT-Dual-Net.

Medical image segmentation is vital for clinical diagnosis and treatment; however, current solutions face three major limitations: (1) the lack of a universal framework capable of handling diverse modalities and anatomical targets, (2) the limited scalability to adapt to evolving clinical needs and new datasets, and (3) the lack of instructive interfaces that make models usable for non-expert users. To address these challenges, this paper presents MedSegAgent, a universal and scalable multi-agent system for instructive medical image segmentation. Specifically, MedSegAgent comprises five agents: one query parsing agent that processes natural language requests, three coarse-to-fine filtering agents (modality filtering, anatomical filtering, and label selection) for identifying relevant datasets and label values, and one execution agent responsible for model inference and result integration. Based on this framework, MedSegAgent utilizes 23 diverse datasets and pre-trained models to perform 343 types of segmentation across various modalities and anatomical targets. Experimental results demonstrate that MedSegAgent simplifies model selection while maintaining high performance, accurately identifying matching datasets and labels in 94.27% of queries and locating at least one suitable match in 99.03% of queries. MedSegAgent offers a universal and scalable solution for diverse medical image segmentation tasks, bridging the gap between user-friendly queries and the complexities of model selection and deployment.

Ultrasound imaging is vital for the early detection of breast cancer, where accurate lesion segmentation supports clinical diagnosis and treatment planning. However, existing deep learning-based methods rely on pixel-level annotations, which are costly and labor-intensive to obtain. This study presents a weakly supervised framework for breast lesion segmentation in ultrasound images. The framework combines morphological enhancement with Class Activation Map (CAM)-guided lesion localization and utilizes the Segment Anything Model (SAM) for refined segmentation without pixel-level labels. By adopting a lightweight region synthesis strategy and relying solely on SAM inference, the proposed approach substantially reduces model complexity and computational cost while maintaining high segmentation accuracy. Experimental results on the BUSI dataset show that our method achieves a Dice coefficient of 0.7063 under five-fold cross-validation and outperforms several fully supervised models in Hausdorff distance metrics. These results demonstrate that the proposed framework effectively balances segmentation accuracy, computational efficiency, and annotation cost, offering a practical and low-complexity solution for breast ultrasound analysis. The code for this study is available at: https://github.com/YueXin18/MorSeg-CAM-SAM-Segmentation.

Automatic pulmonary airway segmentation from thoracic computed tomography (CT) is an essential step for the diagnosis and interventional surgical treatment of pulmonary disease. While deep learning algorithms have shown promising results in segmenting the main and larger bronchi, segmentation of the distal small bronchi remains challenging due to their limited size and divergent spatial distribution. The study aims to address the challenges associated with segmenting the pulmonary airway, particularly focusing on the distal small bronchi. Specifically, we aim to improve the accuracy and completeness of airway segmentation by developing a novel deep-learning model. To achieve this purpose, we propose an Uncertainty-based Double Attention Detail Supplement Network (UDADS-Net) to identify and supply these missing details of the airway. We introduce the Uncertainty-based Double Attention Module (UDA), which utilizes the uncertainty-based attention module to obtain the regions with high uncertainty and utilizes another attention module to identify the missing details. Moreover, we also propose the Adaptive Multi-scale Module (AMS) to optimize the process of extracting details. Evaluation of our method on the ATM’2022 airway segmentation dataset demonstrates its effectiveness, especially for segmenting distal small bronchi. Our method significantly reduces missing and fragmented parts, leading to more accurate and complete airway segmentation, and achieving higher evaluation metrics compared to the state-of-the-art (SOTA) methods.

High-quality synthetic medical images can enlarge training datasets in different deep learning-based applications. Recently, diffusion-based methods for image synthesis have outperformed GAN-based methods, even for medical images. Unfortunately, using diffusion models is costly in terms of training time and computational resources. We propose a two-stage method that combines diffusion models and GANs to tackle this problem. First, we use diffusion models or GANs to generate low-resolution images. Then, we use a GAN-based super-resolution model to interpolate high-resolution images from these low-resolution images. Experimental results on synthetic breast CT slices show that the proposed framework is more efficient and performs better than state-of-the-art methods that generate the images in a single step. The proposed methods will be available at https://github.com/xiaoerlaigeid/Image-Frequency-Score.git.

Deformable image registration is crucial for medical image analysis, yet the complexity of deep learning networks often limits their deployment on resource-limited devices. Current distillation methods in registration tasks fail to effectively transfer complex deformation handling capabilities to non-iterative lightweight networks, leading to insignificant performance improvement. To address this, we propose the Framework for Iterative to Non-iterative Distillation (FIND), which efficiently transfers these capabilities to a Non-Iterative Lightweight (NIL) network. FIND employs a dual-step process: first, using recurrent distillation to derive a high-performance non-iterative teacher assistant from an iterative network; second, using advanced feature distillation from the assistant to the lightweight network. This enables NIL to perform rapid, effective registration on resource-limited devices. Experiments across four datasets show that NIL can achieve up to 60 times faster performance on CPU and 89 times on GPU than compared deep learning methods, with superior registration accuracy improvements of up to 3.5 points in Dice scores.

The force density method (FDM) is a classical method for form finding and static analysis of membrane-cable-strut-beam hybrid structures (HS). This study focuses on the energy principle of FDM, and establishes the potential energies for HS in the form-finding and static analysis stages. The HS model is discretized into link, T and beam elements. Equilibrium equations and stiffness matrices are formulated using the principle of stationary potential energy. In the integrated analysis, the compatibility between the link and beam element is resolved by adopting (x, y, z, θx, θy, θz) as degrees of freedom (DOF) and transforming the beam's DOF into coordinate differences. The global matrix is assembled using nodal global DOF numbers, showing a computational advantage over the conventional topological matrix method. FDM has demonstrated consistency in form-finding and static analysis stages by applying constant and elastic force densities. The algorithm has been implemented in a program called TMCAD with all the calculation details illustrated, and FDM's effectiveness is verified through five HS examples.

Are larger models always better for 3D medical image segmentation? Despite the widespread adoption of 3D U-Net in various medical imaging tasks, this critical question remains underexplored. To challenge the common assumption, we systematically benchmark 18 U-Net variants—adjusting resolution stages, depth, and width—across 42 diverse public datasets. Our findings reveal that the answer is no: optimal architectures are highly task-specific, with smaller models often performing competitively. Specifically, we identify three key insights: (1) increasing resolution stages provides limited benefits for datasets with larger voxel spacing; (2) deeper networks offer limited advantages for anatomically complex shapes; and (3) wider networks provide minimal advantages for tasks with limited segmentation classes. Based on these insights, we provide practical guidelines for optimizing U-Net architectures according to dataset characteristics. Our findings highlight the limitations of the“bigger is better”paradigm while establishing a framework for balancing performance and computational efficiency in 3D medical image segmentation tasks.

In this study, we propose a deep learning based two-stage breast CT reconstruction in the image domain. Unlike most methods, we use two separate models to improve the Breast CT image quality. In the first stage, a deep learning-based denoiser was used to remove the noise. In the second stage, a deep learning based image enhancement model is used to improve the image quality. We evaluated the proposed method on the AAPM 2021 sparse view CT reconstruction challenge dataset.1 The experimental results demonstrate that the proposed method performs better than all comparison methods.

The 3D breast ultrasound is a radiation-free and effective imaging technology for breast tumor diagnosis. However, checking the 3D breast ultrasound is time-consuming compared to mammograms. To reduce the workload of radiologists, we proposed a 2.5D deep learning-based breast ultrasound tumor classification system. First, we used the pre-trained STU-Net to finetune and segment the tumor in 3D. Then, we fine-tuned the DenseNet-121 for classification using the 10 slices with the biggest tumoral area and their adjacent slices. The Tumor Detection, Segmentation, and Classification on Automated 3D Breast Ultrasound (TDSC-ABUS) MICCAI Challenge 2023 dataset was used to train and validate the performance of the proposed method. Compared to a 3D convolutional neural network model and radiomics, our proposed method has better performance.