Wenzhen Zhang

Excellent performance has been achieved on medical image segmentation. Still, existing algorithms perform relatively poorly for annular objects with high intra-class variability and inter-class similarity, which easily leads to regional confusion, especially in medical images with annular regions. In this paper, we first propose an annular prior prompt learning method based on MedSAM, coined APPL, which not only can strengthen the learning for annular region features, but also effectively handles the regional confusion caused by high intra-class variability and inter-class similarity. Specifically, a novel annular prior prompt encoder (APPE) is proposed based on our designed annular constraint to alleviate the negative impact of inter-class similarity, which maps multiple prompt points into a linear regression feature space to provide the standardized annular prompt feature for the mask decoder. In addition, a new region connectivity enhanced image encoder (RCEIE) is developed to introduce morphology attention into supervised training, which reduces feature noise caused by intra-class variability, significantly improving the connectivity of region features. Powered by the collaboration across different annular features, the mask decoder can effectively avoid inter-class confusion to reduce the impact of weak boundaries, further boosting the performance for tackling highly diverse pixel distributions in medical images. Compared with most state-of-the-art methods, the proposed method achieves 90.07% of the mIoU and 94.71% of the DSC on an in-house dataset, and shows strong generalization on two public heart segmentation datasets with end-diastolic DSC scores of 94.2% and 83.9%, and end-systolic DSC scores of 95.5% and 87.3%. The experimental results demonstrate the consistently superior performance of our method quantitatively and qualitatively.

Cardiac magnetic resonance (CMR) is a vital noninvasive imaging modality for evaluating cardiac morphology and function. However, current scanning systems require a long acquisition time, which often introduces motion artifacts and limits the spatiotemporal resolution needed for real-time imaging. Reconstructing images from highly undersampled data offers a promising solution to accelerate clinical examinations. To this end, this article proposes a novel dynamic reconstruction approach—spatial frequency adaptive modulation guided spatialtemporal learning (FASTC)—for accelerating CMR reconstruction. By leveraging the dynamic nature of the CMR acquisition process, we introduce a spatiotemporal adaptive modulation strategy that jointly learns from both temporal and spatial dimensions. Specifically, inspired by linear self-attention (LSA), we introduce an efficient spatiotemporal feature learning (STFL) module to fully exploit the complementary information hidden in consecutive scanning sequences. In addition, the designed frequency adaptive modulation (FAM) mechanism selectively amplifies or suppresses features across multiple frequency subbands to optimize feature information flow. Extensive experiments across diverse accelerated scenarios—such as varying undersampling patterns and acceleration rates—demonstrate that the proposed method outperforms state-of-the-art approaches and effectively reconstructs high-fidelity images.

Integrating the advantages of coronary angiography (CAG) and intravascular ultrasound (IVUS) images can significantly enhance the effectiveness of diagnosing coronary artery disease. However, current techniques for aligning CAG and IVUS rely solely on the manual determination of IVUS. To address this issue, this article introduces an innovative 3-D intelligent reconstruction pipeline for coronary arteries based on CAG and IVUS. The pipeline introduces three main innovations. First, the introduction of a label-text-guided multitask network model for classification and segmentation. This model facilitates end-to-end detection of blood vessel bifurcation and vascular wall segmentation, enhancing segmentation accuracy and calculating bifurcation width information by determining the number of bifurcated blood vessel frames. Second, the proposal of an improved IVUS 3-D reconstruction algorithm that integrates dual-boundary extraction, dual-line segment patch area extraction, and the moving filter-based 3-D smoothing algorithm into the reconstruction processing flow. This method not only presents branch information in the 3-D reconstruction of blood vessels, but also highlights plaque areas, improving the visual impact of the reconstruction. Third, the introduction of a registration algorithm for key branches of blood vessels based on CAG and IVUS. This algorithm aligns CAG and IVUS using key branch information extracted from CAG, achieving intelligent alignment and reconstruction without manual intervention. Experimental results demonstrate that this algorithm effectively aligns CAG and IVUS, assisting doctors in intuitively assessing and locating vascular plaque information, thereby making a significant contribution to the auxiliary diagnosis of coronary heart disease.

Intravenous ultrasound (IVUS) produces important imaging information for the clinical medical detection of intravascular lesions. However, most information hidden in the IVUS image must be manually annotated by doctors and obtained by professional parameter calculation tools, which is time-consuming and labor-intensive. The existing IVUS segmentation algorithms generally ignore the positive effect of expert prior knowledge on the segmentation results. We thus develop a full-scale IVUS Analysis Toolbox (IAT) that can automatically extract more image information through the Image Segmentation Model of Intravascular Ultrasound (ISM-IVUS), allowing experts to modify suboptimal segmentation results and support the calculation and acquisition of instantaneous parameters. The ISM-IVUS integrates the expert prior knowledge extraction module. By fusing edge operators based on intensity changes to obtain prior matrices and constraining the morphological features of the segmentation results, the failure of complex IVUS image segmentation can be effectively avoided. The model uses the U-Net framework and Attention mechanism, combines deep and shallow features, and reduces the loss of important features in the segmentation process. To ensure the generalization and clinical applicability of the toolbox, we used a large-scale professional dataset provided by the hospital and a public dataset to conduct comparison experiments. The mIoU, Dice index, and Hausdorff distance of the segmentation results in IAT reached 0.8616, 0.9343, and 6.0799 respectively, which verified the effectiveness of the model. In the comparison experiments of number of parameters, FLOPs, and robustness, ISM-IVUS also has certain advantages. The correlation and p-value of the calculated parameters also achieved good results. A series of experiments have shown that IAT is in line with clinical needs and provides an application basis for the clinical practice of subsequent IVUS research.