OBJECTIVES:This study aimed to establish a machine-learning model that integrates contrast-enhanced ultrasound (CEUS) radiomics, conventional ultrasound (US) radiomics and clinical features for distinguishing mummified thyroid nodules (MTNs) from papillary thyroid carcinomas (PTCs). The goal is to provide strategies for subsequent clinical decision-making. METHODS:This study included 120 PTCs and 84 MTNs. Clinical information was obtained from the patients to identify independent risk factors. Radiomics features were extracted from the images. After feature selection, Logistic Regression (LR) and support vector machine (SVM) algorithms were used to build various models. Finally, the receiver operating characteristic (ROC) curve, calibration curve, and decision curve analysis (DCA) were used to evaluate the diagnostic performance and clinical utility of the models. A nomogram was used to visualize the final model. RESULTS:The developed US+CEUS radiomics model demonstrated the best diagnostic performance. With seven CEUS and three US features, it achieved an area under the curve (AUC) of 0.936 in the training set and 0.881 in the test set. The DCA indicated that the US+CEUS radiomics model achieved a higher net benefit within the threshold probability range of 0.07-0.97. CONCLUSIONS:The US+CEUS radiomics model shows high diagnostic value in differentiating PTCs from MTNs with the model dominated by CEUS features. However, combining it with clinical features did not improve the diagnostic performance of the model, so we should combine modalities appropriately according to the specific disease.

Automatic tumor segmentation is a critical task in medical image analysis. Positron emission tomography (PET) and computed tomography (CT) are widely used in early cancer diagnosis because they provide complementary imaging information about anatomical structures. However, obtaining complete images from both modalities in clinical practice is often challenging due to constraints such as cost and physical limitations. Missing modality data can hinder multi-modal understanding and degrade the performance of automatic tumor segmentation models. Existing methods struggle to effectively exploit cross-modal correlations and capture essential semantic information. To address these challenges, we propose DMM-Net, an end-to-end Diffusion Mamba Multi-modal Network for incomplete multi-modal automatic tumor segmentation. DMM-Net optimizes performance by combining generation and segmentation in a synergistic manner. Specifically, DMM-Net consists of two key components:conditional missing modality generation and cross-modal spatial-channel interaction segmentation. In the first component, a score-based diffusion model generates the missing modality, leveraging the available modalities as conditional guidance to reduce semantic ambiguities. In the second component, we propose a cross-modal interaction Mamba and a cross-modal channel attention enhancement module. The interaction Mamba efficiently captures global contextual features from PET and CT images, facilitating cross-modal feature interaction. The channel enhancement module highlights essential channel features within the multi-modal representations, captures inter-modal feature correlations, and suppresses redundant information. Next, we validate the effectiveness of DMM-Net through comprehensive ablation studies and visualization experiments on three multi-modal datasets (STS, LMNCLC, and Hecktor 2022). Quantitative comparisons and segmentation results demonstrate that DMM-Net surpasses state-of-the-art methods, regardless of modality incompleteness.

Segmenting muscle compartments and measuring their architecture can facilitate movement function assessment, accurate musculoskeletal modeling, and synergy-based electromyogram simulation. Here, we presented a novel method for volumetric segmentation of muscle compartments using in vivo imaging, focusing on the independent compartments for finger control of flexor digitorum superficialis (FDS). Besides, we measured the architectural properties of FDS compartments and validated the segmentation. Specifically, ultrasound and magnetic resonance imaging (MRI) from 10 healthy subjects were used for segmentation and measurement, while electromyography was utilized for validation. A two-step piecewise segmentation was proposed, first annotating compartment regions in the cross-sectional ultrasound image based on compartment movement, and then performing minimum energy matching to register the ultrasound data to the three-dimensional MRI coordinate system. Additionally, the architectural properties were measured in the compartment masks from the segmentation using MRI tractography. Anatomical correctness was verified by comparing known anatomy with reconstructed fiber tracts and measured properties, while segmentation accuracy was quantified as the percentage of finger electromyogram centers falling within their corresponding compartments. Results demonstrated agreement for the fiber orientation between the tractography and cadaveric photographs. Significant differences in architectural properties (P < 0.001) were observed between compartments. The properties of FDS and its compartments were within the physiological ranges (P < 0.01). 95

BACKGROUND To explore hub genes and pathways related to sarcopenia progression associated with mitochondria and macrophage activation. METHODS GEO datasets GSE8479, GSE1428, and GSE136344 were obtained based on GEO database. MRGs and MARGs were acquired. Thereafter, GO, KEGG, GSEA and PPI analysis were implemented. Later, pooled GEO datasets (Combined Datasets) were further obtained to construct the sarcopenia diagnostic model and sarcopenia subtypes. RESULTS A total of 62 M&MARDEGs were finally identified. There were altogether 9 model genes incorporated into the multifactorial regression model. Immunoinfiltration analysis revealed low scores of M&MARDEGs in the samples with sarcopenia. For macrophage activation in sarcopenia samples, most immune cells were strongly correlated in LowScore group (r = -0.707, p < 0.05). Many immune cells in Mitochondria & Macrophage Activation (HighScore) group were strongly correlated with each other (r = -0.661, p < 0.05). CONCLUSION Screening key genes for sarcopenia associated with mitochondrial and macrophage activation in this study assists in understanding the pathogenesis of sarcopenia and providing new targets for intervention.

Glaucoma is a chronic ocular disease that often remains undiagnosed until advanced stages, highlighting the need for early detection. Current state-of-the-art methods mainly adapt attention mechanisms into classification networks designed for natural images via transfer learning, but they fail to capture domain-specific features and show weak cross-dataset generalization. In this paper, we propose a prior knowledge-driven dual-path network (PK-Net) that integrates medical knowledge into model architecture for glaucoma screening. First, based on the diagnostic importance of the optic disc, we introduce the Global and Local Fusion Network (GloLocNet), which combines high-resolution local optic disc images with global fundus images and applies a triple-loss strategy to improve feature extraction. Second, to leverage the strong inter-eye correlation of glaucoma, we propose the Binocular Fusion Network (BFNet), where paired eye images are processed through parallel GloLocNet encoders and fused to yield joint screening results. PK-Net was validated on multiple datasets, achieving superior intraand cross-dataset results. For PAPILA, AUC, BAcc, Sen, and Spe were 95.67%, 91.18%, 88.72%, and 93.63%; for OIA-ODIR, 92.43 %, 85.64%, 84.62 %, and 86.67%. Trained on ORIGA and tested on REFUGE, the values were 90.00%, 82.29%, 81.25 %, and 83.33 %, demonstrating strong generalization. On GAMMA, results were 95.16%, 83.37%, 75.49%, and 91.29%. These findings indicate that PK-Net effectively enhances glaucoma screening by embedding prior medical knowledge into network design.

Background and aims: Computer-aided diagnosis and prognosis rely heavily on fully automatic seg-mentation of abdominal fat tissue using Emission Tomography images. The identification of subcu-taneous adipose tissue (SAT) and visceral adipose tissue (VAT) in abdomen fat faces two main challenges: (1) the great difficulties in comparison to multi-stage semantic segmentation (VAT and SAT), and (2) the subtle differences due to the high similarity of the two classes in abdomen fat and complicated VAT distribution. Methods: In this research, we built an automated convolutional neural network (A-CNN) for segmenting Abdominal adipose tissue (AAT) from radiology images. Results: We developed a point-to-point design for the A-CNN learning process, wherein the representing features might be learned together with a hybrid feature extraction technique. We tested the proposed model on a CT dataset and evaluated it to existing CNN models. Furthermore, our suggested approach, A -CNN, outperformed existing deep learning methods regarding segmentation outcomes, notably in the AAT segment. Conclusions: Proposed method is extremely fast with remarkable performance on limited-scale low dose CT-scanning and demonstrates the strength in providing an efficient computer-aimed tool for segmen-tation of AAT in the clinic. (c) 2022 Published by Elsevier Ltd on behalf of Diabetes India.

Background Ureteropelvic junction obstruction (UPJO) caused by crossing vessels is a common upper urinary tract abnormal development in which the vessels compress the upper segment of the ureter leading to different grades of hydronephrosis. Compared with routine computed tomography (CT) examination, three-dimensional visualization technology (3DVT) can help surgeons better understand the anatomical structure of the target surgical area. The aim of this study was to investigate the clinical value of 3DVT for the diagnosis, preoperative surgical planning, perioperative outcomes, and patient education of UPJO caused by crossing vessels. Methods In this study, we retrospectively analyzed the clinical data of 28 patients who were admitted to the Department of Urology in Xiangya Hospital between January 2016 and December 2021 presenting with UPJO caused by crossing vessel compression. Among the 28 patients included, 13 patients underwent preoperative 3DVT and 15 patients underwent routine computed tomography (CT) scans. After the initial evaluation, all patients received standardized dismembered LP. The 2 groups of patients were compared in terms of demographic parameters, intraoperative data, and perioperative results. After surgery, all patients were asked to complete a Likert scale questionnaire to gain insight into their understanding of the disease and surgery, as well as their satisfaction with the use of different imaging techniques. Results There were no statistically significant differences in age, gender, body mass index (BMI), side of obstruction, blood vessel compressing the ureteropelvic junction (UPJ), mean duration of hospitalization, and surgical efficacy between the 2 groups. However, the 3DVT group experienced a significant reduction in operation duration (120.8±7.0 versus 144.0±7.9 min, P=0.039), time required for dismemberment of the UPJ (14.8±1.7 versus 24.0±2.2 min, P=0.004), and the amount of intraoperative blood loss (60.8±10.5 versus 95.3±11.9 mL, P=0.041). The 3DVT group was also notably superior to the conventional CT group in terms of the overall levels of patient satisfaction and understanding of the disease and surgery. Conclusions 3DVT is a helpful preoperative examination tool which can clearly show the anatomical relationship between crossing vessels and the UPJ. In addition, 3DVT can also help patients better understand their conditions and surgical plans, thereby improving patient satisfaction.

BACKGROUND:The current diagnosis of major depressive disorder (MDD) is mainly based on the patient's self-report and clinical symptoms. Machine learning methods are used to identify MDD using resting-state functional magnetic resonance imaging (rs-fMRI) data. However, due to large site differences in multisite rs-fMRI data and the difficulty of sample collection, most of the current machine learning studies use small sample sizes of rs-fMRI datasets to detect the alterations of functional connectivity (FC) or network attribute (NA), which may affect the reliability of the experimental results. METHODS:Multisite rs-fMRI data were used to increase the size of the sample, and then we extracted the functional connectivity (FC) and network attribute (NA) features from 1611 rs-fMRI data (832 patients with MDD (MDDs) and 779 healthy controls (HCs)). ComBat algorithm was used to harmonize the data variances caused by the multisite effect, and multivariate linear regression was used to remove age and sex covariates. Two-sample t-test and wrapper-based feature selection methods (support vector machine recursive feature elimination with cross-validation (SVM-RFECV) and LightGBM's "feature_importances_" function) were used to select important features. The Shapley additive explanations (SHAP) method was used to assign the contribution of features to the best classification effect model. RESULTS:The best result was obtained from the LinearSVM model trained with the 136 important features selected by SVMRFE-CV. In the nested five-fold cross-validation (consisting of an outer and an inner loop of five-fold cross-validation) of 1611 data, the model achieved the accuracy, sensitivity, and specificity of 68.90 %, 71.75 %, and 65.84 %, respectively. The 136 important features were tested in a small dataset and obtained excellent classification results after balancing the ratio between patients with depression and HCs. CONCLUSIONS:The combined use of FC and NA features is effective for classifying MDDs and HCs. The important FC and NA features extracted from the large sample dataset have some generalization performance and may be used as a reference for the altered brain functional connectivity networks in MDD.

Dynamic causal modeling (DCM) is a tool used for effective connectivity (EC) estimation in neuroimage analysis. But it is a model-driven analysis method, and the structure of the EC network needs to be determined in advance based on a large amount of prior knowledge. This characteristic makes it difficult to apply DCM to the exploratory brain network analysis. The exploratory analysis of DCM can be realized from two perspectives: one is to reduce the computational cost of the model; the other is to reduce the model space. From the perspective of model space reduction, a model space exploration strategy is proposed, including two algorithms. One algorithm, named GreedyEC, starts with reducing EC from full model, and the other, named GreedyROI, start with adding EC from one node model. Then the two algorithms were applied to the task state functional magnetic resonance imaging (fMRI) data of visual object recognition and selected the best DCM model from the perspective of model comparison based on Bayesian model compare method. Results show that combining the results of the two algorithms can further improve the effect of DCM exploratory analysis. For convenience in application, the algorithms were encapsulated into MATLAB function based on SPM to help neuroscience researchers to analyze the brain causal information flow network. The strategy provides a model space exploration tool that may obtain the best model from the perspective of model comparison and lower the threshold of DCM analysis.

肝脏分割是计算机辅助肝脏疾病诊断的重要前提和基础.本文提出了一种新的基于水平集和形状描述符的腹部CT序列图像肝脏自动分割方法.首先,对原始腹部CT序列图像进行预处理,去除与肝脏不相关的器官和组织.然后,利用灰度偏移场,结合周长项、距离正则项和相邻切片肝脏分割结果构建水平集能量函数,实现CT序列肝脏自动分割.为避免分割误差累积,提出一种基于形状描述符和瓶颈率的肝脏边缘优化方法,在每张切片分割完毕后去除由于灰度重叠造成的过分割.通过对XHCSU14数据库和Sliver07数据库中腹部CT序列的肝脏分割实验,以及与其他肝脏分割算法的比较,表明了本文方法的有效性,且分割精度高,鲁棒性强.

BACKGROUND AND OBJECTIVE:Liver segmentation from abdominal CT volumes is a primary step for computer-aided surgery and liver disease diagnosis. However, accurate liver segmentation remains a challenging task for intensity inhomogeneity and serious pathologies occurring in liver CT volume. This paper presents a novel framework for accurate liver segmentation from CT images.METHODS:Firstly, a novel level set integrated with intensity bias and position constraint is applied, and for normal liver, the generated liver regions are regarded as the final results. Then, for pathological liver, a sparse shape composition (SSC)-based method is presented to refine liver shapes, followed by an improved graph cut to further optimize segmentation results. The level set-based method is capable of overcoming intensity inhomogeneity in object regions, and the SSC- and graph cut-based strategy has outstanding power to address under-segmentation appearing in pathological livers.RESULTS:The experiments conducted on public databases SLIVER07 and 3Dircadb show that the proposed method can segment both healthy and pathological liver effectively. The segmentation performance in terms of mean ASD, RMSD, MSD, VOE and RVD on SLIVER07 are 0.9mm, 1.8mm, 19.4mm, 5.1% and 0.1%, respectively, and on 3Dircadb are 1.6mm, 3.1mm, 27.2mm, 9.2% and 0.5%, respectively, which outperforms many existing methods.CONCLUSIONS:The proposed method does not require complex training procedure on numerous liver samples, and has satisfying and robust segmentation performance on both normal and pathological liver in various shapes.