A modified model for predicting mortality after transjugular intrahepatic portosystemic shunt: A multicentre study.

BACKGROUND AND AIMS: The transjugular intrahepatic portosystemic shunt has controversial survival benefits; thus, patient screening should be performed preoperatively. In this study, we aimed to develop a model to predict post-transjugular intrahepatic portosystemic shunt mortality to aid clinical decision making. METHODS: A total of 811 patients undergoing transjugular intrahepatic portosystemic shunt from five hospitals were divided into the training and external validation data sets. A modified prediction model of post-transjugular intrahepatic portosystemic shunt mortality (ModelMT ) was built after performing logistic regression. To verify the improved performance of ModelMT , we compared it with seven previous models, both in discrimination and calibration. Furthermore, patients were stratified into low-, medium-, high- and extremely high-risk subgroups. RESULTS: ModelMT demonstrated a satisfying predictive efficiency in both discrimination and calibration, with an area under the curve of .875 in the training set and .852 in the validation set. Compared to previous models (ALBI, BILI-PLT, MELD-Na, MOTS, FIPS, MELD, CLIF-C AD), ModelMT showed superior performance in discrimination by statistical difference in the Delong test, net reclassification improvement and integrated discrimination improvement (all p < .050). Similar results were observed in calibration. Low-, medium-, high- and extremely high-risk groups were defined by scores of ≤160, 160-180, 180-200 and >200, respectively. To facilitate future clinical application, we also built an applet for ModelMT . CONCLUSIONS: We successfully developed a predictive model with improved performance to assist in decision making for transjugular intrahepatic portosystemic shunt according to survival benefits.

3D automatic liver and spleen assessment in predicting overt hepatic encephalopathy before TIPS: a multi-center study.

BACKGROUND: Overt hepatic encephalopathy (HE) should be predicted preoperatively to identify suitable candidates for transjugular intrahepatic portosystemic shunt (TIPS) instead of first-line treatment. This study aimed to construct a 3D assessment-based model to predict post-TIPS overt HE. METHODS: In this multi-center cohort study, 487 patients who underwent TIPS were subdivided into a training dataset (390 cases from three hospitals) and an external validation dataset (97 cases from another two hospitals). Candidate factors included clinical, vascular, and 2D and 3D data. Combining the least absolute shrinkage and operator method, support vector machine, and probability calibration by isotonic regression, we constructed four predictive models: clinical, 2D, 3D, and combined models. Their discrimination and calibration were compared to identify the optimal model, with subgroup analysis performed. RESULTS: The 3D model showed better discrimination than did the 2D model (training: 0.719 vs. 0.691; validation: 0.730 vs. 0.622). The model combining clinical and 3D factors outperformed the clinical and 3D models (training: 0.802 vs. 0.735 vs. 0.719; validation: 0.816 vs. 0.723 vs. 0.730; all p < 0.050). Moreover, the combined model had the best calibration. The performance of the best model was not affected by the total bilirubin level, Child-Pugh score, ammonia level, or the indication for TIPS. CONCLUSION: 3D assessment of the liver and the spleen provided additional information to predict overt HE, improving the chance of TIPS for suitable patients. 3D assessment could also be used in similar studies related to cirrhosis.

Semi-supervised hybrid spine network for segmentation of spine MR images.

Automatic segmentation of vertebral bodies (VBs) and intervertebral discs (IVDs) in 3D magnetic resonance (MR) images is vital in diagnosing and treating spinal diseases. However, segmenting the VBs and IVDs simultaneously is not trivial. Moreover, problems exist, including blurry segmentation caused by anisotropy resolution, high computational cost, inter-class similarity and intra-class variability, and data imbalances. We proposed a two-stage algorithm, named semi-supervised hybrid spine network (SSHSNet), to address these problems by achieving accurate simultaneous VB and IVD segmentation. In the first stage, we constructed a 2D semi-supervised DeepLabv3+ by using cross pseudo supervision to obtain intra-slice features and coarse segmentation. In the second stage, a 3D full-resolution patch-based DeepLabv3+ was built. This model can be used to extract inter-slice information and combine the coarse segmentation and intra-slice features provided from the first stage. Moreover, a cross tri-attention module was applied to compensate for the loss of inter-slice and intra-slice information separately generated from 2D and 3D networks, thereby improving feature representation ability and achieving satisfactory segmentation results. The proposed SSHSNet was validated on a publicly available spine MR image dataset, and remarkable segmentation performance was achieved. Moreover, results show that the proposed method has great potential in dealing with the data imbalance problem. Based on previous reports, few studies have incorporated a semi-supervised learning strategy with a cross attention mechanism for spine segmentation. Therefore, the proposed method may provide a useful tool for spine segmentation and aid clinically in spinal disease diagnoses and treatments. Codes are publicly available at: https://github.com/Meiyan88/SSHSNet.