Volumetric segmentation is a major challenge in medical imaging, as current methods require extensive annotations and retraining, limiting transferability across objects. We present PAM, a propagation-based framework that generates 3D segmentations from a minimal 2D prompt. PAM integrates a CNN-based UNet for intra-slice features with Transformer attention for inter-slice propagation, capturing structural and semantic continuity to enable robust cross-object generalization. Across 44 diverse datasets, PAM outperformed MedSAM and SegVol, improving average DSC by 19.3%. It maintained stable performance under variations in prompts (P ≥ 0.5985) and propagation settings (P ≥ 0.6131), while achieving faster inference (P < 0.001) and reducing user interaction time by 63.6%. Gains were strongest for irregular objects, with improvements negatively correlated with object regularity (r < -0.1249). By delivering accurate 3D segmentations from minimal input, PAM lowers reliance on manual annotation and task-specific training, providing an efficient and generalizable tool for automated clinical imaging.

Cytokine release syndrome (CRS) greatly impacts survival in patients who undergo chimeric antigen receptor (CAR)-T cell therapy, and the identification of its determinants is still challenging. We analysed the impact of systemic immunoinflammatory index (SII) and body composition parameters derived from CT images on the severity of CRS in 45 patients with advanced gastric cancer treated with CLDN18.2-targeted CAR-T cells. The waist circumference, skeletal muscle index (SMI), skeletal muscle density (SMD), subcutaneous fat area (SFA), visceral fat area (VFA) and VFA-to-SFA ratio (VSR) on baseline CT were automatically segmented and calculated using a deep learning-based tool. The relationship between SII, body composition, and CRS severity was investigated by using ROC analysis, univariate and multivariate binary logistic regression. There were no significant differences in SMI, SMD, SFA and waist circumference between patients with CRS grade 1 and 2. CRS grade 2 patients exhibited significantly higher VSR and SII than patients with CRS grade 1 (P = 0.003 and 0.012, respectively). ROC analysis showed that the AUCs of VSR and SII for predicting CRS grade were 0.762 (0.620–0.905) and 0.721 (0.563–0.879), respectively. Logistic regression analysis demonstrated that SII > 553 × 10⁹/L and VSR ≥ 0.21 were significantly linked with high grade CRS (P = 0.035 and 0.014, respectively). We constructed a two-step scoring CRS prediction model based on VSR and SII, and the AUC of this model achieved 0.802 (0.665–0.939) for predicting high-grade CRS in advanced gastric cancer patients receiving CLDN18.2-targeted CAR-T cell therapy, providing a practical tool for early risk stratification and clinical intervention.

Our study aimed to identify CT features of locally advanced gastric cancer (LAGC) associated with deficient mismatch repair (dMMR), explore the underlying pathological basis, and develop a morphology-based model for dMMR prediction. This multicenter, retrospective analysis included 397 patients with LAGC treated at three institutions between January 2016 and January 2022. Mismatch repair (MMR) status was determined by immunohistochemical analysis of postoperative specimens. The patients were divided into training and external validation groups. CT features of primary tumours and lymph nodes associated with MMR status were identified, and the underlying pathological basis for these features were explored. Multivariable logistic analysis was used to identify independent CT features and to develop a morphology-based model. Tumours with dMMR were characterized by increased thickness (p = 0.04), location in the lower stomach (p < 0.001), heterogeneous enhancement patterns (p = 0.02), a well-defined margin (p < 0.001), the presence of mushroom sign (p < 0.001), and the presence of abnormal lymph nodes (p = 0.07). Pathologically, tumours with dMMR exhibited extensive lymphocytic infiltration and solid growth with a pushing border, consistent with the distinctive CT features observed. Multivariable logistic analysis identified lower tumour location, well-defined tumour margin, and mushroom sign as independent predictors of dMMR. The morphology-based model achieved AUC values of 0.73-0.74 for dMMR prediction in the training and external validation groups. LAGC with dMMR showed distinct CT features. A morphology-based model constructed from these CT features has the potential to predict dMMR status for LAGC.