Back to the list
Congress: ECR25
Poster Number: C-11627
Type: Poster: EPOS Radiologist (scientific)
DOI: 10.26044/ecr2025/C-11627
Authorblock: Y. Xu, J. Liu, J. Zhou; Lanzhou City/CN
Disclosures:
Yuan Xu: Nothing to disclose
Jianli Liu: Nothing to disclose
Junlin Zhou: Nothing to disclose
Keywords: Abdomen, Artificial Intelligence, Liver, Conventional radiography, CT-Quantitative, Decision analysis, Surgery, Cancer, Neoplasia
Results

Multivariate logistic regression analysis revealed that kurtosis[5] (P < 0.01), platelet-to-lymphocyte ratio (PLR)[6] (P = 0.03), international normalized ratio (INR)[7] (P < 0.01), albumin-bilirubin (ALBI)[8] grade (P < 0.01), and microvascular invasion (MVI)[7, 9] (P < 0.01) were independent risk factors for ER. The clinical-imaging-histogram (CIH) model demonstrated the best predictive performance for ER (AUC = 0.90), surpassing the clinical model and the imaging-histogram (IH) model (AUCs of 0.86 and 0.71, respectively). The CIH model's calibration curve showed good consistency, and the DCA indicated a high level of clinical utility.多变量logistic回归分析显示,峰度[5](P 0.01)、血小板-淋巴细胞比(PLR)[6](P = 0.03)、国际标准化比值(INR)[7](P 0.01)、白蛋白-胆红素(ALBI)[8]分级(P 0.01)和微血管浸润(MVI)[7,9](P 0.01)是ER的独立风险因素。临床成像直方图(CIH)模型表现出对ER的最佳预测性能(AUC = 0.90),超过了临床模型和成像直方图(IH)模型(AUC分别为0.86和0.71)。CIH模型的校准曲线显示出良好的一致性,DCA表明具有较高的临床实用性。

GALLERY