Preoperative Contrast-Enhanced CT-Based Three-Dimensional Fractal Analysis of Automatically Segmented Tumors for Predicting PD-L1 Expression in Hepatocellular Carcinoma: A Preliminary Study
Zhengfeng Fu, Yuan Xu, Yufeng Li, Xiaotao Guo, Guojian Ren, Yizhao Zhang, Pengfei SunObjectives:
To evaluate whether 3-dimensional fractal features extracted from automatically segmented hepatocellular carcinoma (HCC) tumor volumes on preoperative contrast-enhanced CT (CECT) can predict programmed death-ligand 1 (PD-L1) expression.
Methods:
This retrospective study included patients with pathologically confirmed HCC between January 2022 and March 2026 who underwent multiphasic liver CECT and PD-L1 immunohistochemical staining of surgical or biopsy tumor specimens. PD-L1 positivity was defined as a combined positive score ≥1. Patients were randomly divided into training and validation cohorts. Tumor VOIs were automatically segmented on portal venous phase CECT using the uAI Research Portal. Three-dimensional fractal dimension (FD 3D ) and 3-dimensional lacunarity (Lac 3D ) were extracted using the box-counting method. LASSO logistic regression was used to select predictors of PD-L1 positivity and to build a combined model.
Results:
A total of 134 patients with HCC were included, of whom 69 had PD-L1-positive tumors, and 65 had PD-L1-negative tumors. Patients were randomly assigned to the training cohort (n=87; 44 PD-L1-positive) and validation cohort (n=47; 25 PD-L1-positive). In both cohorts, tumor FD
3D
and Lac
3D
were significantly higher in PD-L1-positive patients than in PD-L1-negative patients (both
Conclusion:
Tumor 3-dimensional fractal features derived from preoperative CECT can quantitatively reflect the spatial structural complexity and heterogeneity of HCC. A combined model integrating FD 3D , Lac 3D , and RVI may facilitate noninvasive preoperative prediction of PD-L1 expression and serve as an imaging biomarker for characterizing tumor immune-related features.