Exploratory whole-tumor and habitat radiomic associations with study-documented Clonorchis sinensis status in hepatocellular carcinoma

Aim: To explore associations of whole-tumor and habitat radiomics with study-documented Clonorchis sinensis (Cs) status in hepatocellular carcinoma. Methods: This retrospective study included 161 patients: 77 with computed tomography (CT) and 84 with magnetic resonance (MR) imaging. A patient-level split yielded 128 training and 33 internal holdout patients. Modality-specific habitats were constructed from training voxels with patient-balanced sampling. An ExtraTrees model combined whole-tumor features and label-invariant habitat summaries. Modality-only and MR acquisition-only controls were compared on the same holdout. Additional MR analyses examined resampling and acquisition sensitivity. Case-level infection ascertainment and clinical covariates were unavailable. Results: Holdout area under the receiver operating characteristic curve (AUROC) was 0.833 [95% confidence interval (CI): 0.683-0.955]. The model exceeded the modality-only control (AUROC 0.643; paired difference +0.190, 95%CI: 0.048-0.339; P = 0.008). The pooled-model MR subgroup did not outperform the acquisition control (AUROC 0.846 vs. 0.885; difference -0.038, 95%CI: -0.232-0.115; P = 0.742). Among 90 MR development patients, AUROC from probabilities averaged across three nested cross-validation repeats was 0.854 (95%CI: 0.765-0.926); 14 patients overlapped the original holdout. Restriction to shared acquisition support yielded AUROC 0.684 in 34 patients. Conclusion: Radiomics showed exploratory associations with study-documented Cs status. Acquisition coupling, endpoint uncertainty, unmeasured clinical confounding and reused evaluation data prevent interpretation as infection diagnosis or independent validation.