Preoperative prediction of lymphovascular and perineural invasion in colorectal cancer using quantitative spectral CT measures

Perineural invasion (PNI) and lymphovascular invasion (LVI) are adverse prognostic factors in colorectal cancer (CRC), but their preoperative identification remains challenging. This study evaluated a comprehensive set of quantitative spectral computed tomography (CT) parameters, including arterial enhancement fraction (AEF) and normalized AEF (NAEF), for detecting an exploratory composite endpoint defined as the presence of either PNI or LVI. PNI and LVI were also evaluated separately. This single-center retrospective study included 120 patients with surgically confirmed colorectal adenocarcinoma who underwent preoperative dual-energy spectral CT between November 2023 and August 2024. For the exploratory composite endpoint, patients were classified as PNI/LVI-positive if either PNI or LVI was present ( n = 48) and PNI/LVI-negative if both were absent ( n = 72); PNI and LVI were also analyzed separately as secondary outcomes. Quantitative parameters—including iodine concentration (IC), normalized IC (NIC), monoenergetic attenuation at 40 and 100 keV, spectral slope (λ), effective atomic number (Eff-Z), AEF, and NAEF—were extracted from arterial- and venous-phase images. Parameters that differed significantly between groups in univariable analyses ( P < 0.05) were entered jointly into multivariable logistic regression models. Discrimination was assessed using receiver operating characteristic (ROC) analysis. Internal validation used 1,000 bootstrap resamples in which the complete model-development process, including univariable screening and multivariable model fitting, was repeated. Optimism was estimated as the difference between performance in each bootstrap sample and the performance of the corresponding bootstrap-fitted model in the original cohort; mean optimism was subtracted from apparent performance to obtain optimism-corrected estimates. The same procedure was applied to the PNI/LVI, PNI-specific, and LVI-specific models. The PNI/LVI-positive group had significantly higher arterial- and venous-phase IC, NIC, 40-keV attenuation, and λ (all P < 0.001), as well as higher venous-phase 100-keV attenuation, Eff-Z, AEF, and NAEF (all P ≤ 0.036). Individual parameters yielded apparent areas under the ROC curve (AUCs) of 0.612–0.810. The combined PNI/LVI model had an apparent AUC of 0.874 (95% CI, 0.809–0.939), which decreased to an optimism-corrected AUC of 0.815 after bootstrap internal validation. The corresponding optimism-corrected AUCs were 0.745 for PNI and 0.766 for LVI. In this single-center cohort, quantitative spectral CT parameters were associated with PNI and LVI status in CRC. The combined model retained an optimism-corrected AUC of 0.815 after bootstrap internal validation; however, its external performance and generalizability remain unknown. These hypothesis-generating findings require independent multicenter validation before clinical application and support further investigation of quantitative spectral CT parameters as potential preoperative imaging biomarkers of adverse microscopic invasion.