Comparative accuracy of spectral computed tomography and conventional computed tomography in colorectal cancer staging

Background Accurate pre-treatment staging is essential for risk stratification and treatment planning in colorectal cancer (CRC). Spectral computed tomography (CT) provides energy-resolved reconstructions that may improve local and nodal assessment compared with conventional contrast-enhanced CT. Methods This retrospective study included 340 adults with pathologically confirmed CRC who underwent spectral CT (n = 151) or conventional CT (n = 189) before definitive treatment between September 2023 and September 2025. The reference standard was postoperative histopathologic staging for resected cases or a composite standard incorporating biopsy, imaging, and clinical follow-up for non-surgical cases. Primary outcomes were overall TNM accuracy and component T, N, and M staging accuracy. Agreement was assessed using kappa statistics. Binary diagnostic performance was calculated for nodal metastasis and distant metastasis. Multivariable logistic regression was performed to account for potential residual confounding. Results Spectral CT showed higher overall TNM accuracy than conventional CT (74.2% vs 63.0%, P = 0.028). It also demonstrated superior T-stage accuracy (74.2% vs 57.7%, P = 0.002) and N-stage accuracy (74.8% vs 64.0%, P = 0.032), whereas M-stage accuracy was similar (90.7% vs 91.0%, P = 0.930). Agreement with the reference standard was higher for spectral CT for overall TNM (weighted κ = 0.815 vs 0.795), T stage (0.759 vs 0.663), and N stage (0.599 vs 0.430). For M1 detection, spectral CT showed higher sensitivity (83.8% vs 72.2%) with slightly lower specificity (93.0% vs 95.4%). In adjusted analyses, spectral CT remained independently associated with correct overall TNM staging (adjusted odds ratio 1.91, 95% confidence interval 1.14–3.19, P = 0.014). Conclusions Under standardized protocols, spectral CT demonstrated higher overall TNM staging accuracy than conventional CT, primarily through improved T and N classification, while M staging performance was comparable.