Radiomics of skeletal muscle helps to predict gastrointestinal toxicity in locally advanced rectal cancer patients receiving neoadjuvant chemoradiotherapy

Abstract

BackgroundThe skeletal muscle index (SMI) can serve as a surrogate for a patient’s nutritional status, which is associated with treatment toxicity. This study aims to investigate the potential of baseline skeletal muscle radiomics features to predict gastrointestinal toxicity of neoadjuvant chemoradiotherapy for rectal cancer. MethodsA total of 214 rectal cancer patients (115, 49 and 50 in the training, internal and external validation set, respectively) who underwent neoadjuvant pelvic radiotherapy with capecitabine and irinotecan were retrospectively identified. The skeletal muscle at the level of the third lumber vertebra was contoured, and the radiomics features were extracted from computed tomography scans. In the training set, the least absolute shrinkage and selection operator (LASSO) regression algorithm was applied to select features that were most significantly associated with grade 3–4 gastrointestinal toxicity (diarrhea, nausea, vomiting and proctitis). The predictive performance and clinical utility were estimated using the area under the receiver operator characteristic curve (AUC), F1-score and decision curve analysis (DCA). ResultsNine features, including the SMI and eight radiomics features, were associated with grade 3–4 gastrointestinal toxicity and included in the logistic regression. This combined predictive model, which incorporated the SMI and radiomics features, showed better discrimination than the SMI alone, with an AUC of 0.856 (95 % CI: 0.782–0.929) in the training cohort, 0.812 (95 % CI: 0.667–0.956) in the internal validation cohort and 0.745 (95 % CI: 0.600–0.890) in the external validation cohort. DCA further verified the clinical utility of the combined predictive model. ConclusionRadiomics features of skeletal muscle were significantly associated with gastrointestinal toxicity. The predictive model incorporating the SMI and radiomics features exhibits favorable discrimination and may be highly informative for clinical decision-makings.