Pretreatment Coronary Ultrasound Radiomics Combined With the Prognostic Nutritional Index for Predicting IVIG...

Intravenous immunoglobulin (IVIG) resistance in children with Kawasaki disease (KD) is associated with an increased risk of coronary artery injury and subsequent adverse cardiovascular outcomes. Early identification of patients at risk remains difficult in clinical practice. Conventional coronary ultrasound is largely based on geometric parameters, such as luminal diameter and Z scores, and may not fully capture local heterogeneity within coronary images. In this study, we investigated whether pretreatment coronary artery ultrasound radiomics could improve early identification of IVIG resistance in children with KD. We further developed a multimodal prediction model incorporating radiomics, the prognostic nutritional index (PNI), and clinical characteristics.

A total of 352 children diagnosed with KD and treated at The First Affiliated Hospital of Guangxi Medical University between January 2020 and January 2026 were retrospectively included. All patients received standardized treatment after diagnosis. Demographic characteristics, pretreatment laboratory results, conventional coronary ultrasound parameters, and two-dimensional static ultrasound images of the left coronary artery (LCA), left anterior descending artery (LAD), and right coronary artery (RCA) were collected. The pretreatment two-dimensional static coronary ultrasound images were imported into 3D Slicer for region-of-interest segmentation. A clinical prediction model was constructed using multivariable logistic regression. PNI was defined a priori as the primary nutritional-immunologic indicator. In addition, the potential incremental predictive value of other nutritional-inflammatory indices, including the C-reactive protein-to-albumin ratio (CAR), neutrophil-to-albumin ratio (NAR), neutrophil percentage-to-albumin ratio (NPAR), and C-reactive protein-albumin-lymphocyte (CALLY) index, was further evaluated. Radiomics features were selected through repeated segmentation, intraclass correlation coefficient (ICC)-based stability filtering, correlation filtering, and least absolute shrinkage and selection operator (LASSO) regression. The Clinic, Clinic + PNI, Radiomics, and Clinic + PNI + Radiomics models were assessed using receiver operating characteristic curves, DeLong tests, calibration curves, decision curve analysis, and Shapley Additive Explanations (SHAP) interpretation. The Kobayashi, Egami, and Sano scores were evaluated in the same training and validation cohorts.

Among the 352 children, 57 (16.2%) were classified as IVIG resistant. Multivariable analysis showed that higher pretreatment maximum coronary artery Z score (Z_max), longer fever duration, and higher neutrophil count were associated with an increased risk of IVIG resistance. In the validation cohort, the Clinic + PNI model showed favorable discriminatory ability, with an AUC of 0.855 (95% CI, 0.767-0.911), a sensitivity of 0.882, and a negative predictive value of 0.969. Seven radiomics features were selected and used to construct the radiomics score (Rad-score). In the training cohort, the Clinic + PNI + Radiomics model achieved higher AUC values than the Clinic, Clinic + PNI, and Radiomics models. This pattern was also observed in the validation cohort, where the integrated model achieved the highest AUC of 0.880 (95% CI, 0.817-0.944), although the between-model differences did not reach statistical significance. Decision curve analysis showed that the integrated model offered a favorable net benefit within clinically meaningful threshold ranges. SHAP analysis suggested that its predictions reflected the joint influence of systemic inflammatory burden, nutritional and immune status, and local imaging heterogeneity captured from coronary ultrasound images. The Kobayashi, Egami, and Sano scores yielded validation AUCs of 0.616, 0.567, and 0.449, respectively.

Pretreatment coronary ultrasound radiomics may provide complementary information to conventional clinical, laboratory, and ultrasonographic parameters for assessing the risk of IVIG resistance in children with KD. By integrating inflammatory burden, nutritional-immunologic status, and local coronary imaging heterogeneity, the Clinic + PNI + Radiomics model showed favorable discrimination and potential clinical net benefit in internal validation. However, its incremental value and clinical applicability require further confirmation in external cohorts.