CT delta-radiomics predicts the risks of blood transfusion and massive bleeding during spinal tumor surgery

Abstract

Background: Intraoperative bleeding is a serious complication of spinal tumor surgery. Preoperative identification of patients at high risk of intraoperative blood transfusion (IBT) and intraoperative massive bleeding (IMB) before spinal tumor resection surgery is difficult but critical for surgical planning and blood management. This study aims to develop and validate delta radiomics prediction models for IBT and IMB in spinal tumor surgery.

Methods: Patients diagnosed with spinal tumors who underwent spinal tumor resection surgery were retrospectively recruited. CT, CTE, delta, and clinical models based on CT native phase, CT arterial phase images, and clinical factors were constructed using 10-fold cross-validation and logistic regression (LR), random forest (RF), and support vector machine (SVM) in the training cohort. Receiver operating characteristic (ROC) curves, integrated discrimination improvement (IDI), accuracy, sensitivity, specificity, positive predictive value, and negative predictive value were used to evaluate and compare the diagnostic performance of these models.

Results: 231 patients were randomly divided into training (n = 161) and test (n = 70) cohorts, comprising 146 IBT and 85 no-IBT patients, 35 IMB and 196 no-IMB patients, respectively. The delta model performed best in predicting IBT and IMB risk, with better predictive ability than the clinical model (IDI = 0.11-0.13 for IBT, and IDI = 0.02-0.08 for IMB, p < 0.05, respectively). Calibration curves indicated that the predicted probabilities of IBT and IMB in the model did not differ significantly from the actual probabilities (p > 0.05).

Conclusion: The CT delta model we constructed may be a valuable tool to improve risk stratification before spinal tumor surgery, thus contributing to preoperative planning and improving patient prognosis.

Trial registration: Retrospectively registered (M2020435).

Keywords: Blood transfusion; Delta-radiomics; Intraoperative bleeding; Logistic regression; Spinal tumor.

Fig. 1

Flowchart of patient inclusion

Fig. 2

Workflow for VOI segmentation, radiomics feature extraction, model building, and downstream analysis. AUC, area under the receiver operating characteristic curve; DCA, decision curve analysis; ICC, intraclass correlation coefficient; RFE, recursive feature elimination; LASSO, least absolute shrinkage and selection operator; LR, logistic regression; RF, random forest; SVM, support vector machine; VOI, volume of interest

Fig. 3

Assessment of models for the ability to predict intraoperative transfusion. ROC curves of clinical, CT, CTE, and delta models are used to predict intraoperative transfusion in the training (A, C, E, G) and test sets (B, D, F, H). ROC, receiver operating characteristic

Fig. 4

Assessment of models for the ability to predict intraoperative massive bleeding model. ROC curves of clinical, CT, CTE, and delta models are used to predict intraoperative massive bleeding in the training (A, C, E, G) and test sets (B, D, F, H). ROC, receiver operating characteristic

Fig. 5

Calibration curves and Nomogram of the Delta model for predicting Intraoperative transfusion (A, C) and Intraoperative massive bleeding (B, D)

Fig. 6

Importance of logistic regression features of the delta model for predicting intraoperative transfusion (A) and intraoperative massive bleeding (B)