Predicting intraoperative blood loss during cesarean sections based on multi-modal information: a two-center study

Area of Science:

• Medical imaging and machine learning
• Obstetrics and Gynecology
• Anesthesiology

Background:

• Intraoperative blood loss (IBL) during cesarean sections is a significant concern for maternal morbidity.
• Accurate prediction of IBL is crucial for effective perioperative management.
• Existing prediction methods may not fully integrate diverse patient data.

Purpose of the Study:

• To develop and validate a nomogram model for predicting intraoperative blood loss (IBL) in cesarean sections.
• To integrate radiomics features, clinical factors, and coagulation function indexes (CFI) into a predictive model.
• To assess the model's potential for optimizing perioperative care and reducing maternal complications.

Main Methods:

• Retrospective analysis of 346 patients undergoing cesarean sections with magnetic resonance imaging.
• Development of machine-learning prediction models using clinical factors, radiomics features, and CFI.
• Validation of models using internal and external test sets with ROC analysis for IBL diagnosis (IBL+ > 1000 mL).

Main Results:

• The combined model achieved an AUC of 0.873 (internal) and 0.806 (external), demonstrating good predictive performance.
• The nomogram incorporating all three modalities showed high AUCs (0.960 internal, 0.869 external) in a subset of patients.
• The proposed model's performance was significantly better than CFI alone but not statistically superior to the Clinical-CFI model in validation sets.

Conclusions:

• The developed nomogram shows strong potential for predicting IBL in cesarean sections, with high sensitivity and generalizability.
• The model may be applicable to other obstetric procedures like vaginal delivery and postpartum hysterectomy.
• Further research is needed as the model did not demonstrate statistically significant superiority over the Clinical-CFI model.