The latest study from our group integrates Imaging and Clinical Data using a Multimodal Machine Learning Model. This time, the question under investigation is the prediction of Anastomotic Leakage after Esophageal Cancer Surgery.
Take-home messages:
Logistic regression demonstrated that CT-based aortic calcification score, a CRP at postoperative day 2 (CRPd2) > 211 mg/L (P < 0.001) and at postoperative day 3 (CRPd3) > 222 mg/L (P < 0.001), as well as ASA score (P = 0.003), were significant predictors of postoperative AL.
An XGBoost model was constructed using the parameters found significant in logistic regression. For the prediction of AL, the model achieved an AUC of 79.2% (95% CI 69%–89.4%) with a specificity of 77.46%, a sensitivity of 65.22%, a PPV of 48.39%, an NPV of 87.3% and F1-score of 56%, Figure 4a. SHAP analysis showed that the aortic calcification score had the highest contribution to the model output, with a low aortic score predicting negative and a high aortic score predicting positive AL.
Karolinska University Hospital, Karolinska Institutet, Ioannis Rouvelas, Michail Klontzas, MD, PhD, George Kalarakis, Emmanouil Koltsakis, Erik Stenqvist et al
Kudos to this incredible team - we are honored to be part of this study & believe it is a testament to the power of AI in healthcare, transforming clinical data into actionable insights & bettering patient care!