Prediction of prognosis in patients with cerebral contusions based on machine learning.
Journal:
Scientific reports
PMID:
39738368
Abstract
Traumatic brain injury (TBI) is a global issue and a major cause of patient mortality, and cerebral contusions (CCs) is a common primary TBI. The haemorrhagic progression of a contusion (HPC) poses a significant risk to patients' lives, and effectively predicting changes in haematoma volume is an urgent clinical challenge that needs to be addressed. As a branch of artificial intelligence, machine learning (ML) can proficiently handle a wide range of complex data and identify connections between data for tasks such as prediction and decision making. We collected data from 673 CCs patients who were hospitalized in the neurosurgery department of The First Affiliated Hospital of Wannan Medical College (Yijishan Hospital of Wannan Medical College) from September 2019 to September 2022. Selecting three popular machine learning algorithms, Decision Tree (DT), Random Forest (RF), and Multilayer Perceptron (MLP) to predict hematoma. Machine learning algorithms were run on the Python 3.9 platform. The model was evaluated for sensitivity, specificity, F1 score, accuracy, receiver operating characteristic (ROC) curves, and the area under the receiver operating characteristic curve (AUC). Using sensitivity as the evaluation metric, the best model is DT model. The DT model included the initial haematoma volume, GCS score, Fib level, blood sugar level, multiple CCs, Male, PT, blood sodium level and PLT count. The evaluation indicators of the DT model were as follows: sensitivity = 0.9545 (0.857, 1.0), specificity = 0.9803 (0.9602, 0.9952), F1 score = 0.8936 (0.7742, 0.9778), accuracy = 0.9778 (0.9556, 0.9956), and AUC-ROC = 0.9674 (0.9143, 0.9975). The DT model is the machine learning algorithm most closely aligned with the research objectives, allowing for the scientific and effective prediction of hematoma changes.