Combining clinical-radiomics features with machine learning methods for building models to predict postoperative recurrence in patients with chronic subdural hematoma: Retrospective Cohort Study (Preprint)

Abstract

Background: Chronic subdural hematoma (CSDH) represents a prevalent medical condition, posing significant challenges in postoperative management due to risks of recurrence. Currently, prognosis determination largely depends on clinician expertise, revealing a dearth of precise prediction models in clinical settings.

This study sought to employ machine learning (ML) techniques for the construction of predictive models to assess the likelihood of CSDH recurrence post-surgery.

Methods: Data from 133 patients were amassed and partitioned into a training set (n=114) and a test set (n=19). Radiomics features were extracted from preoperative cranial computed tomography (CT) scans utilizing 3D Slicer software. These features, in conjunction with clinical data and composite clinical-radiomics features, served as input variables for model development. Four distinct ML algorithms were utilized to build predictive models, and their performance was rigorously evaluated via accuracy (ACC), area under the curve (AUC), and Recall metrics. The optimal model was identified, followed by recursive feature elimination (RFE) for feature selection, leading to enhanced predictive efficacy. External validation was conducted using datasets from additional healthcare facilities.

Results: Following rigorous experimental analysis, the Support Vector Machine (SVM) model, predicated on clinical-radiomics features, emerged as the most efficacious for predicting postoperative recurrence in CSDH patients. Subsequent to feature selection, key variables exerting significant impact on the model were incorporated as the input set, thereby augmenting its predictive accuracy. The model demonstrated robust performance, with metrics including accuracy (ACC) of 92.72%, area under the curve (AUC) of 91.34%, and Recall of 93.16%. External validation further substantiated its effectiveness, yielding an ACC of 90.32%, AUC of 91.32%, and Recall of 88.37%, affirming its clinical applicability.

Conclusion: The present study substantiates the feasibility and clinical relevance of a machine learning-based predictive model, utilizing clinical-radiomics features, for precise prognostication of postoperative recurrence in CSDH patients. This model holds considerable import for enhancing the quality and efficiency of clinical decision-making processes.