A deep learning approach to predict temporal changes of subdural hemorrhage on computed tomography

Abstract

Subdural hemorrhage (SDH) is a critical condition requiring prompt assessment of its progression using computed tomography (CT). This study aimed to develop a deep-learning model to predict temporal changes in SDH by leveraging Hounsfield Units (HU) to estimate hemorrhage age across acute, subacute, and chronic stages. A total of 825 pre-processed CT slices from the RSNA dataset were balanced across SDH stages and analyzed using a convolutional neural network (CNN) implemented in Python on Google Colab. Model performance was evaluated using accuracy, sensitivity, specificity, precision, F1-score, and the Area Under the Receiver Operating Characteristic Curve (AUC-ROC). The model achieved 83.11% training accuracy and 85.33% prediction accuracy. Sensitivity for acute, subacute, and chronic SDH was 86.67%, 84%, and 85.33%, respectively, with specificity values of 94%, 88%, and 96%. Precision scores were 87.84%, 77.78%, and 91.43%, while F1 scores were 87.25%, 80.77%, and 88.28%. AUC-ROC values ranged from 0.9394 to 0.9731 across five folds, reflecting robust classification performance. The results highlight the model's potential to support radiologists as a second-reader tool, streamline emergency triage, and enhance diagnostic efficiency within clinical workflows.

Ähnliche Arbeiten