Automatic Diagnosis of Traumatic Brain Injury Using Deep Learning with CT scan Images
نویسندگان
1 Associate Professor of Emergency medicine, Department of Emergency Medicine, Faculty of Medicine, Mashhad University of Medical Sciences, Mashhad, Iran
2 Associate Professor of Emergency medicine, Department of Emergency Medicine, Faculty of Medicine, Mashhad University of Medical Sciences, Mashhad, Iran
3 Master of Biomedical Engineering, Clinical Research Development Unit ,Shahid Hasheminejad Hospital, Medical Sciences University Of Mashhad, Mashhad,Iran
4 Associate Professor of Emergency medicine, Department of Emergency Medicine, Faculty of Medicine, Mashhad University of Medical Sciences, Mashhad, Iran
5 Assistant professor of Emergency medicine, Department of Emergency Medicine, Faculty of Medicine, Mashhad University of Medical Sciences, Mashhad, Iran
6 Associate Professor of Emergency medicine, Department of Emergency Medicine, Faculty of Medicine, Mashhad University of Medical Sciences, Mashhad, Iran
doi
10.22038/ijmp.2025.78206.2392چکیده
Introduction: Traumatic brain injury (TBI) results from external mechanical forces to the head, leading to brain dysfunction. The severity of injury significantly impacts patient health outcomes. Rapid and accurate diagnosis is essential for timely clinical intervention. Computed Tomography (CT) scans are currently the primary imaging modality for identifying intracranial injuries. However, manual analysis of CT images is time-consuming and highly dependent on radiologists’ expertise.Material and Methods: This study proposes an automated approach for detecting intracranial hemorrhage and skull fractures using Convolutional Neural Networks (CNNs). CT scan images containing various pathologies were collected from the Picture Archiving and Communication System (PACS). The dataset was divided into two classes: pathological and non-pathological. Images were resized to 128 × 128 pixels to reduce computational complexity and split into training (90%) and validation (10%) sets. Pre-trained ResNet18 and ResNet34 models were employed for classification. Evaluation metrics such as accuracy, precision, recall, and F-score were computed using a confusion matrix.Results: The CNN model achieved an accuracy of 0.94, a precision of 1.0, and a recall of 0.88 in classifying CT images.Conclusion: These findings indicate that CNN-based models can assist radiologists in faster and more consistent diagnosis of traumatic brain injuries. Further improvements may be achieved by increasing dataset size, refining preprocessing steps, and applying advanced optimization techniques to enhance generalization and robustness.
