Vision Transformer-based Decision Support for Neurosurgical Intervention in Acute Traumatic Brain Injury: Automated Surgical Intervention Support Tool (ASIST-TBI)

• Published in: Radiology. Artificial intelligence Vol. 6; no. 2; p. e230088
• Main Authors: Smith, Christopher W, Malhotra, Armaan K, Hammill, Christopher, Beaton, Derek, Harrington, Erin M, He, Yingshi, Shakil, Husain, McFarlan, Amanda, Jones, Blair, Lin, Hui Ming, Mathieu, François, Nathens, Avery B, Ackery, Alun D, Mok, Garrick, Mamdani, Muhammad, Mathur, Shobhit, Wilson, Jefferson R, Moreland, Robert, Colak, Errol, Witiw, Christopher D
• Format: Journal Article
• Language: English
• Published: United States Radiological Society of North America 01-03-2024
• Subjects: Original Research
• ISSN: 2638-6100; 2638-6100
• DOI: 10.1148/ryai.230088

Purpose To develop an automated triage tool to predict neurosurgical intervention for patients with traumatic brain injury (TBI) without using image-level labels. Materials and Methods A provincial trauma registry was reviewed to retrospectively identify patients with TBI from 2005-2022 treated at a specialized Canadian trauma center. Model training, validation, and testing was performed using head CT scans with binary reference standard patient-level labels corresponding to whether the patient received neurosurgical intervention. Performance and accuracy of model, termed Automated Surgical Intervention Support Tool for TBI (ASIST-TBI), were also assessed using a held-out consecutive test set of all patients with TBI presenting to our center between March 2021 and September 2022. Results Head CT-scans from 2,806 patients with TBI (mean age, 57 years (SD, 22); 1955 (70%) male) were acquired between 2005-2021 and used for training, validation, and testing. Consecutive scans from an additional 612 patients (mean age, 61 years (SD, 22); 443 (72%) male) were used to assess the performance of ASIST-TBI. There was accurate prediction of neurosurgical intervention with an area under the receiver operating curve (AUC) of 0.92 [95% CI: 0.88-0.94], 87% (491/562) accuracy, 87% (196/225) sensitivity, and 88% (295/337) specificity on the test dataset. Performance on the held-out test dataset remained robust with an AUC of 0.89 [95% CI: 0.85-0.91], 85% (199/235) sensitivity, 84% (318/377) specificity, and 84% (517/612) accuracy. Conclusion A novel deep learning model was developed that could accurately predict the requirement for neurosurgical intervention using acute TBI CT scans. ©RSNA, 2024.