Automated Hematoma Detection and Outcome Prediction in Patients With Traumatic Brain Injury.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Traumatic brain injury (TBI) often involves intracranial hemorrhages (ICH).
• Accurate and timely assessment of ICH is crucial for patient management and outcome prediction.
• Current methods for ICH analysis can be time-consuming and subjective.

Purpose of the Study:

• To develop an automated tool for subtype classification and segmentation of ICH on CT scans.
• To predict patient outcomes, specifically 14-day in-hospital mortality.
• To enhance clinical decision-making for TBI patients with ICH.

Main Methods:

• A cascade framework was developed for two-stage segmentation and multi-label classification of ICH.
• Hematoma region of interest (ROI) localization, cropping, and resizing were performed.
• A deep learning approach fused local and global features for classification.
• Hematoma features were integrated with the CRASH model for mortality prediction.

Main Results:

• The segmentation method achieved high accuracy, indicated by Dice similarity and Jaccard Index.
• The multi-label classification achieved an average accuracy of 95.91%.
• The mortality prediction model demonstrated an average AUC of 0.91.

Conclusions:

• The developed method improves the precision of ICH segmentation and subtype classification.
• This tool can streamline ICH evaluation for radiologists.
• Automated feature extraction is expected to aid in prognosis assessment for TBI patients.