Related Experiment Video
Updated: Nov 25, 2025
10:25
Brain Infarct Segmentation and Registration on MRI or CT for Lesion-symptom Mapping
Published on: September 25, 2019
48.9K
Multiclass semantic segmentation and quantification of traumatic brain injury lesions on head CT using deep learning:
Miguel Monteiro1, Virginia F J Newcombe2, Francois Mathieu2
1Biomedical Image Analysis Group, Department of Computing, Imperial College London, London, UK.
The Lancet. Digital Health
|December 17, 2020
Summary
Deep learning models can now automatically detect and quantify traumatic brain injury lesions from CT scans. This AI-driven approach aids in assessing injury severity and tracking progression for personalized TBI treatment.
Area of Science:
- Neuroimaging
- Artificial Intelligence
- Medical Diagnostics
Background:
- Computed Tomography (CT) is the primary imaging method for traumatic brain injury (TBI).
- Conventional CT interpretation requires expertise and lacks detailed quantitative data crucial for prognosis.
- There is a need for automated, quantitative analysis of TBI lesions.
Purpose of the Study:
- To develop and validate a deep learning model for automated quantification and detection of various TBI lesion types.
- To enable reliable and efficient analysis of CT scans for TBI assessment.
Main Methods:
- A convolutional neural network (CNN) was trained on expert-annotated CT scans for lesion segmentation.
- The CNN was iteratively refined using manually corrected segmentations from a large dataset.
- Performance was evaluated on lesion volume quantification, progression, detection, and classification using independent test and external validation sets.
Main Results:
- The developed CNN accurately segmented and quantified multiple hemorrhagic lesion types and perilesional edema.
- Volumetric estimates showed minimal differences compared to manual references across lesion types.
- The model demonstrated reliable performance in lesion detection and classification.
Conclusions:
- Deep learning enables precise segmentation, quantification, and detection of TBI lesions.
- Volumetric lesion data can significantly aid in assessing TBI burden and progression.
- This technology holds potential for personalized TBI treatment strategies and research.
