Related Experiment Video
Updated: Sep 20, 2025
06:44
Modeling Brain Metastases Through Intracranial Injection and Magnetic Resonance Imaging
Published on: June 7, 2020
7.5K
Convolutional neural network using magnetic resonance brain imaging to predict outcome from tuberculosis meningitis
Trinh Huu Khanh Dong1,2, Liane S Canas2, Joseph Donovan1,3
1Oxford University Clinical Research Unit, Viet Nam.
Plos One
|May 23, 2025
Summary
Brain MRI aids in predicting outcomes for tuberculous meningitis (TBM). A fused model combining MRI and clinical data improved predictions of complications and death in TBM patients, highlighting imaging
Area of Science:
- Neurology
- Infectious Diseases
- Medical Imaging
Background:
- Tuberculous meningitis (TBM) is a severe condition with high mortality, particularly in individuals with HIV.
- Predicting complications in TBM is challenging, and the role of brain magnetic resonance imaging (MRI) remains underexplored.
- Convolutional neural networks (CNNs) offer a potential tool to analyze complex imaging data for prognostic insights.
Purpose of the Study:
- To investigate the added value of brain MRI in predicting mortality and neurological complications in tuberculous meningitis (TBM).
- To compare the predictive performance of models using clinical/demographic data, MRI data alone, and a combination of both.
- To explore the utility of CNNs in analyzing T1-weighted MRI volumes for TBM prognostication.
Main Methods:
- Data from two randomized controlled trials involving HIV-positive and HIV-negative adults with TBM were pooled.
- Three predictive models were developed: logistic regression (clinical data), CNN (MRI data), and a fused model (all data).
- Models were validated using cross-validation and a 70/30 training/test split, stratified by outcome and HIV status.
Main Results:
- The fused model demonstrated superior predictive performance (AUC = 77.3%) compared to the non-imaging (AUC = 71.2%) and imaging-only (AUC = 67.3%) models.
- Clinical data were more informative for HIV-positive patients, while MRI features were more predictive for HIV-negative patients.
- Interpretability maps indicated the model focused on specific brain regions, including lateral fissures, corpus callosum, midbrain, and peri-ventricular tissues.
Conclusions:
- Brain MRI provides valuable complementary information for predicting adverse outcomes in tuberculous meningitis.
- A combined approach using both imaging and clinical data offers the most robust prediction of TBM complications.
- Larger datasets are required to further validate these findings and confirm the prognostic value of MRI in TBM.
