Tuberculous meningitis: understanding and improving diagnosis and prognosis using novel tools and approaches

by Trinh Dong (Université de Bordeaux)

Thursday Nov 28, 2024, 1:00 PM → 2:00 PM Europe/Paris

Amphi PA Louis (ISPED)

Speaker: Trinh Dong from the Bordeaux Population Health U1219

Abstract: Tuberculous meniningitis (TM) is the most devastating manifestation of Tuberculosis in the sub-archnoid space. It leads to most certain death if left untreated. Even under standard of care, mortaility rates can go as high as 30-50%, with unreversible disablity amongst survivors. However, despite decade-long of research, the diagnosis and progression of TM is still unclear. 
Leveraging state-of-the-art modelling techniques with new data modality, I shed light on three main questions around TM: a timely diagnosis, an early prognosis, and an understanding of treatment effects. Firstly, an early diagnosis tool for TM using easy-to-acquired biomarkers was developed, taking into account the fuzziness in disease detection by the current microbiological assays, I implemented a latent class model with local bacillary as random effects. Based on the final design, I created a web-based diagnostic application and a simplified scoring system requiring fewer features, aiming for an early screening. Secondly, a proof-of-principle prognostic model with imaging formation was developed to explore the added value of magnetic resonance brain imaging (MRI) collected from early disease courses to predict unwanted outcomes of TM. My implementation utilised a convolutional neural network with guiding auxiliary tasks. Based on the optimised model, I explored the interpretability maps to search for brain regions contributing to model decisions. Finally, to better understand the progressing trajectory of the disease manifesting on brain structures, I extracted neuromorphometric biomarkers from longitudinal MRIs and analysed the effect of standard anti-inflammatory treatment (AIT) on TM progression. I employed a Bayesian joint modelling to correct for informative dropout due to premature death. The findings were stratified by host genotypes and disease severity to identiy groups of patients that were benefitted the most under the administration of AIT.

This seminar will be in English

 

Calendar subscription link for the complete seminar series:
https://indico.math.cnrs.fr/category/711/events.ics

Program of the Biostatistics seminars:
https://indico.math.cnrs.fr/category/711/

Subscribe to the seminar mailing list:
https://diff.u-bordeaux.fr/sympa/subscribe/seminaire.biostat.bph

Former e-seminars on our YouTube channel (mostly in French):
https://www.youtube.com/channel/UCURp-hEQL7k23UzGfqgEurA/videos

Biostatistics seminar series from the Department of Public Health from the University of Bordeaux and the Bordeaux Population Health UMR 1219 research center