Research Projects

Tubercular uveitis (TBU)—an eye infection caused by Mycobacterium tuberculosis leads to vision loss and multiple ocular problems. The significant challenges with ocular TB are that it affects different eye tissues, leading to diverse manifestations of symptoms such as anterior, intermediate, or posterior uveitis compared to specific signs. Globally, there is a considerable variation in TBU prevalence that ranges from 0.2% to 10.5%. The prevalence is significantly higher in TB-endemic regions like India, from 5.6% to 10.5%. The lack of specific signs and symptoms, diagnostic disagreements and uniformity in treatment protocol led to the formation of the Collaborative Ocular Tuberculosis Study (COTS) group—A participation of 25 multi-national centers. The COTS group made significant strides in understanding TBU. The next step is to develop diagnostic (probability that a specific condition is present) and predictive (a particular event will occur in the future) models for TBU to help clinicians make evidence-based informed decisions. A model-to-data approach is desirable for rare diseases such as tubercular uveitis in particular and multi-centric studies in general due to challenges of data ownership, legal, administrative, security and privacy issues. we will, therefore, develop and evaluate the models using frequentists and the Bayesian neural network approach—to understand and address issues in longitudinal, small sample size and heterogeneous data. Data ownership, patient privacy and security are perplexing issues for modelling sensitive medical data; therefore, we will evaluate the model-to-data approach—federated machine learning against frequently used data-to-model approach—centralised machine learning.

Tubercular uveitis, Federated Machine Learning, Bayesian analysis, Multivariable modelling, Recurrent neural network, LSTM network