Introduction: Tuberculosis (TB) is an infectious disease caused by the bacterium Mycobacterium tuberculosis. It kills about 1.5 million people per year and about 95% of its victims are from Africa. A major problem of TB is the difficulty in its diagnosis due to the fact that in most cases, it is either asymptomatic or latent. This difficulty in diagnosis has motivated the development of various algorithms for TB diagnosis most of which have poor diagnosis power.
Aim: This research focused on systematic review and analysis of computational models for diagnosis of TB with the view to identifying their strengths and weaknesses. The overall target is to develop a standard and robust computational model with improved diagnostic power.
Method: Selection was from peer-reviewed articles on Google scholar assessing strictly computational TB diagnostic models. Search terms include: Diagnosis, Tuberculosis, Computational, Mathematics, Bayes, Soft computing, Fuzzy logic, Neural Network. Exclusions were made based on some criteria.
Results: Initial search returned 303 of which only 42 studies met the inclusion criteria. 19 were on neural network or neuro-fuzzy, 2 studies were on Expert System. 7 analysed fuzzy logic/hybrids and Bayesian/data mining appeared in 7 reports. 5 studies were on Genetic Algorithm and its hybridized forms while 2 papers were on other methods.
Conclusion: Results suggest that accuracy and speed need to be improved due to weaknesses in existing models. Hybridization of Genetic algorithm, Neuro-fuzzy and Bayesian techniques will most likely guarantee improved diagnosis, however, further quantitative analysis is required to confirm this.