The importance of a diagnostic test that is simple and quick to identify Mycobacterium tuberculosis infection needs no emphasis. The tuberculin skin test (TST – 1 TU RT23) is the diagnostic tool for identifying M. tuberculosis infection at present. The test reaction on the skin is measured after 48–72 h. It is observed that often multi-modes are seen, when the reactions are drawn as a graph and the bimodality is seen very feebly. Because of the difficulties in the administration of TST, several serological tests were developed over three decades, but none of the studies showed the desired results. One study to evaluate purified protein derivative (PPD) antigen resulted in a claim of 80% sensitivity and 4% false-positivity rate (14), while other researchers were not able to obtain similar results. In addition, several problems were encountered due to the non-availability of antigens, and data analyses from an ELISA-based diagnostic test showed considerable overlap of distributions of optical density (OD) values among patients and healthy individuals (10). Classical statistical techniques cannot explain the cause of these overlaps. Hence, an attempt is made in this article to resolve these difficulties by the pattern recognition technique (PRT). The technique lies in splitting the data into clusters using a supervised algorithm. The data set is normally split into a training set, a test set and a validation set. The PRT gets “trained” through the training data set until the infected and uninfected groups of individuals are correctly classified. The training occurs based on an algorithm on the training set. On successful completion of the training, this technique is further tested and validated in the respective data sets.

Setting: A total of 273 finger-prick specimens were collected from five categories (Al, A2, B, C, D, E) of subjects not vaccinated with (bacille Calmette Guerin) (BCG) from Trivellore BCG Trial area adopted by the Tuberculosis Research Center, Chennai, India.

Objective: The study was conducted with the primary aim of evaluating purified antigens – r38kDa, PPD and 30kDa – for their usefulness as diagnostic tools and to test the applicability of the PRT in the evaluation of diagnostic tests. Individuals in two main categories (definitely not infected categories Al, A2 and D, and definitely infected categories B, E and C based on reaction to TST) were assembled for the purpose.

Results: The overall PRT performance of 30kDa was 72.3% sensitivity and 90.9% specificity for identifying M. tuberculosis infection, while the r38kDa antigen recorded a sensitivity of 73.8% and a specificity of 84.6%. In the case of PPD, the results were not promising.

Conclusion: This paper on ELISA-based diagnostic tests attempts to implement an optimal decision support system through PRT that would identify the outcome (as infected or non-infected) based on the OD values. The PRT was able to predict the outcome for individual suspects. Further, Kullback–Leibler distance measurement has validated the PRT in distinguishing infected individuals from healthy subjects (based on the OD values).