Tuberculosis (TB) is an infectious disease which is caused by a bacterium called Mycobacterium tuberculosis and it affects human lungs. The number of TB related deaths worldwide in 2021 was 1.6 million. TB is the second most lethal infectious disease after COVID-19 and the 13th major cause of death globally. World-wide cases of tuberculosis (TB) are predicted to reach 10.6 million in 2021. The interesting fact is that the bacterium which causes TB continues to exist in human body without exhibiting any symptoms. So, it is necessary to predict the occurrence of TB at an early stage. This prediction is possible by performing classification using deep learning model. For classification, features from Chest X ray images need to be extracted. More number of features sometimes leads to less accuracy and this issue can be rectified using feature selection strategy. Genetic Algorithm (GA) and Dragonfly are the evolutionary algorithms which finds its application in feature selection and has the capability to provide promising results. In this research, a novel neuro fuzzy based system to classify tuberculosis namely OMDNFIS (Optimized Modified Dragonfly Neuro Fuzzy Inference System) with DenseNet121 for better classification accuracy. Experimental results are compared with the individual as well as few existing hybrid algorithms and it is proved that proposed OMDNFIS model outperforms the individual deep learning algorithms and few existing deep learning models.