Convolutional Neural Network based Automated Detection of Mycobacterium Bacillus from Sputum Images

Tuberculosis is the maximum life-threatening ailment that is the result of the species of pathogenic bacteria named Mycobacterium tuberculosis. Nowadays, loss of life rates due to tuberculosis growing gradually. So, the early detection of the disease is the most vital component to be considered. Recent studies primarily based on the Ziehl-Neelsen (ZN) sputum smear microscopy image which contains the ailment statistics of patients who gave their sputum samples to the laboratory for the examination. The motive of this work is to categorize the sputum images based on bacillus present in it and to lessen the dependency of technicians and pathologists through automated detection. In this work, the pre-processing is performed on ZN based sputum images by using noise reduction and intensity modification. Segmentation is based totally on the Channel Area Thresholding (CAT) that is the multilevel-thresholding method. Further, features inclusive of Histogram of Oriented Gradients (HOG), Speed up Robust Feature (SURF) is extracted. Classification is performed using Convolutional Neural Network (CNN) classifier and it classifies the bacillus as mild, moderate and severe primarily based at the wide variety of pixels with 99.5 percentage accuracy, 94.7 percent sensitivity, and 99 percent specificity.