Automated object and image level classification of TB images using support vector neural network classifier

In this work, digital Tuberculosis (TB) images have been considered for object and image level classification using Multi Layer Perceptron (MLP) neural network activated by Support Vector Machine (SVM) learning algorithm. The sputum smear images are recorded under standard image acquisition protocol. The TB objects which include bacilli and outliers in the considered images are segmented using active contour method. The boundary of the segmented objects is described by fifteen Fourier Descriptors (FDs). The prominent FDs are selected using fuzzy entropy measures. These selected FDs of the TB objects are fed as input to the SVM learning algorithm of the MLP Neural Network (SVNN) and the result is compared with the state-of-the-art approach, Back Propagation Neural Network (BPNN). Results show that the segmentation method identifies the bacilli which retain their shape in-spite of artifacts present in the images. The methodology adopted has significantly enhanced the SVNN accuracy to 91.3% for object and 92.5% for image level classification than BPNN.