Automatic Identification and Reconstruction of Tuberculosis in Microscopic Images Using Convolutional Auto-Encoder Network

A Tuberculosis (TB) is an infectious disease caused by the Mycobacterium that can be prevented and treated. The TB automatic identification as an AI tool can help physicians to see the TB bacteria on the microscopic sputum smear images of patients, then, to choose type of the treatment, the amount of medication prescribed, and other treatment measures. In this paper, an automatic method is proposed to identify and reconstruct the TB bacteria on the microscopic images. First, the input images are resized and enhanced. Next, the Convolutional Auto-Encoder Neural Network (CAENN) is applied. The convolution part of the CAENN identifies the TB in the input images during the training phase, extracts the features of the TB, and optimizes the weights of the CAENN. The Auto-Encoder part of the CAENN reduces the dimensions of the feature vector and uses this vector to reconstruct the shape of the TB on the detected locations in each input image. The proposed method simulation is done using the Python software. The simulation results of the proposed method on 10,000 images of the database show the identification accuracy of 99.93%, which is the highest compared to the state-of-art methods.