Title: An empirical approach towards detection of tuberculosis using deep convolutional neural network

Authors: Syed Azeem Inam; Daniyal Iqbal; Hassan Hashim; Mansoor Ahmed Khuhro

Addresses: Department of Artificial Intelligence and Mathematical Sciences, Sindh Madressatul Islam University, Karachi, Pakistan ' Department of Computer Science, Shaheed Zulfiqar Ali Bhutto Institute of Science and Technology, Karachi, Pakistan ' Department of Artificial Intelligence and Mathematical Sciences, Sindh Madressatul Islam University, Karachi, Pakistan ' Department of Artificial Intelligence and Mathematical Sciences, Sindh Madressatul Islam University, Karachi, Pakistan

Abstract: Tuberculosis remains among the top disease, causing death all over the globe and its timely detection is a major concern for medical practitioners, especially after the emergence of the SARS-CoV-2 pandemic. Even with the recent advances in the methods for medical image classification, it is still challenging to diagnose tuberculosis without considering the associated historical and biological factors. There has been a great contribution of unsupervised learning in the development of techniques for image classification and the present study has utilised a deep convolutional neural network for detecting tuberculosis. It proposes a network comprising 54 layers having 59 connections. After computations, our proposed deep convolutional neural network attained an accuracy of 99.79%, 99.46%, and 99.5% for the classes of healthy, sick, and tuberculosis (TB) respectively for a public dataset, achieving higher accuracy as compared to other pre-trained network models.

Keywords: tuberculosis; image classification; deep convolutional neural network; DCNN; accuracy; F1 score.

DOI: 10.1504/IJDMMM.2024.136232

International Journal of Data Mining, Modelling and Management, 2024 Vol.16 No.1, pp.101 - 112

Received: 24 Oct 2022
Accepted: 06 Feb 2023
Published online: 22 Jan 2024 *