Detection & Classification of Tuberculosis HIV- Positive Patients using Deep Learning Princy K T M Department of Computer Science and Engineering Amrita School of Engineering, Bengaluru Amrita Vishwa Vidyapeetham, India bl.en.p2dsc21022@bl.students.amrita.edu Tripty Singh Department of Computer Science and Engineering Amrita school of Engineering, Bengaluru, Amrita Vishwa Vidyapeetham, India Bangalore tripty_singh@blr.amrita.edu Vineet Vinayak Department of Computer Science and Engineering Amrita school of Engineering, Bengaluru, Amrita Vishwa Vidyapeetham, India bl.en.p2dsc21027@bl.students.amrita.edu Prakash Duraisamy Department of Computer Science University of South Alabama, South Alabama United States of America prakashduraisamy@southalabama.edu Abstract— Tuberculosis [TB] recently struck several countries all through the entire globe. According to the World Health Organization [WHO], there have been a calculated 2.8 million TB fatalities in 2022, with an extra 0.3 million deaths mainly due to TB illness in HIV-positive patients. The majority of Deaths can’t be avoided unless the disease is diagnosed early. Conventional diagnostic procedures, such as blood and urine tests or sputum testing, not only are inconvenient, and they also take a lot of time to analyze and cannot compare different drug-resistant phases of TB. In this paper, authors look at how deep learning-based method might be a good alternative to decision forest medical image- categorization systems. These experiments are conducted on chest X Ray of both tuberculosis and normal patient and identify the X-Rays of normal and tuberculosis patient separately. This experiment objective is to create generalized model to overcome the problems of the existing model. In this paper authors are experimenting various models such as normal CNN and Transfer Learning Methods. Keywords— Transfer learning, VGG16, tuberculosis, deep learning, CNN I. INTRODUCTION With just an expected 3.1 million fatalities in 2024, tuberculosis is the most dangerous infectious diseases in the world. Even though the lungs are the most commonly affected organ, they can also damage the intestines. Most commonly used method to detect tuberculosis is chest x ray scanning or scanning TB in the lungs [1]. But due to inexperience of physicians and other inefficient therapy most of the time it is misclassified. Machine learning and deep learning are widely used in medical image classification over the years. So applying deep learning model to classify the tuberculosis detection is great idea. Deep learning models like CNN and transfer learning are showing good result on other images classification for these years [2]. Vgg16, Vgg 19 & RESNET 50 are the commonly used transfer learning methods. These methods have already been trained on large dataset. So, training time and cost is very less while applying this method [4]. The primary objective of this study is to identify tuberculosis as early as possible. Using CXR pictures, this procedure will help in the rapid detection of TB. Avoiding incorrect outcomes can be achieved by constructing a model with high precision. Implementing such a test would lead to a more reliable system, allowing for the rapid evaluation of a larger number of people, effectively reducing the transmission of the virus. II. STATE OF ART Computer aided diagnosing is commonly used to diagnose any diseases related to lung. There has been some investigation into computerized tumor detection. Even though the identification of computed tomography with these imaging techniques may aid in the detection and analysis of tuberculosis, some other diseases symptoms also affecting lung in the same way. So it is very difficult to diagnose & hence physicians have a hard time figuring out the disease [5]. Many strategies have been used, including shape-based segmentation, decision trees, pixels, and so on. To conclude, the results are compared to other datasets. Evangelista and Guedes devised an intelligent pattern recognition- based computer-assisted approach [6]. Deep machine learning approaches can be utilized to evaluate tuberculosis by adjusting the parameters of deep convolutional neural networks (CNNs). [7]. Chikara et al. wanted to see if CXR images might be utilized to diagnose pneumonia. To verify the efficacy of training dataset models, they employed preprocessing procedures such as filtering and gamma adjustment [8]. Some research groups focused on single distinguishing characteristics of tuberculosis in the lungs [9]. While substantial research has been done on CXR datasets to diagnose tuberculosis, the disadvantage is that many of these algorithms rely on rule-based judgments that differ from person to person. Each model has a restricted number of parameters, thus several aspects that may be important contributors to its evaluation are left out [11]. Inside this field of computer vision, ConvNets are used and adds another dimension to medical data processing images. ConvNets were also used to successfully detect lung nodules. Several large archives, like as ImageNet, include terabytes of photos that are used to teach ConvNets. Then, using our training datasets, we may fine-tune them to improve accuracy [The idea of hidden layers in ConvNets’ topologies is another benefit]. These layers enable us to find hidden patterns in the photographs by working with a range of image densities and filters [13]. 2023 IEEE 8th International Conference for Convergence in Technology (I2CT) Pune, India. Apr 7-9, 2023 979-8-3503-3401-2/23/$31.00 ©2023 IEEE 1 2023 IEEE 8th International Conference for Convergence in Technology (I2CT) | 979-8-3503-3401-2/23/$31.00 ©2023 IEEE | DOI: 10.1109/I2CT57861.2023.10126469 Authorized licensed use limited to: Charles Darwin University. Downloaded on May 25,2023 at 00:38:43 UTC from IEEE Xplore. Restrictions apply.