© 2021 IJRTI | Volume 6, Issue 11 | ISSN: 2456-3315 IJRTI2111005 International Journal for Research Trends and Innovation (www.ijrti.org) 22 A Survey: Identification of Different Thoracic Disease Using Convolutional Neural Network Richa Tiwari 1 , Monika Verma 2 , Sumit Kumar Sar 3 *1 PG Student, 2 Assistant Professor, 3 Assistant Professor Dept. of Computer Science &Engineering, Bhilai Institute of Technology, Durg, India, Abstract: Thoracic disease affects various parts of the organ around the chest at different paces, and accounts for the number of deaths in India. We have vaccines and medicines to prevent the spread of infection-causing bacteria, viruses, and fungi in the organs, but many patients still die as a result of the inability in early detection of the disease. The diagnosis of thoracic disease relies on chest x-ray images that are manually interpreted by an expert. Chest x-ray images have their own set of flaws, which can lead to inaccuracy in judging infected areas or even the presence of infections. This paper focuses on using computer-assisted techniques, different algorithms present in machine learning and deep learning pre-trained CNN models (ResNet, DenseNet, CheXNet and VGG), and classification techniques(Logistic Regression, SVM, K-nn) in the medical and healthcare field for the classification of chest x-ray and diagnosis of thoracic disease. Keywords: CNN, Deep Learning, Thoracic Disease, X-ray image, Automation, Machine Learning, DenseNet, K-nn, SVM, ResNet. 1. INTRODUCTION Abnormalities in the human body are life threatening if not treated properly. There are several health problems that are quickly diagnosed and treated when they are discovered early on, while others go unnoticed until they become alarming. The majority of them were caused by unintended medical malpractice, ineffective care, and flaws in technology used to diagnose the disorders. As a result, a modern and creative approach to technology is critical for eliminating the flaws that endanger lives. It’s healthy to have more than one way to serve people having severe disorders. 1.1. Thoracic Disease Thorax is the region between the inferior abdomen and the root of the superior neck. Thoracic Disease is a condition that affects the region around the chest. There are various kinds of thoracic infections that affect lungs, heart, esophagus, great vessel, diaphragm, mediastinum and chest wall. Its initial diagnosis is mostly carried out with the help of X-ray machines which generate x-ray images scanning through the body. As we know, unless the infection, cause of infection or the health condition is detected in its early phase, it is quite impossible to continue the exactly required treatment. 1.2. Image Pre-processing The chest X-ray has been useful for a long time, but it is far from sufficient. The high beam of x- ray light passes through the body and shows the insides based on the density of organs. The x-ray images however have noise that causes difficulties in extraction of actual data. Thus in our model, we are implementing an Image Pre-processing technique where images are evaluated using complex algorithms and it will contribute in extracting only necessary attributes. To combat the disadvantages of diagnosis through X-ray images, we propose incorporating computer-assisted techniques into the field along with different classifiers (Logistic Regression, SVM and K-nn), and algorithms present in machine learning and deep learning pre-trained CNN models (DenseNet, ResNet, VGG and CheXNet), for the detection of thoracic disease. The use of machine learning, artificial intelligence and various algorithms has proven beneficial even in the past for handling the patient’s re cords, disease monitoring and more, which is why, in the study of the chest x-ray, it could result in wonders. 2. LITERARTURE SURVEY Deep Learning and Machine Learning approaches have gained popularity in the healthcare industry in recent years. Several chest x-ray datasets have been released, and various methods for disease classification, diagnosis, and localization in chest x-ray images have been applied. Datasets that have been used for detection of pneumonia and different thoracic diseases included JSRT datasets containing x-ray images of lung nodules; X-rays images from a radiologist and health clinical center and Chest x-ray 14 dataset, released by Wang et al.[4], which is by far the largest dataset of different thoracic disease containing 112,120 frontal view x-ray images. Aside from x-ray datasets, some researchers have used the cough sound of patients as an input for pneumonia detection. Recently, DimpyVarshni and Kartik Thakral, et al. (2019) [1] used various pre-trained CNN Models and classifiers to explore the NIH Chest X-ray dataset for detecting Pneumonia. They found DenseNet-169 as a feature extractor and Support Vector Machine (SVM) as a classifier with Radial Basis Function kernel (RBF kernel) to be the best models for detecting Pneumonia. On the same dataset, Benjamin Antin et al. [2] used the logistic regression approach to identify pneumonia in 2017. Similarly, Pranav Rajurkar et al. (2017) [3] used the same NIH Chest X-ray dataset for Pneumonia detection and named their model CheXNet. The model uses a 121-layer convolutional neural network to correctly detect all 14 common thorax diseases using chest x-ray images as data. Meanwhile, Wang, et al. (2017) [4] released a large dataset consisting of 108,948 frontal-view x-ray images from 32,717 unique patients. Kalyani Kadam, et al. (2019) [5] used data preprocessing techniques and trained the model with smaller image sizes before