faced by the community prior to model building. These challenges include orientation, data augmentation, colour space augmentation, and feature space augmentation of image but are pragmatically handle by various parameters and functions in deep learning models by maintaining their accuracies.
This chapter aims to analyse critically the deep learning techniques utilized for thoracic image analysis with respective accuracy achieved by them. The various deep learning techniques are described along with dataset, activation function and model used, number and types of layers used, learning rate, training time, epoch, performance metric, hardware used, and type of abnormality detected. Moreover, a comparative analysis of existing deep learning models based on accuracy, precision, and recall is also presented with emphasize on the future direction of research.
Keywords: Accuracy, classification, deep learning, localization, precision, recall, radiography, thoracic disorder
2.1 Introduction
Identification of pathologies from chest x-ray (CXR) images plays a crucial role in planning of threptic strategies for diagnosis of thoracic diseases because thoracic diseases create severe burden on the overall health of a person and ignorance to them may lead to the sudden death of patients. Tuberculosis (TB) continues to remain major health threat worldwide and people continue to fall sick due to TB every year. According to WHO report of 2019 [24], 1.4 million people died due to TB in 2019. When a person suffers with TB, it takes time for symptoms to express prominently. Till then, the infection transfers from person to person results in increase in prevalence of infection and mortality rate. Moreover, Pneumonia is also creating major burden in many countries. Since December 2019, people all over the world were suddenly suffered with Pneumonia with unknown etiology which results in COVID-19 Pandemic. TB, Pneumonia and other disorders are located in the thoracic region are known as thoracic disorders and location of Thoracic pathology is captured through x-ray of thoracic region. There are 14 different classes of thoracic pathologies such as Atelectasis, Pneumonia, Hernia, Edema, Emphysema, Cardiomegaly, Fibrosis, Pneumothorax, Consolidation, Pleural Thickening, Effusion, Infiltration, Nodules, and Mass. CXR is the most preferable screening tool used by radiologist for prediction of abnormalities in thorax region. To understand the exact location and type of pathology, expert radiologists are needed who have enough experience in the field of detecting pathologies from the chest radiology x-ray and are seldom available.
Furthermore, economically less developed countries lack with experienced radiologist and growing air pollution is badly impacting human lungs for which x-ray is the only reasonable tool to detect abnormalities in them. Recent pandemic of COVID-19 motivates researchers to design deep learning models which can predict the impact of SARS CoV-2 virus on human lungs at early stages as well as to identify post COVID effect on lungs. Patients suffering with thoracic abnormalities are associated with major change in their mental, social, and health-related quality of life. For some of them, early chest radiography can save their life and prevent them from undergoing unnecessary exposure to x-rays radiations. Thus, this motivates to explore new findings from it. Therefore, to make accurate prognosis of chest pathologies, researchers have automated the process of localization using Deep Leaning (DL) techniques. These models will act as an assistant to less experienced radiologist with the help of which prediction can be tallied. Due to significant contribution of deep learning in medical field in terms of object detection and classification, researchers have developed many models of DL for classification and localization of thoracic pathologies and compared their efficacy with other models on the basis of various evaluation parameters. This automated system of thoracic pathology detection helps radiologists in faster asses to patients and their prioritization as per the severity.
The overall outline of the chapter is presented as follows: Section 2.2 consists of broad view of existing research employed by various researchers, challenges in detecting thoracic pathology, datasets used by researchers, thoracic pathologies, and parameters used for comparison of models. Section 2.3 details the models implemented by researchers using deep learning and demonstrates the comparison of models on various parameters. Section 2.4 present the conclusion and future directions.
2.2 Broad Overview of Research
Though ample number of models has been implemented for localization and classification of thoracic pathologies, the broad view of research carried out till date (i.e., December 2020) is represented using Figure 2.1.
It is observed that researchers have either designed their own ensemble models using existing pretrained models or use pre-trained models like GoogleNet, AlexNet, ResNet, and DenseNet for localization of yCXR. In ensemble models, either weights of neural networks are trained on widely available ImageNet dataset or averaging outputs of existing models were used for building new models. Input to these models could be either radiography CXR images or radiologist report in text form. One of the most regular and cost-effective medical imaging tests available is chest x-ray exams. Chest x-ray clinical diagnosis is however more complex and difficult than chest CT imaging diagnosis. The lack of publicly accessible databases with annotations makes clinically valid computer-aided diagnosis of chest pathologies more difficult but not impossible. Therefore, annotation of CXR images is done manually by researchers and is made available publicly to test the accuracy of novel models. These datasets include ChestX-ray14, ChestX-ray8, Indiana, JSRT, and Shenzhen dataset. The final output of existing models is identifying one or more than one pathology out of fourteen located in either x-ray image or radiologist report. DL models devised by researchers are compared on the basis of various metric such as Receiver Operating Characteristic (ROC), Area Under Curve (AUC), and F1-score, which are listed in Figure 2.1. While designing deep learning models for detection of thoracic pathologies various challenges are faced by researchers which are discussed in next section.
2.2.1 Challenges
There are various challenges while implementing deep learning models for analysis of thoracic images and these challenges are listed below. They could be in terms of availability of dataset and nature of images.
Figure 2.1 Broad view of existing research.
1 Non-availability of large number of labeled dataset [16].
2 Due to use of weights of neural nets trained on ImageNet dataset for ensemble model, overfitting problem occurs. Theses pre-trained models are computationally intensive and are less likely to generalize for comparatively smaller dataset [7].
3 Presence of lesions at different location in x-ray and varying proportion of images of each pathology might hinders performance of models [41].
4 X-ray images contain large amount of noise (healthy region) surrounding to lesion area which is very small. These large numbers of safe regions make it impossible for deep networks to concentrate on the lesion region of chest pathology, and the location of disease regions is often unpredictable. This problem is somewhat different from the classification of generic images [67], where the object of interest is normally located in the center of the image.
5 Due to the broad inter-class similarity of chest x-ray images, it is difficult for deep networks to capture
