A Literature Review on Techniques for Detection of Lung Diseases

Abstract

Computer aided diagnosis (CAD) is one of the potential technologies in today’s medical world that assist doctors to interpret and evaluate medical images in a short time. CAD offers support to medical professionals to make decisions on possible diseases. Various systems and approaches are implemented to serve this technology, and many hospitals have deployed the system for diagnosis of diseases. The detection of the proportion of disease would aid in determining if more in-depth tests are required for confirmation of the condition, hence avoiding risky biopsies. This survey focuses on such methodologies implemented by authors of several works to detect lung infections by analyzing tissue patterns and inflammations in lungs and classifying the same. Along with this, work related to the patient database system is also reviewed and a comparison is made between the works.

Introduction

I. INTRODUCTION

Lungs are the primary organs for respiration in mammals and most other vertebrates. Lungs perform gas exchange, i.e, they absorb oxygen from the air and transport it into the bloodstream, as well as release carbon dioxide from the bloodstream into the atmosphere[1]. Lung diseases prevent the lungs from functioning properly, reducing the oxygen capturing ability. Common causes for lung diseases would be bacterial, viral or fungal infections. Genetic defects and exposure to dangerous, poisonous compounds are also factors in some instances. Few lung diseases would spread to other people by sneezing and coughing. Diagnosis of the diseases at the right time helps the patient recover quickly and avoids spreading of infection inside the body and to the people around. Computer aided diagnosis is now one of the most powerful and accurate tools to analyze Chest X-rays, CT Scans or other medical images to detect lung disease. CAD provides additional support to the medical practitioners to take decisions on the confirmation of the infection. This survey on literature works explores some of the methods used to identify the tissue patterns and classification, detecting lung disease and also a work which proposes the patient database system to acquire and manage relevant information from patients for better decisions.

One of the powerful and accurate approaches for CAD is Convolutional Neural Network (CNN). It is a class of deep-learning artificial neural network, developed to recognize and analyze patterns from pixel images. It is playing a vital role in providing accurate results, serving the doctors and healthcare staff to a greater extent. Several other approaches are also serving for CAD, which are explored and compared in this survey.

II. LITERATURE SURVEY

A. Namrata Bondfale, D.S. Bhagwat, “Convolutional Neural Network for categorization of Lung Tissue Patterns in Interstitial Lung Diseases”, (ICICCT 2018)

The goal of this work is to classify the tissue patterns of Interstitial Lung Diseases(IDLs) as healthy, reticulate, honeycombing, etc by developing a framework using Convolutional Neural Network (CNN). MATLAB software is used for the detection of tissue patterns. Four steps are performed in the Training Stage. Pre-processing began with downsizing the large, high-resolution input pictures from CT scans or chest X-rays to a predetermined size. Second, for easier analysis, the image was segmented by separating it into numerous segments. Thresholding, Inversion, and Masking are done as a part of Segmentation to get the Region of Interest (ROI). Then, Feature Extraction is done on the segmented information to identify the tissue patterns of ILDs. Finally, the CT scans are stored in a dedicated database. At the testing stage, preprocessing is carried out, followed by segmentation and feature extraction, and then CNN is applied. Based on the extracted information and CNN output, tissue patterns of ILDs are classified. The accuracy is calculated using the HMM technique, which is described as (TP+TN)(P+N), where TP, TN, P, and N stand for True Positive, True Negative, Positive Sample, and Negative Sample, respectively. This system is said to be 82% accurate. In the future, 3D images of 3CT scans can be used, integrating the system with CAD for better diagnosis.

B. Quang H. Nguyen, Binh P. Nguyen, et al., “Deep Learning Models for Tuberculosis Detection from Chest X-ray Images”, (ICT 2019)

The use of transfer learning on medical imaging to identify tuberculosis is studied in this work. ImageNet weights were updated with a better method for transfer learning, which was previously insufficient and ineffective. By training the models in a multiclass multilabel setting, a new technique for capturing low-level features has been developed. In comparison to training from an initial random setting, an efficient technique for training in a data constrained situation and tuberculosis classification has been provided. The detection is based on binary classification, which determines if the X-ray has tuberculosis or not. Two sets of publicly available datasets of Chest X-rays have been used for this work. Training stage uses Shenzhen dataset from Shenzhen People’s Hospital and the testing stage uses Montgomery dataset, and different architectures are used for experimentation. The datasets are augmented for training to avoid distortions in the images. Rotations, horizontal flipping and transformations are performed, along with rescaling, before feeding the image into the neural network. Tests on several architectures were conducted and it was observed that performance of Inception ResNetV2 and DenseNet were equally good on the test dataset. DenseNet model was finalized as it performed well for three times lesser parameters than ResNet. Class Activation Mapping(CAM) is done to visualize the image regions having the highest resemblance of the disease. In future, inputs from medical professionals which include patient history, lab reports etc. can also be considered along with the X-rays to improve the final decision.

C. Enes Ayan, Halil Murat Ünver, “Diagnosis of Pneumonia from Chest X-Ray Images using Deep Learning”,  (EBBT-2019)

This paper examines the diagnosis of Pneumonia, that can only be determined by a professional radiologist using a chest X-ray. The disease's appearance in chest X-ray images might be obscure for a variety of reasons, and it can be mistaken for other disorders. Thus, computer-aided diagnosis systems are needed to guide clinicians. The authors use two renowned convolutional neural network models, Vgg16 and Xception, to identify and diagnose Pneumonia.In addition, in the training stage transfer learning and fine-tuning are used. When the results of two networks are compared on data with varied measures, the Xception model outperforms the Vgg16 model in diagnosing pneumonia (pneumonia precision 94%,91% respectively). The Vgg16 model has shown better execution over the Xception model (with accuracy  87%, 82% respectively) in diagnosing normal cases. A combination of these neural network models, Xception and Vgg16 or a hybrid model will work best in diagnosing pneumonia from chest X-ray pictures.

D. Ali Serener, Sertan Serte, "Deep learning to distinguish COVID-19 from other lung infections, pleural diseases, and lung tumors", Medical Technologies Congress (2020)

The research and the groundwork of this system is focused on detecting COVID-19 and eliminating the possibilities of all other lung infections. The system works on image processing of the X-ray and segmenting the processed image to six different neural networks such as  AlexNet, MobileNet-v2, DenseNet-121, VGG and ResNet-50. The models are trained using the Caffe deep learning framework and the ImageNet database of chest X-rays. Based on the results from all the different neural networks, the condition of the patient is distinguished from other pleural infections.Chest radiographs are utilised to develop deep learning models to differentiate COVID-19 from other pleural illnesses, infections, and lung mass. Every 20 chest radiograph image was trained to discriminate COVID-19 from each of the three diseases using four distinct types of chest radiographs of the same or comparable infection. In differentiating COVID-19 from pleural effusion, MobileNet-v2 and AlexNet infrastructures have been deployed. ResNet-18 and AlexNet architectures are used to distinguish Covid-19 from Pneumonia. DenseNet-121 is used for detecting lung mass anomalies. The results from these have been tabulated and deployed to make one specific model analysis to suit all the kinds of detection. ResNet-18 architecture is then considered the most accurate and efficient in separating COVID-19 from pleural effusion,lung mass and pneumonia. The system would further focus on elaborating the detection of other ailments apart from COVID-19. Manifestation of the ailment could also be introduced to the system along with the detection.

E. LiuLiu Fu, Ling Li, “A Smart Decision Making System for Managing Patient Database”, 2016

The suggested system gives healthcare professionals useful recommendations and assists them in gathering more clear feedback about patients in order to improve appointments, to narrow the gap between patient needs and offered or delivered services. Since the study concentrated on a decision-making module for smart scheduling. For the storage of data (MySQL) is used as a local database. There are two phases to it. The first phase is a new global scheme that entails developing a new overall classification system. The method will be based on the physicians of the personal preferences. The second phase is to analyze and merge the record of the updated patient. Automatic classification utilizes the support vector machine technique. The framework consists of a Web-based point of interaction that empowers individuals to make and change classes like making due files. Furthermore, the local database may be deployed in the cloud since cloud databases offer affordability, security, accessibility, collaboration, and sharing.

III. COMPARISON TABLE

IV. DECLARATION

We, the project team hereby declare that the details enclosed in the manuscript is true and correct to the best of our knowledge and belief. In terms of Competing interests and Conflict of Interests, we have ensured we have not produced any work as our own but we have used the research to draw our own comparison analysis. Our work does not need any assistance with respect to funding as all the tools and resources used are through open source, open access and institutional access. We have received the ethical approval from our institution regarding the content of the paper and we are willing to give our approval for the publishing of the paper. We are glad to give our consent to participate in the journal and publish our manual in the same.

Supreeth S - supreethjs2000@gmail.com

Moiz Ahmed Khan - moizahmed713@gmail.com

Sri Krishnan K L - srikrishnan1947@gmail.com

Conclusion

From the survey of existing methodologies and the techniques used to detect lung infections, it is clearly observable that the models have to be trained with a lot of sample X-ray images. The techniques have been effectively proven to be accurate, quick and deployable. The detection of ailments without visible symptoms has been made easier. The reach of the systems to the medical practitioners has led to further research paths on deep learning models. Categorizing the ailments has led to simpler diagnosis and quicker treatment procedures. Accuracy rate of the systems is reliable to judge the ailment and clear the suspicion, testing of various other infections. Image processing, though a part of the system, is not used intensively to capture the anomaly. There is still a paved path for the improvement of the detection of specific ailments by different deep learning models. The distinction of different diseases can be improved upon. And the rate of manifestation of the ailments can be further looked into. Models can be merged and hybrid models can be trained for better performance and results. Apart from X-rays, other scanning reports can be used to detect the ailments.

References

REFERENCES [1] Namrata Bondfale, D.S.Bhagwat, “Convolutional Neural Network for categorization of Lung Tissue Patterns in Interstitial Lung Diseases”, Second International Conference on Inventive Communication and Computational Technologies (ICICCT 2018) [2] Quang H. Nguyen , Binh P. Nguyen , Son D. Dao, “Deep Learning Models for Tuberculosis Detection from Chest X-ray Images”, 2019 26th International Conference on Telecommunication (ICT 2019) [3] Enes Ayan, Halil Murat Ünver, “Diagnosis of Pneumonia from Chest X-Ray Images using Deep Learning”, Scientific Meeting on Electrical-Electronics & Biomedical Engineering and Computer Science (EBBT-2019) [4] Ali Serener, Sertan Serte,“Deep learning to distinguish COVID-19 from other lung infections, pleural diseases, and lung tumors”, Medical Technologies Congress (2020) [5] LiuLiu Fu, Ling Li, “A Smart Decision Making System for Managing Patient Database”, 4th International Conference on Enterprise Systems (2016) [6] Marios Anthimopoulos, Stergios Christodoulidis, Lukas Ebner, Andreas Christe, Stavroula Mougiakakou, “Lung Pattern Classification for Interstitial Lung Diseases Using a Deep Convolutional Neural Network”, IEEE Transactions on Medical Imaging (2016) [7] Araya Chatchaiwatkul, Pasuk Phonsuphee, “Lung Disease Detection and Classification using Deep Learning Approach”, 2021 36th International Technical Conference on Circuits/Systems, Computers and Communications (ITC-CSCC 2021) [8] R. Uppaluri et al., “Computer recognition of regional lung disease patterns,” Am. J. Respir. Crit. Care Med., vol. 160, no. 2 [9] L. Brunese, F. Mercaldo, A. Reginelli, and A. Santone, “Explainable deep learning for pulmonary disease and coronavirus covid-19 detection from x-rays,” Computer Methods and Programs in Biomedicine, p. 105608, 2020. [10] Qing Li et al., “Medical image classification with convolutional neural network,” in Proc. 13th Int Conf. Control Automat. Robot. Vis., Dec. 2014 [11] M. Gao et al., “Holistic classification of CT attenuation patterns for interstitial lung diseases via deep convolutional neural networks,” in 1st Workshop Deep Learn. Med. Image Anal., 2015 [12] K. He, X. Zhang, S. Ren, and J. Sun, \"Deep residual learning for image recognition,\" in Proceedings of the IEEE conference on computer vision and pattern recognition, 2016

Copyright

Copyright © 2022 Supreeth S, Moiz Ahmed Khan, Sri Krishnan K L, Prof. Chetan Umadi, Prof. Dr. Smitha Sasi. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.