Volume 6, Issue 02, February 2020 (ISSN: 2394 – 6598)
20
©IJETIE 2020
Optimal method to Detect Pneumonia in Chest X-Rays using deep learning
Framework
Ajith R1, Aravindha Raja G1, Bahrul Ameen S1, Kuppan MR1, Reshma S2
1Department of IT, Sri Krishna College of Technology
2Assistant Professor, Department of IT, Sri Krishna College of Technology
ABSTRACT
Physicians often use chest X-rays to quickly and cheaply diagnose disease associated with the area.
However, it is much more difficult to make clinical diagnoses with chest X-rays than with other imaging modalities such as CT or MRI. With computer-aided diagnosis, physicians can make chest X-ray diagnoses more quickly and accurately. Pneumonia is often diagnosed with chest X-Rays and Kills around 50,000 people each year. With computer aided diagnosis of pneumonia specifically, physicians can more accurately and efficiently diagnose the disease. In this project, we hope to train a model using the dataset described below to help physicians in making diagnoses of pneumonia in chest X-Rays. Our problem is thus a binary classification where the inputs are chest X-ray images and the output is one of two classes: pneumonia or non-pneumonia.
KEYWORDS: Pneumonia, Chest X-Rays, Diagnosis and Binary Classification
1. INTRODUCTION
LUNG diseases are very often examined by using of screening methods. The better is the examination the higher are the chances to help patients in time and save their life’s. Lung cancer is one of the most dangerous diseases, therefore a rapid identification of nodules in the early stadium can reduce deaths. However in-depth analysis of diagnostic images requires large experience and skills. Nowadays due to the recent advancements of clinical and medial sciences, the clinicians dispose of many non-invasive techniques for oncological assessment and diagnosis that are generally preferred to potentially dangerous and more radical approaches (e.g. exploration surgery). The imaging
diagnostic tools for medicinal applications are essentially founded on echography, x-rays, magnetic resonance imaging (MRI), computer tomography scan (CTS), and positron emission tomography (PET). X-ray imagery is a paramount exam for a correct diagnosis of potentially cancerous lung nodules. Such a diagnostic method is fundamental for radiologists and clinicians in order to select a correct treatment process. The examination of x-ray images is very important for many lung disorders and diseases even in an early stage, ie. we can detect pre symptomatic changes in the structure of lung tissues. From X-ray images we can evaluate several pathological conditions, and various research present that even 90% of peripheral neoplasms and 65-70% of central neoplasms can be efficiently
21
©IJETIE 2020 diagnosed from X-ray images in the early stadium of
disease. In any case unfortunately the most vulnerable factor is human, since many variables can influence the radiologist ability to detect the illness.
The quality of the images can tamper with a clear visibility of all tissues as well as visualize the tissues as opaque or disformed. Sometimes several tissues can be placed in superimpositions with bones which can cover them. Unfortunately, the extreme variability of the matter and the subjective evolution of the illness in each patient can potentially tamper with the correctness of the diagnosis made by a radiologist. Therefore, to improve the accuracy of their diagnosis the medical specialists can benefit from intelligent methods of diagnostic support.
Moreover, automated approaches to the process of analysis bring many advantages for faster and accurate diagnosis. Intelligent classification systems, and particularly convolutional neural networks (CNNs) and other deep learning techniques can be used as an efficient aid for lung cancer diagnosis and the consequent determination of a therapeutic approach. Among the possible applications of CNNs based classification techniques we can include solid tumours diagnosis and discrimination between malignant cancers with respect to benignant nodules. Deep learning strategies can be partitioned into four classifications:
CNN-based methods, restricted Boltzmann machines (RBMs), autoencoders, and Neuro-Fuzzy methods.
2. RELATED WORKS
The convolutional neural network (CNN) is proven to be very effective in image recognition and classification tasks. The development of CNNs starts from, LeNet, AlexNet [6], ZFNet [7], VGG [8], Inception [9] [10], ResNet [11], Inception- ResNet [12], Xception [13], DenseNet [14], and NASNet [15]. There are many studies on the use of
deep CNNs to detect abnormalities in chest x-rays.
For instance, M. T. Islam et al., [16] use several CNNs to detect abnormalities in chest x-rays. There is also a study by X. Wang et al., on the use of CNNs to detect thoracic pathologies from chest x-ray images. Their study also provides a large dataset as is the case in this study [17]. Among current research, some studies on the application of Densely Connected Convolutional Networks (DenseNet) [14] to detect thoracic pathologies such as ChexNet [18] and the Attention Guided Convolutional Neural Network (AG-CNN) [19]. Both studies train the neural network on a very large chest x-ray image dataset. Lung cancer prediction with CNN faces the problem of small sample size. Indeed, CNN contains a large number of parameters for adjustment on a large image dataset. In practice, researchers often pre-train CNNs on ImageNet, a standard image dataset containing more than one million images.
The trained CNNs are then adjusted on a specific target image dataset. Unfortunately, the available lung cancer image dataset is too small for this transfer learning to be effective, even with a data augmentation trick.
3. PROPOSED FRAMEWORK
The chest X-ray is one of the most commonly accessible radiological examinations for screening and diagnosis of many lung diseases. A tremendous number of X-ray imaging studies accompanied by radiological reports are accumulated and stored in many modern hospitals' Picture Archiving and Communication Systems (PACS). Machine learning has been used in various fields of medical industry to derive patterns from data. We demonstrate the effectiveness of this iterative refinement framework via extensive experimental evaluations on the publicly available data. And propose a deep learning framework using algorithms to extract knowledge from image data.
22
©IJETIE 2020 Fig.1. Proposed Work
4. DATASET
The dataset is organized into 3 folders (train, test, val) and contains subfolders for each image category (Pneumonia/Normal). There are 5,863 X-Ray images and 2 categories (Pneumonia/Normal). Source: Mendeley. Chest X-ray images (anterior-posterior) were selected from retrospective cohorts of pediatric patients of one to five years old from Guangzhou Women and Children’s Medical Center, Guangzhou. Format: . JPEG. The diagnoses for the images were then graded by two expert physicians before being cleared for training the AI system.
Fig.2. Class Distribution
Data is highly imbalanced. We have almost with thrice pneumonia cases here as compared to the normal cases.
5. CONVERT IMAGES TO NUMPY ARRAYS
Often in machine learning, we want to work with images as NumPy arrays of pixel data. This can be achieved using the imread () function that loads the image an array of pixels directly. the imshow () function that will display an array of pixels as an image. first loads the image and then reports the data type of the array.
Fig 3. Normal vs Pneumonia
Fig.4. Normal vs Pneumonia
23
©IJETIE 2020 we're applying a technique called Data
Augmentation. We're changing the sizes of the images to 226 x 226 and we'll flip the images horizontally as well so that we can have more data(images) to train on. Data augmentation: to make slight variations to our data so that we have more data, without losing semantic meaning in our data. Data augmentation is a powerful technique which helps in almost every case for improving the robustness of a model.
augmentation can be much more helpful where the dataset is imbalanced.
6. DEEP LEARNING
A CNN structure uses a Feed-Forward Neural Network. CNN uses MLP, Multi-Layer Perceptron's, to do this convolutional process. Deep learning CNN techniques, became well known based on an outstanding, or winning performance, of Image Recognition. CNNs need a minimal amount of preprocessing. The convolutional layer uses multiple filters where each filter moves sequentially across the input data or image to make a 2- dimensional activation map based on each filter.
Accuracy is one metric for evaluating classification models. The fraction of predictions our model got right.
Fig.5. Basic Structure of CNN
6.1. RESULTS AND DISCUSSION
The simplicity of the proposed method is one of the main advantages. Moreover, it has obtained
very good results. In further research we want to apply this methodology to other pulmonary diseases, which have similar symptoms. From our early research we can conclude that this kind of approach can be easily adjusted to other requirements due to the use of fuzzy rules without any loss in precision
Table.1.1 Summary of Results
Fig.6. Accuracy of CNN 7.CONCLUSION
Data Analytics is the procedure of retrieve a pattern from large data set in connection with machine learning, data base, and statistics.
Machine learning can be a very good help in deciding the line of treatment to be followed by extracting knowledge from such suitable databases. Our project can assist in proper treatment for Automated methods to detect and classify human diseases from medical images.
CLASSIFIER
ACCURACY
LOSS Epoch
CNN model 1
87% 0.28 20
CNN model 2
93% 0.2 3
24
©IJETIE 2020 REFERENCES
[1]https://www.cdc.gov/nchs/fastats/pneumonia.ht m
[2] Pranav Rajpurkar, Jeremy Irvin, et al. CheXNet:
Radiologist-Level Pneumonia Detection on Chest
XRays with Deep Learning,
https://arxiv.org/pdf/1711. 05225.pdf 2017.
[3] Gao Huang, Zhuang Liu, Laurens van der Maaten. Densely Connected Convolutional Neural Networks https://arxiv.org/abs/1608.06993
[4] https://www.kaggle.com/nih-chest- xrays/datasets
[5] Wang, et al. ChestX-ray8: Hospital-scale Chest X-ray Database and Benchmarks on Weakly- Supervised Classification and Localization of
Common Thorax Diseases
http://openaccess.thecvf.com/content cvpr 2017/
papers/Wang ChestX-ray8 Hospital-Scale Chest CVPR 2017 paper.pdf
[6] Diederik P. Kingma, Jimmy Lei Ba. Adam: A Method for Stochastic Optimization https://arxiv.org/ pdf/1412.6980.pdf
[7] Laurens van der Maaten, Geoffrey Hinton Visualizing Data using t-SNE 2008 http://www.cs.toronto. edu/⇠hinton/absps/tsne.pdf.
[8] Y. Freund and R. E. Schapire. Experiments with a new boosting algorithm. In Proceedings of International
Conference on Machine Learning, volume 96, pages 148–156, 1996.
[9] D. H. Wolpert. Stacked generalization. Neural Networks, 5(2):241–259, 1992.
[10] A. C. Tan and D. Gilbert. Ensemble machine learning on gene expression data for cancer classification. Applied
Bioinformatics, 2003.
[11] S. B. Cho and H. H.Won. Machine learning in dna microarray analysis for cancer classification. In Asia-Pacific
Bioinformatics Conference, pages 189–198, 2003.
[12] P. Mamoshina, A. Vieira, E. Putin, and A.
Zhavoronkov. Applications of deep learning in biomedicine.
Molecular Pharmaceutics, 2016.
[13] H. Hijazi and C. Chan. A classification framework applied to cancer gene expression profiles. Journal of
Healthcare Engineering, 4(4):255–84, 2012.
[14] S. Anders and W. Huber. Differential expression analysis for sequence count data.
Genome Biology, 11(10):1– 12, 2010.
[15] C. M. Bishop. Pattern Recognition and Machine Learning, chapter 1.3. Springer Science+Business Media, 2006.
[16] L. Breiman. Random forest. Machine Learning, 45:5–32, 2001.
[17] http://scikit-learn.org/stable/.
[18] X. Glorot, A. Bordes, and Y. Bengio. Deep sparse rectifier neural networks. Journal of Machine Learning
Research, 15, 2010.
[19] The TCGA Database.
http://cancergenome.nih.gov/.
[20] Cancer Genome Atlas Research Network.
Nature, 513(7517):202–209, 2014.
[21] J. Wu, X. Zhao, Z. Lin, and Z. Shao. A system level analysis of gastric cancer across tumor stages with RNA-seq data. Molecular BioSystems, 11(7):1925–1932, 2015.
