Lung Disease Classification and Identification using Deep Learning Techniques

Authors: simran shivhare, dr.manish Dixit

Certificate: View Certificate

Abstract

Machine learning is a subfield of artificial intelligence that uses a wide range of statistical, probabilistic, and optimization techniques. These methods allow computers to \"learn\" from previous examples and find patterns that are hard to spot in large, noisy, or complicated data sets. Learning by machine is a key strategy for making complex algorithms that can work on their own and are completely objective. This makes it perfect for analysing high-dimensional and multimodal biomedical data. Machine learning is an important part of the modern medical systems we have today. Many lives can be saved and stress on the system can be eased if diseases are found earlier and more accurately. Lung problems are one of the most common reasons why people die. During this investigation, we will suggest and evaluate the convolution neural network that was made to classify ILD patterns. ILD patterns can have one of seven different results: healthy, ground glass opacity (GGO), honeycombing, consolidation, reticulation, or micro nodules. Another choice is to use both GGO and reticulation. For training and testing the CNN, we first used a deep CNN that was made for the problem at hand. Lastly, we put into groups how well CNNs did at analysing lung patterns, which showed that they had a 91% success rate. We suggested that a deep convolutional neural network (CNN) be used to divide lung CT image patches into seven different groups, including six different ILD patterns and healthy tissue. The method is easy to train on different types of lung patterns, and its effectiveness could be improved even more. Randomizing the weights at the beginning makes the results for the same input slightly different. This is because the weights are not always the same. This means that the number of normal and non-pathological images in the training data is skewed. This is another important issue in medical image analysis. Rare diseases are an extreme example of this because they can be missed if there aren\'t enough training examples. Even though there is a risk of overfitting, the effect of this data imbalance can be lessened by using data augmentation to create more training images of unusual or atypical data. In addition to looking at data-level approaches, algorithmic modification strategies and costsensitive learning have also been looked at.

Introduction

The lungs, which are very important, make it possible for the body to expand and relax in order to take in oxygen and get rid of carbon dioxide. This is how the body takes in oxygen and gets rid of carbon dioxide. Lung diseases are a type of respiratory disease that affect the many organs and tissues that help us breathe. This can lead to a number of problems, such as problems with the airways, lung tissue, and circulation in the lungs. Some respiratory diseases, like the common cold and flu, only cause mild pain and trouble breathing, while others, like pneumonia, tuberculosis, and lung cancer, can be fatal and cause severe acute breathing problems [1]. According to the results of a study called "Te Global Impact of Respiratory Disease," which was done by the Forum of International Respiratory Societies, 10.4 million people had mild or severe symptoms of tuberculosis, and 1.4 million of those who had the disease died [2]. The Forum of International Respiratory Societies was in charge of the study. Lung cancer kills an astonishing number of people every year. During the time the poll was done, it was said that more than 1.6 million people had died. The "Pneumonia and Diarrhea Progress Report 2020" from the Johns Hopkins Bloomberg School of Public Health says that pneumonia is one of the most common respiratory infections and that it killed 1.23 million children under the age of 5 around the world in [3]. If any of the above diseases are found in their early stages, not only will the chance of survival be much higher, but it may also be possible to keep people from dying. X-rays of the chest and CT scans are two common types of tests used to find out if these diseases are present [4]. For the scanned photos to be looked at and the infections to be found, qualified professionals must be present. Official numbers from the Ministry of Health

Conclusion

We proposed a deep CNN to classify lung CT image patches into7classes, including 6 different ILD pattern sand healthy tissue. The method can be easily trained on additional textural lung patterns while performance could be further improved. The slight fluctuating of the results, for the same input, due to the random initialization of the weights. Data or class imbalance in the training set is also a significant issue in medical image analysis. this refers to the number of images in the training data being skewed towards normal and non-pathological images. Rare diseases are an extreme example of this and can be missed without adequate training examples. This data imbalance effect can be ameliorated by using data augmentation to generate more training images of rare or abnormal data, though there is risk of over fitting. Aside from data-level strategies, algorithmic modification strateg