IMAGE PROCESSING FOR LUNG DISEASES CLASSIFICATION

Meshesha, Kasanesh

dc.contributor.author	Meshesha, Kasanesh
dc.date.accessioned	2020-03-20T05:44:19Z
dc.date.available	2020-03-20T05:44:19Z
dc.date.issued	2020-03-20
dc.identifier.uri	http://hdl.handle.net/123456789/10766
dc.description.abstract	The diagnosis of tuberculosis and lung cancer is difficult as symptoms of both diseases are similar. A missed or wrong diagnosis of lung cancer or TB by clinician can lead to delays in diagnosis and treatment and hence progression of the disease. This indicates that lung cancer is often misdiagnosed as pulmonary tuberculosis, and vice versa in most cases. Its challenges and problems are big concerns in most developing countries, including Ethiopia. Research Issue: As the problem of lung disease is raising over time, unlike the Ethiopian context which stuck on a diagnostic radiologist, globally, the means and techniques for detecting same have also been increasing. However, approaches integrating two or more lung disease together are rare. Methods: In order to achieve the objective of the research, image processing based lung disease classification techniques using MATLAB proposed and defined. Accordingly, a digital image analysis technique based on morphological and Texture features was developed to classify the two lung diseases. Sample lung images taken from three hospitals and the internet, and on average 100 images taken from each; Normal Lung, Lung TB and Lung Cancer. Finding: Approaches of KNN, Naïve Bayes and Neural Network classifiers on each classification parameters of morphology, texture and the combination of the two are compared. To evaluate accuracy of the classifier, 70% of the data set used for training and the remaining 30% for testing. The classification system is supervised corresponding to the predefined classes of the lung image. It is found that the classification performance of KNN is better than Naïve Bayes and ANN classifier. It is also identified that the discrimination power of texture feature is better than morphology feature, but when two of the features are used together the classification accuracy is greater. Of all the classification approaches, the best classification performance is obtained using KNN (specificity of 90%, and Sensitivity of 86.67% for Lung TB and 83.33% for Lung Cancer). The accuracy obtained from this approach is 86.67%. Conclusion/Originality: The finding of this study revealed that the two major and ever deadly lung diseases can be classified more accurately from an x-ray image than a radiologist can do. This will pave the way in treating the two diseases before progression and saves the lives of many in developing countries like Ethiopia.	en_US
dc.language.iso	en	en_US
dc.subject	Information Technology	en_US
dc.title	IMAGE PROCESSING FOR LUNG DISEASES CLASSIFICATION	en_US
dc.type	Thesis	en_US