Description

Lung diseases are the most common disease which causes mortality worldwide .In this study, the computed tomography images are used for the diagnosis of the lung diseases such as normal, small cell lung carcinoma, large cell lung carcinoma and non-small cell lung carcinoma by the effective extraction of the global features of the images and feature selection techniques. The images are recognized with the statistical and the shape based features. The texture based features are extracted by Gabor filtering, the feature outputs are combined by watershed segmentation and the fuzzy C means clustering. Feature selection techniques such as Information Gain, correlation based feature selection are employed with Genetic algorithm which is used as an optimal initialization of the clusters. The dataset of lung diseases for four classes are considered and the training and testing are done by the Naive Bayes and random forest classifier. Results of this work show an accuracy of above 80% for the correlation based feature selection method using naive Bayes classifier. In this study, texture based segmentation and recognition of the lung diseases from the computed tomography images are presented. The texture based features are extracted by Gabor filtering, feature selection techniques such as Information Gain, Principal Component Analysis, correlation based feature selection are employed with Genetic algorithm which is used as an optimal initialization of the clusters. The feature outputs are combined by watershed segmentation and the fuzzy C means clustering. The images are recognized with the statistical and the shape based features. The four classes of the dataset of lung diseases are considered and the training and testing are done by the Naive Bayes classifier to classify the datasets. Results of this work show an accuracy of above 90% for the correlation based feature selection method for the four classes of the dataset. The diagnostic results obtained are found to be very promising. As high as 83% accuracy in classification is achieve d using training data sets of reasonable size . Classification accuracy is improved as the numbers of training samples are increased. The present study also concludes that, back – propagation algorithm of ANN is a good choice for classification of cancer and TB images. This supervised training algorithm produce results faster than the other traditional classifier. Graphical representation of the classification on accuracy.