SUPERVISED LEARNING AT CHEST RADIOGRAPHY USING ARTIFICIAL NEURAL NETWORK
chest radiograph is a low-cost medical screening technique widely used to screen interstitial lung diseases. Since the images produced are 2D, it requires a highly experienced and qualified radiologist to review and detect the disease correctly. Also, X-Rays are more prone to noise and therefore it is arduous to see the findings with the naked eye. Lack of qualified radiologists paves way for CAD techniques to interpret the X-Rays. The approach presented in this paper employs a range of image processing techniques along with supervised learning at CXRs, to screen the diseases and classify them into normal and abnormal. K-means clustering is used to split the lung region and markers alike as mean, variance, entropy, kurtosis, and skewness are using the local data patterning statistics to derive. The features are validated using t-test and significant features are used to train the classifier. ANN is utlized for classification as it produced better classification results with 85% accuracy, 80% specificity and 90% sensitivity.
Canny Edge Detector, Structured Edge Detection (SED), Watershed Algorithm, K-means clustering, Local Binary Pattern (LBP), Artificial Neural Network (ANN)