Detection of abnormalities in ultrasound lung image using multi-level RVM classification

The classification of abnormalities in ultrasound images is the monitoring tool of fluid to air passage in the lung. In this study, the adaptive median filtering technique is employed for the preprocessing step. The preprocessed image is then extracted the features by the convoluted local tetra pattern, histogram of oriented gradient, Haralick feature extraction and the complete local binary pattern. The extracted features are selected by applying particle swarm optimization and differential evolution feature selection. In the final stage, classifiers namely relevance vector machine (RVM), and multi-level RVM are employed to perform classification of the lung diseases. The diseases respiratory distress syndrome (RDS), transient tachypnea of the new born, meconium aspiration syndrome, pneumothorax, bronchiolitis, pneumonia, and lung cancer are used for training and testing. The experimental analysis exhibits better accuracy, sensitivity, specificity, pixel count and fitness value than the other existing methods. The classification accuracy of above 90% is accomplished by multi-level RVM classifier. The system has been tested with a number of ultrasound lung images and has achieved satisfactory results in classifying the lung diseases.