A support vector machine approach for identification of pleural effusion.

In this research, we investigated the method which was based on a support vector machine (SVM) to identify pleural effusion on the thoracic image. SVM is a method of machine learning that works well when applied to data outside the training set. We formulated the detection of pleural effusion and applied SVM to develop the identification algorithm. We applied SVM to detect thoracic images whether they identified as pleural effusion or normal. The identification of pleural effusion on the thoracic image was conducted through some processes such as the determination of the region of interest (ROI), segmentation, morphology operation, measurement of the sharpness value and slope value, training as well as testing. Determining ROI was intended to focus the measurement on the left side of the chest. Segmentation was carried out to separate lungs object from the background. Morphology operation was carried out for cavities on the object as the segmentation result to obtain the entire object so that the measurement of the slope's lower part image could be done perfectly. The training was carried out on 100 thoracic images, 50 of them were identified with pleural effusion and the other 50 were normal. The objective was to find the hyperplane with the parameter input such as the sharpness value and slope value of the lungs on the thoracic image. We tested the method proposed based on doctors' diagnosis using 50 thoracic images, 25 of which were identified with pleural effusion and the other 25 were normal. From the result of the test, the accuracy of the method we proposed was 96%.