Classification of features shape of Gram-negative bacterial using an extreme learning machine

Abstract

Gram-negative bacteria are one of microorganism responsible for nosocomial infections in Indonesia. Nosocomial bacteria can cause nosocomial diseases, which are difficult to cure with antibiotic treatment. This bacterial observation was carried out using image processing to replace visual inspection. The process of this research consists of four stages, namely Pre-processing preparing image data, dividing objects by segmentation, obtaining and selecting features, ending with classification. At the segmentation stage, the bacterial image object was chosen that best suits the expert representation, in this case, a medical analyst. Feature extraction is done to get the pixel object information to be processed. At the classification stage, the use of extreme learning machines was chosen due to its shorter training process time. Two different bacteria were used, namely nosocomial bacteria and Gram-negative bacteria. In this research, selected bacterial Klebsiella pneumonia was obtained from 50 patients until a total of 2520 images were obtained. At the classification stage, the results of the bacterial object feature extraction are used 120 photographs in the training process and 40 image data in the testing process, with the total amount of data used, is 160 images with 512x512 pixel size and 24-bit depth. Extreme learning accuracy results are obtained in 96.71% of the process testing.