SUPPORT VECTOR MACHINES FOR AUTOMATIC DETECTION OF TUBERCULOSIS BACTERIA IN CONFOCAL MICROSCOPY IMAGES

This paper presents an image segmentation method based on support vector machines classifiers at a pixel level. We apply this method to quantify the amount of Mycobacterium tuberculosis in confocal microscopy images for drug-discovery within the context of high content screening (HCS). To deal with the performance constraints of HCS, we propose a model-selection algorithm that finds the best classifier's hyperparameters by optimizing both classification rate and complexity. We validate our HCS adapted approach against commonly used readout techniques