A hybrid feature-based segmentation and classification system for the computer aided self-diagnosis of otitis media

We propose a novel hybrid otitis media (OM) computer aided detection (CAD) system, designed to aid in the self-diagnosis of various forms of OM. OM is a prevalent disease in both children and adults. Our system is able to differentiate normal ear from acute otitis media (AOM), otitis media with effusion (OME) and the multi-categories of chronic otitis media including perforation, retraction, cholesteatoma, etc. We propose a modified double active contour segmentation method designed for use with otoscope images, and enabled to handle user acquired data. To describe the visual symptoms (e.g., red, bulging, effusion, perforation, retraction, etc.) of otitis media accurately, we extract color, geometric and texture features by grid color moment, Gabor filter, local binary pattern and histogram of oriented gradients. A powerful classification structure based on Adaboost is used to select the most useful features and build a strong classifier. Our system achieves classification accuracy as high as 88.06% and is suitable for real use. In addition, some interesting observations about OM otoscope images are also discussed.