Towards low cost automated smartphone- and cloud-based otitis media diagnosis. (January 2018)

• Record Type: Journal Article
• Authors: Myburgh, Hermanus C.; Jose, Stacy; Swanepoel, De Wet; Laurent, Claude
• Is Part Of: Biomedical signal processing and control. Volume 39(2018)
• Page Start: 34
• Page End: 52
• Publication Date: 2018-01
• Subjects: Otitis media -- Image processing -- Feature extraction -- Decision tree -- Neural network -- Otoscope -- Tympanic membrane; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2017.07.015
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A smartphone- and cloud-based system for automated otitis media diagnosis is proposed. A image analysis and a neural network are used to perform diagnosis. The system diagnoses five pathologies with high accuracy (>86%). Abstract: Otitis media is one of the most common childhood illnesses. Access to ear specialists and specialist equipment is rudimentary in many third world countries, and general practitioners do not always have enough experience in diagnosing the different otitis medias. In this paper a system recently proposed by three of the authors for automated diagnosis of middle ear pathology, or otitis media, is extended to enable the use of the system on a smartphone with an Internet connection. In addition, a neural network is also proposed in the current system as a classifier, and compared to a decision tree similar to what was proposed before. The system is able to diagnose with high accuracy (1) a normal tympanic membrane, (2) obstructing wax or foreign bodies in the external ear canal (W/O), (3) acute otitis media (AOM), (4) otitis media with effusion (OME) and (5) chronic suppurative otitis media (CSOM). The average classification accuracy of the proposed system is 81.58% (decision tree) and 86.84% (neural network) for images captured with commercial video-otoscopes, using 80% of the 389 images for training, and 20% for testing and validation.