Otoscopic diagnosis using computer vision: An automated machine learning approach

Abstract

Objective: Access to otolaryngology is limited by lengthy wait lists and lack of specialists, especially in rural and remote areas. The objective of this study was to use an automated machine learning approach to build a computer vision algorithm for otoscopic diagnosis capable of greater accuracy than trained physicians. This algorithm could be used by primary care providers to facilitate timely referral, triage, and effective treatment.Methods: Otoscopic images were obtained from Google Images (Google Inc., Mountain View, CA), from open access repositories, and within otolaryngology clinics associated with our institution. After preprocessing, 1,366 unique images were uploaded to the Google Cloud Vision AutoML platform (Google Inc.) and annotated with one or more of 14 otologic diagnoses. A consensus set of labels for each otoscopic image was attained, and a multilabel classifier architecture algorithm was trained. The performance of the algorithm on an 89-image test set was compared to the performance of physicians from pediatrics, emergency medicine, otolaryngology, and family medicine.Results: For all diagnoses combined, the average precision (positive predictive value) of the algorithm was 90.9%, and the average recall (sensitivity) was 86.1%. The algorithm made 79 correct diagnoses with an accuracy of 88.7%. The average physician accuracy was 58.9%.Conclusion: We have created a computer vision algorithm using automated machine learning that on average rivals the accuracy of the physicians we tested. Fourteen different otologic diagnoses were analyzed. The field of medicine will be changed dramatically by artificial intelligence within the next few decades, and physicians of all specialties must be prepared to guide that process.

A method for utilizing automated machine learning for histopathological classification of testis based on Johnsen scores

“…Other research groups have already reported on medical image classification or otoscopic diagnosis performed using an automated deep learning approach with no coding 15 , 19 . We consider that an AI-based algorithm would also be suitable for quantification such as that in Gleason grading for prostate cancer.…”

Deep Learning for Classification of Pediatric Otitis Media

“…This method can automatically learn complex tasks in different fields. With 1,366 unique otoscopic images of 14 diagnoses collected from Google, Livingstone and Chau 22 have obtained sensitivities (recall) of 75% and 90.9% in the classification of AOM and OME. Our study has gained a total of 10,703 otoendoscopic images in train set and achieved sensitivities of 98.38% and 97.59% in the classification of AOM and OME in test set with Xception (Table I).…”

“…One of the important challenges of AI in otolaryngology is high‐quality and large quantities of patient data collection 13 . Based on a large amount of otoendoscopic images, several studies have developed the feature‐extractions‐based algorithms for automatic diagnosis of OM 15–18 . Recently, deep learning methods, which reduce the need for manual feature extraction are used to automated diagnosis of ear diseases 19,20 .…”

Shortwave infrared otoscopy for diagnosis of middle ear effusions: a machine-learning-based approach

“…Multiple studies have applied machine learning to visible light otoscopy for identification chronic otitis media with perforation, cerumen, myringosclerosis, and retraction pockets 45 , 62 – 64 . Livingstone and colleagues uploaded visible light otoscopy images into a Google Cloud automated algorithm, AutoML, and compared the diagnostic accuracy of the algorithm to physicians from a variety of specialties 44 . Serous otitis media was correctly diagnosed by the algorithm 80% of the time, compared to a rate of 44% among physicians.…”