Deep metric learning for otitis media classification

Josefine Vilsbøll Sundgaard,Søren Laugesen,Rasmus R Paulsen,Yosuke Kamide,Anders Nymark Christensen,Peter Bray,Chiemi Tanaka,James Harte

doi:10.1016/j.media.2021.102034

Josefine Vilsbøll Sundgaard, Søren Laugesen + Show 6 more

Open Access

https://doi.org/10.1016/j.media.2021.102034

Copy DOI

Abstract

In this study, we propose an automatic diagnostic algorithm for detecting otitis media based on otoscopy images of the tympanic membrane. A total of 1336 images were assessed by a medical specialist into three diagnostic groups: acute otitis media, otitis media with effusion, and no effusion. To provide proper treatment and care and limit the use of unnecessary antibiotics, it is crucial to correctly detect tympanic membrane abnormalities, and to distinguish between acute otitis media and otitis media with effusion. The proposed approach for this classification task is based on deep metric learning, and this study compares the performance of different distance-based metric loss functions. Contrastive loss, triplet loss and multi-class N-pair loss are employed, and compared with the performance of standard cross-entropy and class-weighted cross-entropy classification networks. Triplet loss achieves high precision on a highly imbalanced data set, and the deep metric methods provide useful insight into the decision making of a neural network. The results are comparable to the best clinical experts and paves the way for more accurate and operator-independent diagnosis of otitis media.

Highlights

Otitis media is a group of diseases in the middle ear, which can be divided into two major diagnostic groups: acute otitis media (AOM) and otitis media with effusion (OME)
The results show that utilising class-weighted crossentropy has increased the precision on the under-represented class by 5%, at the expense of a lower AOM recall compared to standard cross-entropy loss, which was expected when introducing classweights in the loss function, while the rest of the performance measures are very similar to those of the standard cross-entropy measure
We demonstrate that it is possible to do automated classification of otitis media, and develop a diagnostic tool for detecting acute otitis media, otitis media with effusion, or no effusion

Summary

Introduction

Otitis media is a group of diseases in the middle ear, which can be divided into two major diagnostic groups: acute otitis media (AOM) and otitis media with effusion (OME). Around 11% of the world’s population suffer from AOM (Monasta et al, 2012), and it is the second most common reason for a visit to the doctor (Worrall, 2007). Otitis media with effusion is the most common cause of acquired hearing loss in childhood (Robb and Williamson, 2016) and The disease is usually treated with antibiotics, and it is the single diagnosis responsible for most prescriptions of antibiotics (Worrall, 2007), even though ’watch-and-wait’ is advised by many clinical guidelines to limit the overuse of antibiotics.

Methods

Results

Discussion

Conclusion