Associate Professor Narinder Singh
MBBS (Syd) MS (Syd) FRACS (OHNS)
ENT Specialist Surgeon, Rhinologist, Westmead Hospital; Clinical Associate Professor, University of Sydney, Australia
ENT Specialist Surgeon, Rhinologist, Westmead Hospital; Clinical Associate Professor, University of Sydney, Australia
|
Narinder Singh is a Rhinologist/ Anterior Skull Base Surgeon, Head of Department at Westmead Hospital, Australia’s largest healthcare campus and Clinical Associate Professor at the University of Sydney.
He has published over 80 peer-reviewed publications, university textbooks and book chapters and has given over 230 guest and keynote presentations at international scientific meetings. He has supervised 12 PhD and Masters students to completion with another 5 currently in progress. His team has received major grants from the Australian Research Council, the Passe and Williams Memorial Foundation, The Ramsay Hospital Research Foundation and Microsoft as well as major research funding from ResMed, Optinose and GSK. A/Prof Singh has 2 key research themes: 1) Artificial Intelligence (AI). He founded the Society for AI in Medicine, Surgery and Healthcare (www.amsah.org), convening the first Australian congress in Sydney, 2019, with a recent meeting in Brisbane, 2023. His flagship project, DrumBeat.ai, aims to address ear disease in rural and remote Indigenous Australian children (www.drumbeat.ai). 2) Computational Fluid Dynamics (CFD). He founded the Society for CFD of the Nose and Airway (www.scona.org), convening the first world congress in London, 2018, with subsequent meetings in Chicago, Oklahoma and Brisbane. |
|
DrumBeat.ai: Using Artificial Intelligence (AI) to Address Ear Disease in Rural and Remote Indigenous Australian Children
Background
In Aboriginal and Torres Strait Islander children living in rural and remote Australia, chronic ear disease can have lifelong impacts on hearing, education, social issues and employment. Access to otolaryngologists is limited in such areas. Telehealth has proven helpful but has several limitations. The aim of this project was to use AI to classify ear disease and predict the likelihood of hearing loss in Aboriginal and Torres Strait Islander children.
Methods
Otoscopic images and audiometric data were were collected by remote nurses and audiologists from Queensland and the Northern Territory from 2010 to 2020. Deep learning methods were used to develop models to a) classify ear disease and b) predict the likelihood of hearing loss using otoscopic images alone. Performance was evaluated by accuracy and agreement (prevalence-and-bias adjusted κ values).
Results
9086 otoscopic images were used to train an AI model to identify normal, acute otitis media, otitis media with effusion, or chronic otitis media, achieving 90% accuracy and near perfect agreement (κ = 0.86). To identify the presence of hearing loss, 4436 otoscopic images with corresponding audiometry results were used, achieving 88% accuracy and substantial agreement (κ = 0.76).
Conclusion
AI models can identify ear disease and predict the likelihood of hearing loss in Aboriginal and Torres Strait Islander children using otoscopic images alone. We are currently undertaking on-site field tests of these AI models at remote sites in Qld, the NT and WA.
In Aboriginal and Torres Strait Islander children living in rural and remote Australia, chronic ear disease can have lifelong impacts on hearing, education, social issues and employment. Access to otolaryngologists is limited in such areas. Telehealth has proven helpful but has several limitations. The aim of this project was to use AI to classify ear disease and predict the likelihood of hearing loss in Aboriginal and Torres Strait Islander children.
Methods
Otoscopic images and audiometric data were were collected by remote nurses and audiologists from Queensland and the Northern Territory from 2010 to 2020. Deep learning methods were used to develop models to a) classify ear disease and b) predict the likelihood of hearing loss using otoscopic images alone. Performance was evaluated by accuracy and agreement (prevalence-and-bias adjusted κ values).
Results
9086 otoscopic images were used to train an AI model to identify normal, acute otitis media, otitis media with effusion, or chronic otitis media, achieving 90% accuracy and near perfect agreement (κ = 0.86). To identify the presence of hearing loss, 4436 otoscopic images with corresponding audiometry results were used, achieving 88% accuracy and substantial agreement (κ = 0.76).
Conclusion
AI models can identify ear disease and predict the likelihood of hearing loss in Aboriginal and Torres Strait Islander children using otoscopic images alone. We are currently undertaking on-site field tests of these AI models at remote sites in Qld, the NT and WA.
