DrumBeat.ai: AI for Indigenous Ear Care

Aim

The aim of this project is to evaluate the usefulness and acceptance of the DrumBeat.ai algorithm as a clinical aid for healthcare workers managing middle ear disease in Indigenous children in primary care settings, in rural and remote locations.

Outcomes

• Development of a world-first AI triage tool (DrumBeat.ai): Co-designed by an Indigenous  and non-Indigenous team and trained on >10,000 otoscopic images collected over 10 years from rural and remote Aboriginal and Torres Strait Islander children in the Northern Territory and Queensland, with linked audiometry, tympanometry and nurse assessments.
• Strong diagnostic performance in validation testing: When tested using otoscopic images alone, DrumBeat.ai achieved 88% accuracy, substantial agreement with specialist otolaryngologists (κ = 0.79), and excellent discriminative ability (AUC 0.963–0.997).
• Potential to improve frontline triage: The algorithm provides an objective, point-of-care decision-support tool for healthcare workers, helping address limitations of current telehealth models.

Background

Aboriginal and Torres Strait Islander children living in rural and remote areas have the highest rates of ear disease in the world (1). The World Health Organisation considers otitis media to be a serious Australian public health issue requiring urgent attention and innovative solutions (2, 3). Indigenous children are five times more likely to be diagnosed with severe otitis media than non-Indigenous children (4). Untreated otitis media can progress to serious complications that include tympanic membrane perforation, permanent hearing loss, mastoiditis, labyrinthitis, meningitis, intracranial abscess, cavernous sinus thrombosis and death (5). The commonest complication of otitis media, conductive hearing loss, has been demonstrated to affect Indigenous children in the first three years of life, which can adversely affect speech, language and cognitive development (6).

Diagnosing ear disease requires multiple modalities including clinical history, otoscopy, pneumatic otoscopy, audiometry and tympanometry (7).

Specialist otolaryngologists are trained to provide the most accurate diagnoses, but access to such specialists
is limited in rural and remote communities. Telehealth services have been demonstrated to be the most effective substitute method of ear disease screening in such communities (8). Ear assessment data is gathered by local or travelling healthcare workers (HCW),
including nurses and audiologists.

Typically, a “store-and-forward” telehealth approach is used, rather than “live” telehealth (7, 9-11). However, store-and-forward telehealth in otolaryngology results in delays between data capture, specialist review and  treatment initiation. This leads to an increased risk of complications from missed opportunities for intervention at the initial point of patient contact. Artificial intelligence (AI) is a rapidly evolving field of computer science that uses computer algorithms  to perform tasks typically requiring human intelligence. Convolutional neural networks (CNNs) are a form of AI that can be trained using large volumes of accurately labelled training data (supervised learning) to automatically classify unlabelled test data.  In otolaryngology, the integration of AI-powered diagnostic tools into healthcare delivery systems represents a novel solution and demonstrates significant potential for enhancement of remote healthcare delivery.

AI algorithms have the potential to provide an instant point-of-contact diagnosis, based on otoscopic images alone. In collaboration with rural and remote Indigenous communities and experts, local healthcare workers, ENT specialists, universities, hospitals, government and Microsoft’s “AI for Good” computer science team, our group has developed and refined such an AI algorithm (DrumBeat.ai) over a five-year period. The algorithm has been trained using over 10,000 otoscopic images, with associated audiometry, tympanometry and nurse impressions, collected over a 10-year period from rural and remote Aboriginal and Torres Strait Islander children in the Northern Territory and Queensland.

When tested on otoscopic images alone, the algorithm has demonstrated an accuracy of 88% with substantial agreement compared with specialist otolaryngologists and an AUC of 0.963 (95% CI: 0.941-0.986), on test images (12-14). The use of such an algorithm in clinical settings has the potential to have a significant impact on the incidence and severity of ear disease in rural and remote Aboriginal and Torres Strait Islander children by providing frontline healthcare workers with a triage tool that mitigates the limitations of current telehealth approaches.

Furthermore, the intrinsic wholistic approach to the provision of care enables nurses to met key areas within clinical governance frameworks. By nature, nurses have an understanding of the wider local social determinants of health and have the ability to provide culturally appropriate and relevant health care (Wood, Brown et al. 2024). Nurses are able to provide targeted care and interventions that empower patients to take ownership over their health care (Htay and Whitehead 2021). The sound development model of care for the NLWIC's provides a framework from which clinicals can provide evidence-based same day care for patients in the Central Queensland region.

Discussion

Drumbeat.ai provides an opportunity to find a novel, feasible and affordable solution by leveraging developing AI technologies to assist with the accurate diagnosis of ear disease to inform appropriate triaging of children thereby enhancing the efficiency of services by identifying children who require further assessment compared with children who can be observed. This will streamline workflow efficiency by prioritising high-risk cases for specialist review and treatment, while allowing low-risk cases to be managed within the community. Translation of our preliminary work requires assessment of this tool in clinical settings. The application of artificial intelligence algorithms in ear disease highlights the potential for improving and modernising screening practices within clinical practice, particularly in under-resourced communities.

This project addresses an urgent health issue in Australia, directly addressing the exceptionally high rates of ear disease among Aboriginal and Torres Strait Islander children in rural and remote regions. The use of artificial intelligence to provide a diagnosis from digital otoscopic eardrum images that matches the accuracy of an ENT specialist is an innovative approach, addressing the gap in access to specialist care. This project originated through liaison with Aboriginal and Torres Strait Islander community representatives who identified critical weaknesses in current telehealth models. The potential impact of earlier detection of ear disease is multifaceted: improved ear health outcomes have the potential to dramatically improve hearing, language skills, school and social participation, community and workforce engagement, reduced interactions with the criminal justice system, academic performance, and quality of life for Aboriginal and Torres Strait Islander children, with far-reaching societal benefits.

The project also aims to empower local  health workers by providing them with a decision support tool to instantly identify and triage ear disease and detect hearing loss, improve their diagnostic capabilities with traditional methods, and improve patient and family understanding of ear disease  with digital imaging accompanying this program, building the capacity of Indigenous health workers and other primary healthcare clinicians. Precious healthcare resources can be re-allocated towards supporting treatment rather than triage. This project supports Queensland Health's commitment to addressing health inequities for Aboriginal and Torres Strait Islander people, contributing to breaking the cycle of disadvantage for this specific cohort.

We expect that successful usage and acceptance of the AI algorithm y the target groups will validate the potential widespread roll-out of the tool across similar communities throughout Australia. Secondary benefits include potential use of the same technology in primary care (General practice) and hospital emergency departments for all Australians. Successful widespread use has the potential to significantly reduce the current unacceptably high burden of ear disease in rural and remote indigenous children, thereby helping to “close-the-gap” on this intractable public health crisis and provide an opportunity for more equitable distribution of healthcare resources.

References

1. Morris PS, Leach AJ, Silberberg P, et al (2005) Otitis media in young Aboriginal children from remote communities in Northern and Central Australia: a cross-sectional survey. BMC Pediatr 5:27.

2. Spurling GK, Askew DA, Schluter PJ, et al (2014) Household number associated with middle ear disease at an urban Indigenous health service: a cross-sectional study. Aust J Prim Health 20(3):285-90.

3. DeLacy J, Dune T, Macdonald JJ (2020) The social determinants of otitis media in aboriginal children in Australia: are we addressing the primary causes? A systematic content review. BMC Public Health 20(1):492.

4. Gunasekera H, Knox S, Morris P, et al (2007) The spectrum and management of otitis media in Australian indigenous and nonindigenous children: a national study. Pediatr Infect Dis J 26(8):689-92.

5. Danishyar A, Ashurst JV. Acute Otitis Media.  StatPearls. Treasure Island (FL)2024.

6. Lehmann D, Weeks S, Jacoby P, et al (2008) Absent otoacoustic emissions predict otitis media in young Aboriginal children: a birth cohort study in Aboriginal and non-Aboriginal children in an arid zone of Western Australia.  BMC Pediatr 8:32.

7. Habib AR, Perry C, Crossland G, et al (2023) Inter-rater agreement between 13 otolaryngologists to diagnose otitis media in Aboriginal and Torres Strait Islander children using a telehealth approach. Int J Pediatr Otorhinolaryngol 168:111494.

8. O'Neil LM, O'Neill M, Whelan F, et al (2024) Novel ENT live telehealth and live video-otoscopy clinics in remote Australia: outcomes and comparisons to traditional clinic models. J Laryngol Otol 138(3):253-7.

9. Gunasekera H, Miller HM, Burgess L, et al (2018) Agreement between diagnoses of otitis media by audiologists and otolaryngologists in Aboriginal Australian children. Med J Aust 209(1):29-35.

10. Jacups SP, Kinchin I (2021) A rapid review of evidence to inform an ear, nose and throat service delivery  odel in remote Australia. Rural Remote Health 21(1):5611.

11. Gotis-Graham A, Macniven R, Kong K, et al (2020) Effectiveness of ear, nose and throat outreach programmes for Aboriginal and Torres Strait Islander Australians: a systematic review. BMJ Open 10(11):e038273.

12. Habib AR, Wong E, Sacks R, et al (2020) Artificial intelligence to detect tympanic membrane perforations. J Laryngol Otol 134(4):311-5.

13. Habib AR, Xu Y, Bock K, et al (2023) Evaluating the generalizability of deep learning image classification algorithms
to detect middle ear disease using otoscopy. Sci Rep 13(1):5368.

14. Habib AR, Crossland G, Patel H, et al (2022) An Artificial Intelligence Computer-vision Algorithm to Triage Otoscopic Images From Australian Aboriginal and Torres Strait Islander Children.  Otol Neurotol 43(4):481-8.