MALIN HEDMAN: Can AI improve eardrum diagnostics in primary care?

Project description

Otitis media is a common diagnosis that, if inadequately managed, may lead to serious complications. In high-income regions, the main challenge is not undertreatment but excessive antibiotic use, partly driven by overdiagnosis. Most patients are managed in primary care, yet knowledge about diagnostic accuracy in this setting is limited. The eHEar research initiative at Umeå University aims to develop a smartphone-based diagnostic tool integrated with a digital otoscope. By combining digital imaging, acoustic reflectometry, and artificial intelligence (AI)—specifically convolutional neural networks for image interpretation—the tool seeks to improve diagnostic accuracy. A critical component of this endeavour involves examining how AI-based technologies influence physicians’ clinical judgments and to explore attitudes toward AI among clinicians, providing a foundation for sustainable implementation. 

Aim

To investigate whether AI-supported technology can enhance diagnostic accuracy in otitis media and determine what is required for it to be considered reliable, safe, and useful in primary care. 

Method  

Studie I: Diagnostic accuracy of otitis media with and without a fictitious AI support among physicians in primary care and medical students. A survey study with evaluation of tympanic membrane images in two separate rounds, with the second round incorporating fictitious AI support. An expert panel determines the reference diagnosis. Diagnostic accuracy and confidence is calculated, alongside an assessment of the technological impact, to determine how evaluations are influenced by the AI support. 

Studie II: Attitudes towards AI-supported tools for ear diagnostics- a study within primary care in Västerbotten. An interview study involving physicians and district nurses. Semi-structured interviews, guided by an interview protocol, will be analysed according to qualitative content analysis. 

Studie III: Diagnostic accuracy of OM by CNN and GPs for classifying TM images and mobility in combination with symptoms. A prospective, clinical multicentre study with patients presenting with symptoms of otitis media. Following routine clinical management, they are examined with digital otoscopy, tympanometry, and acoustic reflectometry. In the lab, these data are entered into the eHear team’s novel AI model, which generates its diagnostic assessment. An expert panel determines the reference diagnosis, against which the diagnostic accuracy of both physicians and the AI model is calculated. Follow-up: Physicians will later review images from patients they previously managed and set a diagnose. This process is repeated with the addition of acoustic reflectometry data, and finally with the inclusion of AI support. Diagnostic accuracy and technological impact are then evaluated. 

Studie IV: To explore the prerequisite for ear and hearing diagnostics and the attitudes towards AI among professionals in primary care. An extension of the follow-up phase of the third study, in which participating physicians take part in a focus group interview. 

Relevance 

Development of a novel method for tympanic membrane diagnostics, with the aim of achieving higher diagnostic accuracy, enhanced patient safety, time saving and reduced pressure on antibiotic resistance. Identification of the prerequisites for implementing AI-based tools in primary care.