Technical characterisation of digital stethoscopes: towards scalable artificial intelligence-based auscultation

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Digital stethoscopes integrate artificial intelligence (AI) for improved diagnostic accuracy and to overcome limitations in manual auscultation.
• Scalable AI development is hindered by sensor bias arising from variations in digital stethoscope acquisition devices.
• Manufacturers often lack transparency regarding device specifications, specifically frequency responses, complicating AI integration.

Purpose of the Study:

• To develop and present an effective methodology for determining the frequency response of digital stethoscopes.
• To characterize the frequency responses of three common digital stethoscopes: Littmann 3200, Eko Core, and Thinklabs One.
• To highlight the inter-device variability and its implications for AI-assisted auscultation.

Main Methods:

• A novel methodology was employed to measure the frequency response of digital stethoscopes.
• Three distinct digital stethoscope models (Littmann 3200, Eko Core, Thinklabs One) were characterized.
• Frequency response data was analyzed to identify inter- and intra-device variations.

Main Results:

• Significant inter-device variability was observed, with distinct frequency responses among the Littmann 3200, Eko Core, and Thinklabs One.
• Moderate intra-device variability was noted when comparing two units of the Littmann 3200.
• The study confirms that digital stethoscope frequency responses are not uniform.

Conclusions:

• There is a critical need for device normalization to ensure the success of AI-assisted auscultation systems.
• The developed methodology provides a foundational approach for technical characterization of digital stethoscopes.
• Addressing device variability is essential for reliable and scalable AI applications in digital auscultation.