FunduScope: a human-centered, machine learning-based interactive tool for training junior ophthalmologists in...

Area of Science:

• Ophthalmology
• Medical Education
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) detection from fundus images is crucial for preventing visual impairment.
• Current training methods for junior doctors are limited by experienced ophthalmologist availability and static learning materials.

Purpose of the Study:

• To develop and evaluate FunduScope, an interactive, ML-based learning tool for training junior ophthalmologists in DR detection.
• To address the need for flexible and accessible medical training resources.

Main Methods:

• A qualitative pre-study identified junior doctors' learning needs.
• A mixed-methods study evaluated FunduScope with 10 junior doctors, assessing usability, cognitive load, and learning experience.
• The tool utilizes a pre-trained ML model for DR classification.

Main Results:

• FunduScope demonstrates potential as an effective e-learning tool in medical education.
• The tool helps overcome time constraints for senior ophthalmologists and offers learning independence to junior doctors.
• Despite technical limitations, the study confirmed the tool's value.

Conclusions:

• ML-based tools like FunduScope can significantly enhance ophthalmology training.
• Future development could incorporate explainable AI (XAI) for improved clinical decision support.
• The tool's scope can be expanded to include other ophthalmic diseases.