An Educational Graphical User Interface to Construct Convolutional Neural Networks for Teaching Artificial...

Area of Science:

• Medical image analysis
• Machine learning in radiology
• Radiology training

Background:

• Deep learning, particularly convolutional neural networks (CNNs), shows promise in medical image analysis.
• Radiology training often lacks comprehensive instruction in deep learning, hindering ML integration.
• A gap exists in accessible tools for teaching machine learning to radiology professionals.

Purpose of the Study:

• To develop and evaluate an educational graphical user interface (GUI) for teaching deep learning concepts to radiology trainees.
• To assess the usability and effectiveness of the GUI as a teaching tool for machine learning in radiology.
• To facilitate the integration of machine learning into radiology research and clinical practice through improved education.

Main Methods:

• Developed an educational GUI using Python with PyQt and PyTorch frameworks.
• Demonstrated GUI functionality via a binary classification task on brain MR images.
• Assessed usability with 5 neuroradiologists/fellows using task completion times, SUS, and qualitative feedback.

Main Results:

• Users successfully completed all assigned tasks after a brief introduction and GUI walkthrough.
• No significant difference in task completion time was observed between users and an ML expert.
• The GUI achieved a high System Usability Scale (SUS) score of 82.5, indicating excellent usability.

Conclusions:

• The educational GUI is a highly usable and effective tool for teaching machine learning to radiology trainees.
• Interactive GUI-based learning can be integrated into radiology training programs.
• This tool addresses the educational gap, empowering radiologists to utilize deep learning in their practice.