Discovery Viewer (DV): Web-Based Medical AI Model Development Platform and Deployment Hub.

Area of Science:

• Medical Artificial Intelligence (AI)
• Deep Learning (DL) in Healthcare
• Biomedical Imaging Analysis

Background:

• Sharing medical data for AI development is challenging due to Protected Health Information (PHI).
• Pre-trained models offer a solution, but adaptable environments are needed for clinical task-specific fine-tuning and peer testing.
• The Biomedical Engineering and Imaging Institute at Mount Sinai developed the Discovery Viewer (DV) platform to address these needs.

Purpose of the Study:

• To demonstrate the potential of the Discovery Viewer (DV) platform in enabling individuals without AI expertise to create high-performing deep learning (DL) models for clinical applications.
• To showcase the platform's utility through various use cases, including segmentation, regression, and classification tasks in musculoskeletal AI projects.

Main Methods:

• Three non-AI experts utilized the DV platform to develop musculoskeletal AI models for segmentation, regression, and classification.
• Participants annotated data and trained models using the platform's "Training Module".
• Model performance was evaluated on a separate 20% hold-off dataset.

Main Results:

• The classification model achieved 0.94 accuracy, 0.92 sensitivity, and 1.0 specificity.
• The regression model reported a mean absolute error of 14.27 pixels.
• The segmentation model attained a Dice Score of 0.93, with 0.9 sensitivity and 0.99 specificity.

Conclusions:

• The Discovery Viewer (DV) platform successfully empowers individuals without AI expertise to develop effective DL models for medical applications.
• The platform facilitates the creation and distribution of cutting-edge AI models, promoting peer testing and refinement.
• This initiative aims to broaden the medical AI developer community and accelerate the transition of AI models from research to clinical practice.