Data Curation for AI with Proper Medical Imaging Physics Context (2023)-R7-CPH11-2023

Video Transcription

Video Summary

The session titled "Data Curation for AI with Proper Medical Imaging Physics Context" emphasizes the critical role of medical physics in AI development for medical imaging. The importance of scientific and technical foundations, including physics and math, quality assurance, and safety, are highlighted as crucial to AI development and deployment in clinical settings. The session begins with a presentation from Jiwa Qi, focusing on data preparation for AI, exemplified by a COVID-19 chest X-ray diagnosis project. The talk stresses the significance of collecting high-quality, generalizable data and navigating the complexities of data storage and sharing under ethical and legal constraints. <br /><br />Dr. Ran Zhang continues by discussing data quality and the importance of avoiding shortcuts in model development. Using COVID-19 classification as an example, he emphasizes the necessity of a well-curated, high-quality data set even from a single clinical site for achieving generalizable models. Shortcut learning can yield misleadingly strong results on internal datasets but falter on external ones. Continuous monitoring of AI models for performance consistency based on external datasets is advocated.<br /><br />Dr. John Garrett focuses on implementing AI models in clinical settings, showing the substantial gap between research and practical application. He discusses various methods for AI integration, ranging from embedding models in PACS systems or EMRs to using dedicated AI platforms. The importance of choosing correct data inputs, considering use-case specific needs, and ongoing validation are emphasized to ensure AI tools are both effective and efficient in practice.

Keywords

Data Curation

Medical Imaging

AI Development

Medical Physics

Quality Assurance

COVID-19 Diagnosis

Data Sharing

Model Validation

Clinical Integration

AI Platforms