Seminar Summary and Learning Objectives
This presentation outlines the development of a comprehensive AI-enhanced workflow for ventilation/perfusion (V/Q) imaging, focused on batched data extraction, improving image quality, pseudo-planar V/Q image generation, diagnostic accuracy, and clinical interpretability in pulmonary embolism (PE) assessment. The talk covers the following key components:
1. Data Acquisition and Integration
- Automated querying of relevant clinical cases from the Electronic Medical Record (EMR)
- Extraction and curation of imaging studies from the Picture Archiving and Communication System (PACS)
- Streamlined downand preprocessing of V/Q scan data for AI training and inference
2. Image Quality Enhancement and Count Correction
- Development of an AI model to enhance the quality of V/Q images
- Pseudo-planar V/Q image generation
3. Defect Segmentation Using Deep Learning
- Design and training of a segmentation model to localize perfusion and ventilation defects
- Handling multiple projections and aligning outputs across views
4. Anatomical Mapping and PE Burden Quantification
- Mapping segmented defects to a standard lung atlas for anatomical localization
- Automated quantification of PE burden across lung lobes for objective reporting
5. Clinical Relevance and Future Directions
- Integration within clinical workflows
- Toward real-time decision support in nuclear medicine and radiology
- Direct application in other diseases (COPD, CTEPH, functional avoidance radiotherapy, and lobectomy) and other departments (Thrombosis)
