BE 4963 - Best Practices in Visualization and Labeling of Medical Images

• Explain the purpose of ground truth information in artificial intelligence applications  
• Describe practical options and challenges associated with clinically acquiring large sets of images and creating ground truths  
• Explain different strategies that can be used to improve quality and consistency of medical image annotations  
• Describe potential consequences of errors and uncertainties in medical image annotations and strategies to mitigate these occurrences   
• Recognize the value of data visualization in health care  
• Identify the key techniques in medical image visualization  
• Be familiar with medical visualization tools 

• Basic understanding of various medical imaging systems, image processing methods, and physiology 

• What is ground truth and why is it essential in AI?  
• Current clinical approaches in obtaining images and creating ground truths (anatomical, diagnosis, etc.)  
• Sources of images, logistics of accumulating large numbers of images  
• Necessity of uniform acquisition protocols  
• Regulatory and ethical issues   
• Who does the labeling?  
• Technical strategies in identifying structures  
• Using information from different slices  
• Incorporating AI into predicting structures (pre-label attempts)  
• Challenges associated with errors and uncertainties in medical image labeling for AI  
• Sources of variability in medical image annotation  
• Adjudication mechanisms to produce gold standards for annotation  
• Potential consequences of variability in AI algorithm validity and subsequent applications   
• Data visualization in health care (clinical records, dashboards, sepsis management, imaging, board and public presentations) 
• Applications of medical image data (diagnosis, treatment planning, intraoperative support, documentation, anatomy education and surgical training, medical research) 
• Requirements for the selection of imaging modalities (relevant anatomy, data resolution, image quality with respect to contrast, signal-to-noise ratio, minimal exposure and burden to the patient and to the medical doctor, minimizations of costs) 
• Medical imaging selection tasks (Selection of points at surfaces, a region of interest, and objects; picking and snapping; selection feedback) 
• Data visualization tools in health care