Segmenting the Data Stream: Harnessing contrast‐enhanced imaging from widely available, anonymized patient data to Generate 3‐D Training Tools for Clinical Gross Anatomy

Understanding the organization of the human body—particularly through cadaveric dissection— is the foundation of modern medical education. A hands‐on appreciation for the remarkable three‐dimensional (3‐D) complexity and integration of nerves, blood vessels, muscle, sinew, and bone allows students to acquire the detailed knowledge of human anatomical structures necessary to build their own conceptual maps of how the body is organized. To forward these goals, educational applications for 3‐D technology are progressively expanding and becoming entrenched in medical and non‐medical classrooms as well as in doctors' offices and surgery bays. In healthcare, 3‐D technology is expected to revolutionize several areas of medicine by providing new methods for enabling mock diagnosis of normal and pathological states during student training, practicing clinical procedures in silico and physically with 3‐D printed models, and facilitating the development of patient‐customized prosthesis for faster recoveries with fewer complications. These initiatives have resulted in a vast amount of preexisting, anonymized 3‐D radiographic datasets, such computed tomography (CT) scans or magnetic resonance imaging (MRI), that can be readily reconstructed into 3‐D models. Here we describe a pipeline to access already available contrast‐enhanced CT and MRI datasets to develop 3‐D anatomical atlases (abdominal, cardiac, and cerebral vasculature using Avizo) for augmenting classroom learning initiatives through digital hubs such as Radiopaedia.org and DiceCT.com. Granted that the datasets have sufficient contrast differences allowing for differentiation of tissues, thick‐section scans allow for rapid reconstruction but may compromise anatomical accuracy, while thin‐section scans provide greater anatomical detail but also require extensive segmentation. We discuss how decimation tools and resampling approaches can be applied to smoothing the resulting 3‐D models. In addition, we illustrate steps that must be taken when choosing both the type of scan and the amount of detail desired in the final product. As new modalities for 3‐D technology arise, teaching styles and training materials that can evolve to embrace these new tools will see expanded applications spanning from student to patient. We anticipate that physicians‐in‐training will take such applications to even loftier heights by sourcing exemplar datasets from digital hubs that provide greater opportunities to share 3‐D models and provide more resources to a wider scope of students and clinicians. Support or Funding Information Funding was provided to Paul Gignac through the National Science Foundation (grants no. 1450850 and 1457180) and the Oklahoma State University Center for Health Sciences.