Post

How iMerit Approaches Medical AI Curriculum Design

November 30, 2020

Expert-annotated ground truth data will remain the bedrock of AI imaging for the foreseeable future. But if the supply of experts is limited to trained medical personnel, the future may not be a sustainable one. The key to training an effective workforce of lay domain specialists in the medical imaging field is multifaceted. Even before a training curriculum can be designed, a methodology must be devised to identify and evaluate prospective candidates who demonstrate both the aptitude and the motivation to complete a rigorous training regime. Then experienced medical professionals must be engaged to design the training curriculum, teaching both the general skills of using the annotation tools, and the specific skills associated with the project focus. Ideally the curriculum would be designed in close consultation with the client, resulting in a highly focused, narrow but deep course of study.

Experienced medical professionals must be engaged to design the training curriculum, teaching both the general skills of using the annotation tools, and the specific skills associated with the project focus

The curriculum begins with learning the toolset, which gets more advanced on a regular basis, from 2D images to multi-planar navigation in 3D imaging to 4D CINE studies. And while broad medical training is not required, the narrow-but-deep curriculum does include anatomy and physiology of the relevant area(s) for the use case under examination, and an understanding of the terminology as well. This elevates the annotation work from simple recall to deeper understanding, creating a more robust skillset. For example, teams working on cardiology projects can understand anatomy and physiology well enough to commonly navigate congenital and pathologic aberrations. Because all training is hands-on, rather than based solely on textbooks and lecture halls, progress is typically quick, and getting quicker as pedagogical techniques are refined. This is a very agile microskilling model that allows people to learn and relearn new use cases, often within one-week cycles.

The modular curriculum is also technology-centric, and includes gamification, anywhere/anytime learning, and a high degree of personalization. For example, a custom-designed tool to assess the trainee’s computer vision skills quickly identifies strengths and weaknesses and allows for additional focused training. A result of iMerit’s social impact model is that motivation levels are very high and attrition levels are very low. Jobs at iMerit are life-changing so the investment in Learning & Development and the accumulation of knowledge and skill over time, is retained in the organization and customers love working with stable teams. Clients report that they like investing in training iMerit’s teams because they see the fruit of their efforts over time.

If you wish to learn more about creating training data sets for machine learning, please contact us to talk to an expert.