What is AI in Medical Imaging?

The main benefits of using in AI in medical imaging include the ability to use algorithms to carry out large-scale projects involving multiple patient datasets, therefore reducing planning time and potentially improving clinical decision-making. By using AI, clinicians can obtain rapid insights or information that could not be easily found using traditional imaging review methods, for example in the heart. These inputs can then help with diagnosis, but also in terms of real-time monitoring and recommendations for patient’s health, managing hospital workflows, and helping with medical device design testing and simulation by speeding up common steps needed to produce 3D models from imaging.

According to the U.S. Food & Drug Administration (FDA), there are several types of AI used in healthcare, including CADe to detect information such as abnormalities within an image, and CADx that may use algorithms that add additional diagnostic information (such as a tumor grade). In addition, there are Non-CADe algorithms that can be used to help segment anatomical structures. More broadly, AI algorithms are trained based on large amounts of data for a specific purpose, for example to identify risk factors, and then developed and validated against new data to show that accurate results can be repeated.

Most AI or ML-based algorithms are ‘locked’, so that they will not adapt or learn beyond the specific training and performance testing, and are generally easier to control and reduce the risk of errors for. By comparison, ‘adaptive’ or ‘continuously learning’ algorithms will update their outputs based on real-world experience or new data; being able to regulate for these types of algorithm is more difficult, as it introduces a significant degree of uncertainty to whether the algorithms are still providing suitable outputs.

A further category is generative AI, for example ChatGPT, which uses very large amounts of data to generate new material, for example text. Adaptive AI is expected to continuously learn and update based on new data, and is more suited to dynamic data-driven situations, compared to the use of generative AI for creating synthetic creative content. Generative AI may have some applications to medical diagnosis, such as for creating realistic and anonymized patient data for medical research and training to test out new clinical inputs.