Application of artificial intelligence in image reconstruction and processing in radiology and nuclear medicine

Imaging procedures that use ionizing radiation are to be optimized from the perspective of radiation protection. AI-based approaches in medical imaging could help provide better diagnostic information at the same patient exposure level or reduce patient exposure while maintaining comparable diagnostic image quality. However, the use of AI-based procedures can also, for example, result in the unintended suppression or artificial addition of relevant information. Thus, it is necessary to assess what potential problems must be considered and how they can be prevented.

• Which AI approaches are appropriate for image optimization in medical imaging and what risks are associated with their use?
• What are the current framework conditions for employing such methods, and what conditions must exist in the future to ensure the safe use of ionizing radiation as well as sufficient reliability of actual diagnostic features?
• Can requirements for the methodology of an AI algorithm in image optimization be developed with reference to the intended field of application and the design of quality assurance?

• There is a wide range of approaches for developing AI-based methods for image reconstruction and noise reduction of medical image data. In principle, such approaches can be suitable for better utilizing the acquired or measured image (raw) data.
• If data can be utilized more efficiently, this can yield better image results from a diagnostic perspective at the same radiation exposure for the patient, or enable a reduction of radiation exposure while maintaining image quality.
• Because of potential sources of error in AI-based methods (e.g., non-transparent algorithmic image data processing, frequently non-reproducible original information, risk of suppression or artificial addition of image content), diagnostic equivalence to standard methods must always be ensured. This is also a fundamental requirement from the perspective of radiation protection law.
• In addition to documenting which models are used and how they are trained, it is particularly important to carefully select the data sets used for training, testing, and validation. The recommendation includes guidance on how to develop, review, and implement safe procedures.
• For quality assurance in terms of radiation protection, not only system checks but also the correct application of methods are crucial. This necessitates a clear understanding of the methods and standard operating procedures for medical professionals.