Receive a weekly summary and discussion of the top papers of the week by leading researchers in the field.

In Der Radiologe ; h5-index 0.0

CLINICAL/METHODOLOGICAL ISSUE : Artificial intelligence (AI) is being increasingly used in the field of radiology. The aim of this review is to illustrate the developments expected in the next 5 to 10 years as well as possible advantages and risks.

STANDARD RADIOLOGICAL METHODS : Currently, all computed tomography (CT) images are reconstructed using programmed algorithms. Pathologies are detected by the radiologist with a high expenditure of time and evaluated using standardized procedures.

METHODOLOGICAL INNOVATIONS : AI can potentially provide a significant improvement to all these standard procedures in the future. CT reconstructions can be significantly enhanced using generative adversarial networks (GAN). Histology can be evaluated using radiomics or deep learning (DL)-based image analysis and the prognosis of the patient can be predicted highly individualized.

PERFORMANCE : The performance of the networks is strongly influenced by data quality and requires extensive validation. The ability and willingness of the manufacturers to integrate these into the existing RIS/PACS systems is also decisive.

EVALUATION : AI will have a large impact on the daily clinical work of radiologists. However, publications on the risks of the technology and on adequate validation are still lacking. In addition to opening new fields of application, further research regarding possible risks is warranted.

PRACTICAL RECOMMENDATIONS : In the next 5 to 10 years, AI will improve and facilitate work in clinical practice. The integration of the applications into the existing RIS/PACS systems is expected to take place via app stores and/or existing teleradiology networks.

Haubold Johannes


Deep learning, Image analysis, Radiomics, Risks, Validation