My User Area
MyUserArea

Artificial intelligence and radiology: a perfect match?

Dr. Michael Forsting (Essen, DE), during the special focus session on “artificial intelligence and radiology” at ECR 2018 discussed where artificial intelligence could improve radiology.
Artificial intelligence (AI) currently uses deep learning, a set of algorithms in machine learning that attempts to model high-level abstractions in data by using model architectures composed of multiple non-linear transformations.

According to Dr. Forsting, these algorithms could be able to read mammograms with an accuracy up to 99% in screening. AI may therefore drastically decrease the workload for radiologists in population-based screening programs and consequently decrease the number of radiologists needed. Hence population based screening may become more widely available, in particular in low-resource regions in the world where dedicated breast radiologists are relatively scarce and the incidence of breast cancer is still increasing. AI may also aid as a decision support tool and might potentially improve the accuracy of breast radiologists.

Moreover AI could be an accurate tool in the follow-up setting in order to detect temporal changes (e.g. in neo-adjuvant therapy monitoring and post-therapy follow up). AI can also be applied to radiomics, where it may be able to assess the biological building blocks of the tumor and may possibly predict the response to treatment and patient outcome.

Finally Dr. Forsting underlined that AI will likely not replace radiologists, but it will change and improve the radiology workflow dramatically, in particular by supporting decision-making. The availability of AI aids in the analysis of the images and in supporting the diagnosis and may also improve the efficiency. Consequently, AI may allow radiologists more time to do what human do better, as patient management and radiologist-patients relationship, thus improving not only the reporting quality but also patient outcome and experience.

Listen to the lecture via ECR Online (http://ecronline.myesr.org/ecr2018/)