MAGIC - AI for Breast Cancer Diagnostics

Breast cancer is the most common form of cancer amongst women and the second leading cause of cancer deaths globally. Population-based organised breast cancer screening is implemented with the overall purpose of reducing mortality through early detection. In Denmark, all women aged 50-69 years are biennially offered mammography screening which is read by at least two expert breast radiologists in accordance with European guidelines. Although double reading is an effective method to reduce missed cancers, it also increases the number of false positives leading to an excess of unnecessary testing and mental distress. Moreover, due to growing demand and a shortage of radiologists, the screening program is very resource demanding.

In recent years, advancements in computer science have presented artificial intelligence (AI) solutions based on novel deep learning (DL) technology as a clinical tool to potentially increase both quality and efficiency of breast cancer screening. DL methods have been tested for numerous applications in breast cancer diagnostics with impressive results in terms of improvement of cancer detection and prediction on mammography in particular as well as reducing false positives. However, all existent studies have tested a DL-based AI system on either a retrospective dataset (mainly single-centre studies) or in limited reader studies, which thereby reduces the generalisability of the results to a real-world clinical setting.

Aim

The objectives of this project were to validate and test an AI system in an existing radiology workflow and analyse the impact on cancer detection and recall rates as well as workload. The project presented the world’s first large-scale validation study and a prospective clinical trial which would assess the impact of a DL-based workflow in Danish clinical screening practice. The project was a multicentre collaboration across all breast cancer screening clinics in the Region of Southern Denmark and included a complete screening cohort representing almost one-fifth of the entire Danish population in order to obtain a representative sample.

The results of this project were expected to elucidate upon the benefits and feasibility of AI solutions for breast cancer screening in an existing radiology workflow in terms of diagnostic accuracy and workflow efficiency. Most importantly, the project would provide important knowledge on the potential of AI systems in detecting clinically significant cancers which could ultimately improve patient outcome.
 

Participants

The project was organised by the Department of Radiology at Odense University Hospital, and all radiology departments in the Region of Southern Denmark were part of the project.

Centre for Innovative Medical Technology (CIMT) contributed with project management.

Funding

The project was funded by the Region of Southern Denmark.