Short summary

The aim of this project is to determine the accuracy and feasibility of a novel deep learning-based artificial intelligence solution to be used as a clinical decision support tool within Danish breast cancer screening.

Rationale

Breast cancer is the most common cancer amongst women and the second leading cause of cancer deaths. Breast cancer screening is implemented with the purpose of reducing death rates through early detection. All examinations are read by at least two expert breast radiologists to ensure high diagnostic accuracy, but this is very resource demanding. Moreover, an existing shortage of radiologists makes it difficult to ensure continued high quality of the screening programme.

New techniques of artificial intelligence (AI) based on deep learning (DL) are proposed as a clinical tool that could potentially increase both quality and efficiency of breast cancer screening. The latest research within this field has shown encouraging results, however, thorough scientific validation on local data and in a real-life clinical setting is still needed before safely implementing such solutions in a Danish screening context.

The main purpose of this project is to investigate the potential and benefits of using an AI solution in Danish mammography screening. Through a large-scale retrospective study, including an entire unselected screening cohort from the Region of Southern Denmark, the project will validate a commercially available CE-marked DL model for breast cancer detection in terms of diagnostic accuracy and feasibility.

Description of the cohort

All women who had a mammography screening performed between August 2014 - August 2018 in the Region of Southern Denmark are eligible for inclusion in the study cohort.

Data and biological material

Screening data and register data is retrieved from local archives, the Danish Quality Database on Mammography Screening (DKMS) and the Danish Breast Cancer Cooperative Group (DBCG) database. Image data is retrieved from the Vendor Neutral Archive of RSD (VNA SYD).

Eligible study cases are processed by the DL model which delivers the analysis results and abnormality assessments of each exam.

Collaborating researchers and departments

Department of Radiology, Odense University Hospital; Research and Innovation Unit of Radiology, University of Southern Denmark

• Ole Graumann, MD, PhD, Associate Professor
• Benjamin Schnack Brandt Rasmussen, MD, PhD, Assistant Professor
• Lisbet Brønsro Larsen, MD, Associate Professor

Department of Computer Science, University of Copenhagen

• Mads Nielsen, PhD, Professor

Department of Nuclear Medicine, Odense University Hospital

• Oke Gerke, PhD, Professor

Centre for Innovative Medical Technology, Odense University Hospital/University of Southern Denmark

• Camilla Stryhn, Innovation Consultant

Regional IT, Region of Southern Denmark

• Henrik Johansen, IT Specialist Consultant