Retrospective Research Projects
Validation of a deep learning model for mammography in a Danish screening population: a multicenter study of diagnostic accuracy, feasibility and clinical attributes
Mohammad Talal Elhakim, MD, PhD student
Breast cancer is the most common cancer amongst women and the second leading cause of cancer deaths globally. Breast cancer screening is implemented with the purpose of reducing death rates through early detection.
To ensure a high diagnostic quality, all examinations are read independently by at least two expert radiologists. Still, not all cancers are detected at the initial mammography screening and there is a considerable false positive recall rate. This is altogether very resource demanding and currently there exists a significant shortage of radiologists.
Novel techniques within the field of computer science and artificial intelligence (AI), particularly deep learning (DL), has shown great potential if applied as a clinical decision support tool in ensuring earlier detection of breast cancer, reducing false positives and improving efficiency in mammography screening.
This study investigates if a commercially available CE marked DL model can optimize workflow in Danish mammography screening while maintaining a non-inferior cancer detection performance. The study is carried out as a multicenter validation study across all mammography units in the Region of Southern Denmark, from which nearly 262.000 consecutive mammograms from an entire screening population between 2014 to 2018 are collected with at least a 2-year follow-up period.
This study is expected to provide important information on how a DL model can improve diagnostic quality and optimize workflow in breast cancer screening. Results from this project are expected to benefit future national and international recommendations on the integration of DL solutions for mammography screening.