Back to the list
Congress: ECR24
Poster Number: C-12462
Type: EPOS Radiologist (scientific)
Authorblock: J. Gommers1, S. D. Verboom1, M. Webster2, C. Abbey3, M. Broeders1, I. Sechopoulos1; 1Nijmegen/NL, 2Reno, NV/US, 3Santa Barbara, CA/US
Disclosures:
Jessie Gommers: Nothing to disclose
Sarah Delaja Verboom: Nothing to disclose
Michael Webster: Nothing to disclose
Craig Abbey: Advisory Board: Izotropic Corp. Consultant: Canon Medical, Izotropic Corp.
Mireille Broeders: Research/Grant Support: Screenpoint Medical, Sectra Benelux, Hologic, Volpara Solutions, Lunit inc., iCAD Speaker: Hologic, Siemens Healthcare
Ioannis Sechopoulos: Research/Grant Support: Siemens Healthcare, Canon Medical, Screenpoint Medical, Sectra Benelux, Hologic, Volpara Solutions, Lunit Inc., iCAD Speaker: Siemens Healthcare Advisory Board: Koning Corp.
Keywords: Breast, Mammography, Screening, Cancer
Purpose Visual adaptation is a fundamental aspect of human perception, enabling individuals to adjust to different environmental conditions and stimuli. In the context of radiology visual adaptation may help with optimizing interpretation and thereby improving diagnostic performance (1). Radiologists encounter diverse sets of images, undergoing continuous adaptation while navigating through these images. Of particular interest is screening mammography, where radiologist must detect and classify subtle breast abnormalities indicative of breast cancer. In this study, we aim to investigate the impact of...
Read more Methods and materials An enriched fully-crossed, multi-reader multi-case (MRMC) study with 150 screening mammograms (75 cancers, 75 normals) was performed. Examinations included bilateral two-view digital mammography images and were interpreted by 13 Dutch screening radiologists. Each radiologist performed three sessions. Within each session the mammograms were ordered differently: randomly, by increasing Volpara®Density™ (v1.5.4.0) volumetric breast density percentage, or ordered to minimize the distance in an SSL encoding between mammograms. The SSL encoding was based on a Resnet-50 trained with a SimCLR framework to...
Read more Results Compared to random-order performance, ordering mammograms by increasing density enabled radiologists to increase their AUC (0.921 vs 0.934, P=0.002), with higher specificity (85.8% vs 89.3%, P=0.024), while maintaining sensitivity (80.8% vs 81.4%, P=0.776), and reducing reading time (27.9 seconds vs 24.3 seconds, P<0.001) and eye-fixation counts (52 vs 47, P<0.001). No significant difference in screening performance was detected for the SSL-ordered reading sequence.
Read more Conclusion Ordering screening mammograms by increasing breast density may improve reading performance, decrease reading time, and increase search efficiency. This could be due to the phenomenon of human visual adaptation. Ordering screening mammograms on increasing density would be relatively easy to implement in screening practice. However, prospective, un-enriched studies are needed to confirm these observations. Additionally, it would be interesting to investigate what happens if screening mammograms were ordered by decreasing instead of increasing breast density. 
Read more References Webster MA. Visual Adaptation. Annual Review of Vision Science. 2015 Nov 1;1:547-567. doi: 10.1146/annurev-vision-082114-035509. Chen T, Kornblith S, Norouzi M, Hinton G, editors. A simple framework for contrastive learning of visual representations. International conference on machine learning; 2020: PMLR.
Read more