Investigating Breast Density and the Feasibility of Personalised Breast Cancer Screening in Saudi Arabia

Abstract

Breast cancer is the commonest cancer among women in Saudi Arabia. It has been established in other screening populations that breast density is an independent risk factor for breast cancer and affects the sensitivity of mammography screening. Thus, women with increased breast density are most likely to develop breast cancer and to face late diagnosis with mammography screening alone. Therefore, the majority of screening programs, including the Saudi National Breast Cancer Screening Programme (SNBCSP), may not sufficiently serve these women, and additional or alternative imaging may be more beneficial. This thesis aims to examine breast density in the Saudi breast cancer screening population and its relationship with breast cancer risk, in addition to investigating the feasibility of personalised breast cancer screening for Saudi women on the basis of breast density. In a retrospective cohort study, the breast density of available screening mammograms from the SNBCSP was assessed visually by the Breast Imaging and Reporting Data System (BI-RADS) and Visual Analogue Scale (VAS) and automatically by predicted VAS (pVAS) for processed (pVASprocessed) and raw images (pVASraw) and VolparaTM. The relationship between density methods was assessed using the intra-class coefficient (ICC) and weighted kappa (κ). A nested case-control study was conducted to assess breast cancer risk based on density. Breast density was assessed using visual methods for 285 cases and 855 controls and visual and automated methods for a subset of 160 cases and 480 controls. Odds ratios (ORs) were estimated using conditional logistic regression. Compared to Saudi radiologists, the diagnostic performance of a stand-alone artificial intelligence (AI) system was estimated by calculating sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC). The diagnostic performance of mammography, synthetic mammography (SM), digital breast tomosynthesis (DBT), ultrasound and abbreviated breast MRI (ABMRI) was also assessed for women with dense breasts. Around one-third of Saudi women of screening age had dense breasts. The inter-reader agreement between radiologists was moderate for BI-RADS and excellent for VAS. The highest correlation between visual and automated density methods was for VAS and pVASraw (ICC= 0.86). Among all continuous methods, VAS was the strongest predictor of breast cancer (OR=7.54, 95% confidence intervals (CI) 3.86-14.74), followed by pVASraw (OR=5.38, 95% CI 2.68-10.77) and Volpara (OR=3.55, 95% CI 1.86-6.75) in the highest vs. lowest quartiles. The sensitivity of the AI system was higher (94.3%) compared to radiologists (91.6%); however, the specificity and AUC were lower (p

Date of Award	14 Dec 2022
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Susan Astley Theodossiadis (Main Supervisor) & Elaine Harkness (Co Supervisor)