Capability and reliability of deep learning models to make density predictions on low-dose mammograms.

Purpose: Breast density is associated with the risk of developing cancer and can be automatically estimated using deep learning models from digital mammograms. Our aim is to evaluate the capacity and reliability of such models to predict density from low-dose mammograms taken to enable risk estimates for younger women. Approach: We trained deep learning models on standard-dose and simulated low-dose mammograms. The models were then tested on a mammography dataset with paired standard- and low-dose images. The effect of different factors (including age, density, and dose ratio) on the differences between predictions on standard and low doses is analyzed. Methods to improve performance are assessed, and factors that reduce the model quality are demonstrated. Results: We showed that, although many factors have no significant effect on the quality of low-dose density prediction, both density and breast area have an impact. The correlation between density predictions on low- and standard-dose images of breasts with the largest breast area is 0.985 (0.949 to 0.995), whereas that with the smallest is 0.882 (0.697 to 0.961). We also demonstrated that averaging across craniocaudal-mediolateral oblique (CC-MLO) images and across repeatedly trained models can improve predictive performance. Conclusions: Low-dose mammography can be used to produce density and risk estimates that are comparable to standard-dose images. Averaging across CC-MLO and model predictions should improve this performance. The model quality is reduced when making predictions on denser and smaller breasts.

An app promoting weight gain prevention via healthy behaviours amongst young women with a family history of breast cancer: Acceptability and usability assessment.

BACKGROUND: Breast cancer is the most frequent female malignancy in the UK. Around 20% of cases are linked to weight gain, excess weight and health behaviours. We designed a weight gain prevention, health behaviour intervention for young women at increased risk. METHODS: The study comprised a single arm observational study over 2 months testing acceptability and usability of the intervention: online group welcome event, app and private Facebook group. Females aged 18-35 years at moderate or high risk of breast cancer (>17% lifetime risk) were recruited via invite letters and social media posts. The app included behaviour change techniques and education content. Online questionnaires were completed at baseline, as well as at 1 and 2 months. We also assessed feasibility of study procedures. RESULTS: Both recruitment methods were successful. Thirty-five women were recruited, 26% via social media posts. Median age was 33 (interquartile range = 28.2-34.5) years, the majority (94.1%) were of White ethnicity. Thirty-four participants were included in the analyses, of which 94% downloaded the app. Median self-monitoring logs per participant during the study period was 10.0 (interquartile range = 4.8-28.8). App quality mean (SD) score was 3.7 (0.6) at 1 and 2 months (scale: 1-5). Eighty-nine per cent rated the app at average or above at 1 month and 75.0% at 2 months. Nineteen women (55.9%) joined the Facebook group and there were 61 comments and 83 reactions and votes from participants during the study period. CONCLUSIONS: This first iteration of the app and intervention was well received and is suitable to progress to the next stage of refining and further testing.

Breast cancer after ovarian cancer in BRCA1 and BRCA2 pathogenic variant heterozygotes: Lower rates for 5 years post chemotherapy.

PURPOSE: The identification of germline BRCA1/BRCA2 pathogenic variants (PV) infer high remaining lifetime breast/ovarian cancer risks, but there is paucity of studies assessing breast cancer risk after ovarian cancer diagnosis. METHODS: We reviewed the history of breast cancer in 895 PV heterozygotes (BRCA1 = 541). Cumulative annual breast cancer incidence was assessed at 2, 5, 10, and >10 years after ovarian cancer diagnosis date. RESULTS: Breast cancer annual rates were evaluated in 701 assessable women with no breast cancer at ovarian diagnosis (BRCA1 = 425). Incidence was lower at 2 years (1.18%) and 2 to 5 years (1.13%) but rose thereafter for BRCA1 with incidence post 10 years in excess of 4% annually. Breast cancer pathology in BRCA1 PV heterozygotes showed less high-grade triple-negative breast cancer and more lower-grade hormone-receptor-positive cancer than women with no prior ovarian cancer. In the prospective cohort from ovarian cancer diagnosis, <4% of all deaths were caused by breast cancer, although 50% of deaths in women with breast cancer after ovarian cancer diagnosis were due to breast cancer. CONCLUSION: Women can be reassured that incidence of breast cancer after ovarian cancer diagnosis is relatively low. It appears likely that this effect is due to platinum-based chemotherapy. Nonetheless women need to be aware that incidence increases thereafter, especially after 10 years.

Breast density prediction from low and standard dose mammograms using deep learning: effect of image resolution and model training approach on prediction quality.

Purpose. To improve breast cancer risk prediction for young women, we have developed deep learning methods to estimate mammographic density from low dose mammograms taken at approximately 1/10th of the usual dose. We investigate the quality and reliability of the density scores produced on low dose mammograms focussing on how image resolution and levels of training affect the low dose predictions.Methods. Deep learning models are developed and tested, with two feature extraction methods and an end-to-end trained method, on five different resolutions of 15,290 standard dose and simulated low dose mammograms with known labels. The models are further tested on a dataset with 296 matching standard and real low dose images allowing performance on the low dose images to be ascertained.Results. Prediction quality on standard and simulated low dose images compared to labels is similar for all equivalent model training and image resolution versions. Increasing resolution results in improved performance of both feature extraction methods for standard and simulated low dose images, while the trained models show high performance across the resolutions. For the trained models the Spearman rank correlation coefficient between predictions of standard and low dose images at low resolution is 0.951 (0.937 to 0.960) and at the highest resolution 0.956 (0.942 to 0.965). If pairs of model predictions are averaged, similarity increases.Conclusions. Deep learning mammographic density predictions on low dose mammograms are highly correlated with standard dose equivalents for feature extraction and end-to-end approaches across multiple image resolutions. Deep learning models can reliably make high quality mammographic density predictions on low dose mammograms.

A novel preclinical model of the normal human breast.

Improved screening and treatment have decreased breast cancer mortality, although incidence continues to rise. Women at increased risk of breast cancer can be offered risk reducing treatments, such as tamoxifen, but this has not been shown to reduce breast cancer mortality. New, more efficacious, risk-reducing agents are needed. The identification of novel candidates for prevention is hampered by a lack of good preclinical models. Current patient derived in vitro and in vivo models cannot fully recapitulate the complexities of the human tissue, lacking human extracellular matrix, stroma, and immune cells, all of which are known to influence therapy response. Here we describe a normal breast explant model utilising a tuneable hydrogel which maintains epithelial proliferation, hormone receptor expression, and residency of T cells and macrophages over 7 days. Unlike other organotypic tissue cultures which are often limited by hyper-proliferation, loss of hormone signalling, and short treatment windows (< 48h), our model shows that tissue remains viable over 7 days with none of these early changes. This offers a powerful and unique opportunity to model the normal breast and study changes in response to various risk factors, such as breast density and hormone exposure. Further validation of the model, using samples from patients undergoing preventive therapies, will hopefully confirm this to be a valuable tool, allowing us to test novel agents for breast cancer risk reduction preclinically.

Pathogenic variant detection rate varies considerably in male breast cancer families and sporadic cases: minimal additional contribution beyond BRCA2, BRCA1 and CHEK2.

BACKGROUND: Male breast cancer (MBC) affects around 1 in 1000 men and is known to have a higher underlying component of high and moderate risk gene pathogenic variants (PVs) than female breast cancer, particularly in BRCA2. However, most studies only report overall detection rates without assessing detailed family history. METHODS: We reviewed germline testing in 204 families including at least one MBC for BRCA1, BRCA2, CHEK2 c.1100DelC and an extended panel in 93 of these families. Individuals had MBC (n=118), female breast cancer (FBC)(n=80), ovarian cancer (n=3) or prostate cancer-(n=3). Prior probability of having a BRCA1/2 PV was assessed using the Manchester Scoring System (MSS). RESULTS: In the 204 families, BRCA2 was the major contributor, with 51 (25%) having PVs, followed by BRCA1 and CHEK2, with five each (2.45%) but no additional PVs identified, including in families with high genetic likelihood on MSS. Detection rates were 85.7% (12/14) in MSS ≥40 and 65.5% with MSS 30-39 but only 12.8% (6/47) for sporadic breast cancer. PV rates were low and divided equally between BRCA1/2 and CHEK2. CONCLUSION: As expected, BRCA2 PVs predominate in MBC families with rates 10-fold those in CHEK2 and BRCA1. The MSS is an effective tool in assessing the likelihood of BRCA1/2 PVs.

A pragmatic qualitative study to explore women's and clinicians' experience of access to systemic anti-cancer therapies for the treatment of secondary breast cancer.

PURPOSE: An estimated 57,000 women are currently living with secondary (metastatic) breast cancer across the UK. Equitable access to treatment has been associated with improved clinical outcomes, however geographical disparities have been reported which remain poorly understood. The purpose of our study was to explore women and clinicians' experience of geographic access to systemic anti-cancer therapies for the treatment of secondary breast cancer. METHOD: The study setting was the integrated cancer system across the northwest region of Greater Manchester UK. A pragmatic qualitative study design was used. Women aged >18 years with a confirmed SBC diagnosis and clinicians responsible for the care and treatment of women with a secondary breast cancer diagnosis were interviewed using semi structured interviews to elicit their experience and perspectives on geographic access to treatment. Data were analysed using thematic analysis to identify emergent themes. RESULTS: Eighteen interviews with women and 12 interviews with clinicians were completed. Four meta-themes were identified for geographic access, the influence of the health care system, person centred factors and the impact of Covid-19 on treatment access and receipt. CONCLUSION: Our study was the first of its kind to explore women and clinicians experience of geographic access to systemic anti-cancer therapies for the treatment of secondary breast cancer. Findings provided a greater understanding of distance decay and the influence of the health care system on treatment access. This included the importance and availability of clinical trials as a potential treatment option. This provided important insights and contributed to ongoing debate.

Tumor circadian clock strength influences metastatic potential and predicts patient prognosis in luminal A breast cancer.

Studies in shift workers and model organisms link circadian disruption to breast cancer. However, molecular circadian rhythms in noncancerous and cancerous human breast tissues and their clinical relevance are largely unknown. We reconstructed rhythms informatically, integrating locally collected, time-stamped biopsies with public datasets. For noncancerous breast tissue, inflammatory, epithelial-mesenchymal transition (EMT), and estrogen responsiveness pathways show circadian modulation. Among tumors, clock correlation analysis demonstrates subtype-specific changes in circadian organization. Luminal A organoids and informatic ordering of luminal A samples exhibit continued, albeit dampened and reprogrammed rhythms. However, CYCLOPS magnitude, a measure of global rhythm strength, varied widely among luminal A samples. Cycling of EMT pathway genes was markedly increased in high-magnitude luminal A tumors. Surprisingly, patients with high-magnitude tumors had reduced 5-y survival. Correspondingly, 3D luminal A cultures show reduced invasion following molecular clock disruption. This study links subtype-specific circadian disruption in breast cancer to EMT, metastatic potential, and prognosis.

Feasibility and acceptability of offering breast cancer risk assessment to general population women aged 30-39 years: a mixed-methods study protocol.

INTRODUCTION: Breast cancer incidence starts to increase exponentially when women reach 30-39 years, hence before they are eligible for breast cancer screening. The introduction of breast cancer risk assessment for this age group could lead to those at higher risk receiving benefits of earlier screening and preventive strategies. Currently, risk assessment is limited to women with a family history of breast cancer only. The Breast CANcer Risk Assessment in Younger women (BCAN-RAY) study is evaluating a comprehensive breast cancer risk assessment strategy for women aged 30-39 years incorporating a questionnaire of breast cancer risk factors, low-dose mammography to assess breast density and polygenic risk. This study will assess the feasibility and acceptability of the BCAN-RAY risk assessment strategy. METHODS AND ANALYSIS: This study involves women undergoing risk assessment as part of the BCAN-RAY case-control study (n=750). They will be aged 30-39 years without a strong family history of breast cancer and invited to participate via general practice. A comparison of uptake rates by socioeconomic status and ethnicity between women who participated in the BCAN-RAY study and women who declined participation will be conducted. All participants will be asked to complete self-report questionnaires to assess key potential harms including increased state anxiety (State Trait Anxiety Inventory), cancer worry (Lerman Cancer Worry Scale) and satisfaction with the decision to participate (Decision Regret Scale), alongside potential benefits such as feeling more informed about breast cancer risk. A subsample of approximately 24 women (12 at average risk and 12 at increased risk) will additionally participate in semistructured interviews to understand the acceptability of the risk assessment strategy and identify any changes needed to it to increase uptake. ETHICS AND DISSEMINATION: Ethical approval was granted by North West-Greater Manchester West Research Ethics Committee (reference: 22/NW/0268). Study results will be disseminated through peer-reviewed journals, conference presentations and charitable organisations. TRIAL REGISTRATION NUMBER: NCT05305963.

Geographic and sociodemographic access to systemic anticancer therapies for secondary breast cancer: a systematic review.

BACKGROUND: The review aimed to investigate geographic and sociodemographic factors associated with receipt of systemic anticancer therapies (SACT) for women with secondary (metastatic) breast cancer (SBC). METHODS: Included studies reported geographic and sociodemographic factors associated with receipt of treatment with SACT for women > 18 years with an SBC diagnosis. Information sources searched were Ovid CINAHL, Ovid MEDLINE, Ovid Embase and Ovid PsychINFO. Assessment of methodological quality was undertaken using the Joanna Briggs Institute method. Findings were synthesised using a narrative synthesis approach. RESULTS: Nineteen studies published between 2009 and 2023 were included in the review. Overall methodological quality was assessed as low to moderate. Outcomes were reported for treatment receipt and time to treatment. Overall treatment receipt ranged from 4% for immunotherapy treatment in one study to 83% for systemic anticancer therapies (unspecified). Time to treatment ranged from median 54 days to 95 days with 81% of patients who received treatment < 60 days. Younger women, women of White origin, and those women with a higher socioeconomic status had an increased likelihood of timely treatment receipt. Treatment receipt varied by geographical region, and place of care was associated with variation in timely receipt of treatment with women treated at teaching, research and private institutions being more likely to receive treatment in a timely manner. CONCLUSIONS: Treatment receipt varied depending upon type of SACT. A number of factors were associated with treatment receipt. Barriers included older age, non-White race, lower socioeconomic status, significant comorbidities, hospital setting and geographical location. Findings should however be interpreted with caution given the limitations in overall methodological quality of included studies and significant heterogeneity in measures of exposure and outcome. Generalisability was limited due to included study populations. Findings have practical implications for the development and piloting of targeted interventions to address specific barriers in a socioculturally sensitive manner. Addressing geographical variation and place of care may require intervention at a commissioning policy level. Further qualitative research is required to understand the experience and of women and clinicians. SYSTEMATIC REVIEW REGISTRATION: PROSPERO CRD42020196490.

Germline testing of BRCA1, BRCA2, PALB2 and CHEK2 c.1100delC in 1514 triple negative familial and isolated breast cancers from a single centre, with extended testing of ATM, RAD51C and RAD51D in over 400.

BACKGROUND: The identification of germline pathogenic gene variants (PGVs) in triple negative breast cancer (TNBC) is important to inform further primary cancer risk reduction and TNBC treatment strategies. We therefore investigated the contribution of breast cancer associated PGVs to familial and isolated invasive TNBC. METHODS: Outcomes of germline BRCA1, BRCA2 and CHEK2_c.1100delC testing were recorded in 1514 women (743-isolated, 771-familial), and for PALB2 in 846 women (541-isolated, 305-familial), with TNBC and smaller numbers for additional genes. Breast cancer free controls were identified from Predicting Risk Of Cancer At Screening and BRIDGES (Breast cancer RIsk after Diagnostic GEne Sequencing) studies. RESULTS: BRCA1_PGVs were detected in 52 isolated (7.0%) and 195 (25.3%) familial cases (isolated-OR=58.9, 95% CI: 16.6 to 247.0), BRCA2_PGVs in 21 (2.8%) isolated and 67 (8.7%) familial cases (isolated-OR=5.0, 95% CI: 2.3 to 11.2), PALB2_PGVs in 9 (1.7%) isolated and 12 (3.9%) familial cases (isolated-OR=8.8, 95% CI: 2.5 to 30.4) and CHEK2_c.1100delC in 0 isolated and 3 (0.45%) familial cases (isolated-OR=0.0, 95% CI: 0.00 to 2.11). BRCA1_PGV detection rate was >10% for all familial TNBC age groups and significantly higher for younger diagnoses (familial: <50 years, n=165/538 (30.7%); ≥50 years, n=30/233 (12.9%); p<0.0001). Women with a G3_TNBC were more likely to have a BRCA1_PGV as compared with a BRCA2 or PALB2_PGV (p<0.0001). 0/743 isolated TNBC had the CHEK2_c.1100delC PGV and 0/305 any ATM_PGV, but 2/240 (0.83%) had a RAD51D_PGV. CONCLUSION: PGVs in BRCA1 are associated with G3_TNBCs. Familial TNBCs and isolated TNBCs <30 years have a >10% likelihood of a PGV in BRCA1. BRCA1_PGVs are associated with younger age of familial TNBC. There was no evidence for any increased risk of TNBC with CHEK2 or ATM PGVs.

Artificial Intelligence for Image-Based Breast Cancer Risk Prediction Using Attention.

Accurate prediction of individual breast cancer risk paves the way for personalised prevention and early detection. The incorporation of genetic information and breast density has been shown to improve predictions for existing models, but detailed image-based features are yet to be included despite correlating with risk. Complex information can be extracted from mammograms using deep-learning algorithms, however, this is a challenging area of research, partly due to the lack of data within the field, and partly due to the computational burden. We propose an attention-based Multiple Instance Learning (MIL) model that can make accurate, short-term risk predictions from mammograms taken prior to the detection of cancer at full resolution. Current screen-detected cancers are mixed in with priors during model development to promote the detection of features associated with risk specifically and features associated with cancer formation, in addition to alleviating data scarcity issues. MAI-risk achieves an AUC of 0.747 [0.711, 0.783] in cancer-free screening mammograms of women who went on to develop a screen-detected or interval cancer between 5 and 55 months, outperforming both IBIS (AUC 0.594 [0.557, 0.633]) and VAS (AUC 0.649 [0.614, 0.683]) alone when accounting for established clinical risk factors.