Breast cancer risk stratification in women of screening age: Incremental effects of adding mammographic density, polygenic risk, and a gene panel.

PURPOSE: There is great promise in breast cancer risk stratification to target screening and prevention. It is unclear whether adding gene panels to other risk tools improves breast cancer risk stratification and adds discriminatory benefit on a population basis. METHODS: In total, 10,025 of 57,902 women aged 46 to 73 years in the Predicting Risk of Cancer at Screening study provided DNA samples. A case-control study was used to evaluate breast cancer risk assessment using polygenic risk scores (PRSs), cancer gene panel (n = 33), mammographic density (density residual [DR]), and risk factors collected using a self-completed 2-page questionnaire (Tyrer-Cuzick [TC] model version 8). In total, 525 cases and 1410 controls underwent gene panel testing and PRS calculation (18, 143, and/or 313 single-nucleotide polymorphisms [SNPs]). RESULTS: Actionable pathogenic variants (PGVs) in BRCA1/2 were found in 1.7% of cases and 0.55% of controls, and overall PGVs were found in 6.1% of cases and 1.3% of controls. A combined assessment of TC8-DR-SNP313 and gene panel provided the best risk stratification with 26.1% of controls and 9.7% of cases identified at <1.4% 10-year risk and 9.01% of controls and 23.3% of cases at ≥8% 10-year risk. Because actionable PGVs were uncommon, discrimination was identical with/without gene panel (with/without: area under the curve = 0.67, 95% CI = 0.64-0.70). Only 7 of 17 PGVs in cases resulted in actionable risk category change. Extended case (n = 644)-control (n = 1779) series with TC8-DR-SNP143 identified 18.9% of controls and only 6.4% of stage 2+ cases at <1.4% 10-year risk and 20.7% of controls and 47.9% of stage 2+ cases at ≥5% 10-year risk. CONCLUSION: Further studies and economic analysis will determine whether adding panels to PRS is a cost-effective strategy for risk stratification.

Mammographic density change in a cohort of premenopausal women receiving tamoxifen for breast cancer prevention over 5 years.

BACKGROUND: A decrease in breast density due to tamoxifen preventive therapy might indicate greater benefit from the drug. It is not known whether mammographic density continues to decline after 1 year of therapy, or whether measures of breast density change are sufficiently stable for personalised recommendations. METHODS: Mammographic density was measured annually over up to 5 years in premenopausal women with no previous diagnosis of breast cancer but at increased risk of breast cancer attending a family-history clinic in Manchester, UK (baseline 2010-2013). Tamoxifen (20 mg/day) for prevention was prescribed for up to 5 years in one group; the other group did not receive tamoxifen and were matched by age. Fully automatic methods were used on mammograms over the 5-year follow-up: three area-based measures (NN-VAS, Stratus, Densitas) and one volumetric (Volpara). Additionally, percentage breast density at baseline and first follow-up mammograms was measured visually. The size of density declines at the first follow-up mammogram and thereafter was estimated using a linear mixed model adjusted for age and body mass index. The stability of density change at 1 year was assessed by evaluating mean squared error loss from predictions based on individual or mean density change at 1 year. RESULTS: Analysis used mammograms from 126 healthy premenopausal women before and as they received tamoxifen for prevention (median age 42 years) and 172 matched controls (median age 41 years), with median 3 years follow-up. There was a strong correlation between percentage density measures used on the same mammogram in both the tamoxifen and no tamoxifen groups (all correlation coeficients > 0.8). Tamoxifen reduced mean breast density in year 1 by approximately 17-25% of the inter-quartile range of four automated percentage density measures at baseline, and from year 2, it decreased further by approximately 2-7% per year. Predicting change at 2 years using individual change at 1 year was approximately 60-300% worse than using mean change at 1year. CONCLUSIONS: All measures showed a consistent and large average tamoxifen-induced change in density over the first year, and a continued decline thereafter. However, these measures of density change at 1 year were not stable on an individual basis.

A case-control evaluation of 143 single nucleotide polymorphisms for breast cancer risk stratification with classical factors and mammographic density.

Panels of single nucleotide polymorphisms (SNPs) stratify risk for breast cancer in women from the general population, but studies are needed assess their use in a fully comprehensive model including classical risk factors, mammographic density and more than 100 SNPs associated with breast cancer. A case-control study was designed (1,668 controls, 405 cases) in women aged 47-73 years attending routine screening in Manchester UK, and enrolled in a wider study to assess methods for risk assessment. Risk from classical questionnaire risk factors was assessed using the Tyrer-Cuzick model; mean percentage visual mammographic density was scored by two independent readers. DNA extracted from saliva was genotyped at selected SNPs using the OncoArray. A predefined polygenic risk score based on 143 SNPs was calculated (SNP143). The odds ratio (OR, and 95% confidence interval, CI) per interquartile range (IQ-OR) of SNP143 was estimated unadjusted and adjusted for Tyrer-Cuzick and breast density. Secondary analysis assessed risk by oestrogen receptor (ER) status. The primary polygenic risk score was well calibrated (O/E OR 1.10, 95% CI 0.86-1.34) and accuracy was retained after adjustment for Tyrer-Cuzick risk and mammographic density (IQ-OR unadjusted 2.12, 95% CI% 1.75-2.42; adjusted 2.06, 95% CI 1.75-2.42). SNP143 was a risk factor for ER+ and ER- breast cancer (adjusted IQ-OR, ER+ 2.11, 95% CI 1.78-2.51; ER- 1.81, 95% CI 1.16-2.84). In conclusion, polygenic risk scores based on a large number of SNPs improve risk stratification in combination with classical risk factors and mammographic density, and SNP143 was similarly predictive for ER-positive and ER-negative disease.

Young adulthood body mass index, adult weight gain and breast cancer risk: the PROCAS Study (United Kingdom).

BACKGROUND: We tested the hypothesis that body mass index (BMI) aged 20 years modifies the association of adult weight gain and breast cancer risk. METHODS: We recruited women (aged 47-73 years) into the PROCAS (Predicting Risk Of Cancer At Screening; Manchester, UK: 2009-2013) Study. In 47,042 women, we determined BMI at baseline and (by recall) at age 20 years, and derived weight changes. We estimated hazard ratios (HRs) and 95% confidence intervals (CIs) for new breast cancer using Cox models and explored relationships between BMI aged 20 years, subsequent weight changes and breast cancer risk. RESULTS: With median follow-up of 5.6 years, 1142 breast cancers (post-menopausal at entry: 829) occurred. Among post-menopausal women at entry, BMI aged 20 years was inversely associated [HR per SD: 0.87 (95% CI: 0.79-0.95)], while absolute weight gain was associated with breast cancer [HR per SD:1.23 (95% CI: 1.14-1.32)]. For post-menopausal women who had a recall BMI aged 20 years <23.4 kg/m2 (75th percentile), absolute weight gain was associated with breast cancer [HR per SD: 1.31 (95% CIs: 1.21-1.42)], but there were no associations for women with a recall BMI aged 20 years of >23.4 kg/m2 (Pinteraction values <0.05). CONCLUSIONS: Adult weight gain increased post-menopausal breast cancer risk only among women who were <23.4 kg/m2 aged 20 years.

Breast cancer pathology and stage are better predicted by risk stratification models that include mammographic density and common genetic variants.

PURPOSE: To improve breast cancer risk stratification to enable more targeted early detection/prevention strategies that will better balance risks and benefits of population screening programmes. METHODS: 9362 of 57,902 women in the Predicting-Risk-Of-Cancer-At-Screening (PROCAS) study who were unaffected by breast cancer at study entry and provided DNA for a polygenic risk score (PRS). The PRS was analysed alongside mammographic density (density-residual-DR) and standard risk factors (Tyrer-Cuzick-model) to assess future risk of breast cancer based on tumour stage receptor expression and pathology. RESULTS: 195 prospective incident breast cancers had a prediction based on TC/DR/PRS which was informative for subsequent breast cancer overall [IQ-OR 2.25 (95% CI 1.89-2.68)] with excellent calibration-(0.99). The model performed particularly well in predicting higher stage stage 2+ IQ-OR 2.69 (95% CI 2.02-3.60) and ER + BCs (IQ-OR 2.36 (95% CI 1.93-2.89)). DR was most predictive for HER2+ and stage 2+ cancers but did not discriminate as well between poor and extremely good prognosis BC as either Tyrer-Cuzick or PRS. In contrast, PRS gave the highest OR for incident stage 2+ cancers, [IQR-OR 1.79 (95% CI 1.30-2.46)]. CONCLUSIONS: A combined approach using Tyrer-Cuzick/DR/PRS provides accurate risk stratification, particularly for poor prognosis cancers. This provides support for reducing the screening interval in high-risk women and increasing the screening interval in low-risk women defined by this model.

A comparison of five methods of measuring mammographic density: a case-control study.

BACKGROUND: High mammographic density is associated with both risk of cancers being missed at mammography, and increased risk of developing breast cancer. Stratification of breast cancer prevention and screening requires mammographic density measures predictive of cancer. This study compares five mammographic density measures to determine the association with subsequent diagnosis of breast cancer and the presence of breast cancer at screening. METHODS: Women participating in the "Predicting Risk Of Cancer At Screening" (PROCAS) study, a study of cancer risk, completed questionnaires to provide personal information to enable computation of the Tyrer-Cuzick risk score. Mammographic density was assessed by visual analogue scale (VAS), thresholding (Cumulus) and fully-automated methods (Densitas, Quantra, Volpara) in contralateral breasts of 366 women with unilateral breast cancer (cases) detected at screening on entry to the study (Cumulus 311/366) and in 338 women with cancer detected subsequently. Three controls per case were matched using age, body mass index category, hormone replacement therapy use and menopausal status. Odds ratios (OR) between the highest and lowest quintile, based on the density distribution in controls, for each density measure were estimated by conditional logistic regression, adjusting for classic risk factors. RESULTS: The strongest predictor of screen-detected cancer at study entry was VAS, OR 4.37 (95% CI 2.72-7.03) in the highest vs lowest quintile of percent density after adjustment for classical risk factors. Volpara, Densitas and Cumulus gave ORs for the highest vs lowest quintile of 2.42 (95% CI 1.56-3.78), 2.17 (95% CI 1.41-3.33) and 2.12 (95% CI 1.30-3.45), respectively. Quantra was not significantly associated with breast cancer (OR 1.02, 95% CI 0.67-1.54). Similar results were found for subsequent cancers, with ORs of 4.48 (95% CI 2.79-7.18), 2.87 (95% CI 1.77-4.64) and 2.34 (95% CI 1.50-3.68) in highest vs lowest quintiles of VAS, Volpara and Densitas, respectively. Quantra gave an OR in the highest vs lowest quintile of 1.32 (95% CI 0.85-2.05). CONCLUSIONS: Visual density assessment demonstrated a strong relationship with cancer, despite known inter-observer variability; however, it is impractical for population-based screening. Percentage density measured by Volpara and Densitas also had a strong association with breast cancer risk, amongst the automated measures evaluated, providing practical automated methods for risk stratification.

Does Reader Performance with Digital Breast Tomosynthesis Vary according to Experience with Two-dimensional Mammography?

Purpose To assess whether individual reader performance with digital breast tomosynthesis (DBT) and two-dimensional (2D) mammography varies with number of years of experience or volume of 2D mammograms read. Materials and Methods After written informed consent was obtained, 8869 women (age range, 29-85 years; mean age, 56 years) were recruited into the TOMMY trial (A Comparison of Tomosynthesis with Digital Mammography in the UK National Health Service Breast Screening Program), an ethically approved, multicenter, multireader, retrospective reading study, between July 2011 and March 2013. Each case was read prospectively for clinical assessment and to establish ground truth. A retrospective reading data set of 7060 cases was created and randomly allocated for independent blinded review of (a) 2D mammograms, (b) DBT images and 2D mammograms, and (c) synthetic 2D mammograms and DBT images, without access to previous examinations. Readers (19 radiologists, three advanced practitioner radiographers, and two breast clinicians) who had 3-25 (median, 10) years of experience in the U.K. National Health Service Breast Screening Program and read 5000-13 000 (median, 8000) cases per annum were included in this study. Specificity was analyzed according to reader type and years and volume of experience, and then both specificity and sensitivity were analyzed by matched inference. The median duration of experience (10 years) was used as the cutoff point for comparison of reader performance. Results Specificity improved with the addition of DBT for all readers. This was significant for all staff groups (56% vs 68% and 49% vs 67% [P < .0001] for radiologists and advanced practitioner radiographers, respectively; 46% vs 55% [P = .02] for breast clinicians). Sensitivity was improved for 19 of 24 (79%) readers and was significantly higher for those with less than 10 years of experience (91% vs 86%; P = .03) and those with total mammographic experience of fewer than 80 000 cases (88% vs 86%; P = .03). Conclusion The addition of DBT to conventional 2D screening mammography improved specificity for all readers, but the gain in sensitivity was greater for readers with less than 10 years of experience.

Increased peri-ductal collagen micro-organization may contribute to raised mammographic density.

BACKGROUND: High mammographic density is a therapeutically modifiable risk factor for breast cancer. Although mammographic density is correlated with the relative abundance of collagen-rich fibroglandular tissue, the causative mechanisms, associated structural remodelling and mechanical consequences remain poorly defined. In this study we have developed a new collaborative bedside-to-bench workflow to determine the relationship between mammographic density, collagen abundance and alignment, tissue stiffness and the expression of extracellular matrix organising proteins. METHODS: Mammographic density was assessed in 22 post-menopausal women (aged 54-66 y). A radiologist and a pathologist identified and excised regions of elevated non-cancerous X-ray density prior to laboratory characterization. Collagen abundance was determined by both Masson's trichrome and Picrosirius red staining (which enhances collagen birefringence when viewed under polarised light). The structural specificity of these collagen visualisation methods was determined by comparing the relative birefringence and ultrastructure (visualised by atomic force microscopy) of unaligned collagen I fibrils in reconstituted gels with the highly aligned collagen fibrils in rat tail tendon. Localised collagen fibril organisation and stiffness was also evaluated in tissue sections by atomic force microscopy/spectroscopy and the abundance of key extracellular proteins was assessed using mass spectrometry. RESULTS: Mammographic density was positively correlated with the abundance of aligned periductal fibrils rather than with the abundance of amorphous collagen. Compared with matched tissue resected from the breasts of low mammographic density patients, the highly birefringent tissue in mammographically dense breasts was both significantly stiffer and characterised by large (>80 µm long) fibrillar collagen bundles. Subsequent proteomic analyses not only confirmed the absence of collagen fibrosis in high mammographic density tissue, but additionally identified the up-regulation of periostin and collagen XVI (regulators of collagen fibril structure and architecture) as potential mediators of localised mechanical stiffness. CONCLUSIONS: These preliminary data suggest that remodelling, and hence stiffening, of the existing stromal collagen microarchitecture promotes high mammographic density within the breast. In turn, this aberrant mechanical environment may trigger neoplasia-associated mechanotransduction pathways within the epithelial cell population.

Breast cancer risk feedback to women in the UK NHS breast screening population.

INTRODUCTION: There are widespread moves to develop risk-stratified approaches to population-based breast screening. The public needs to favour receiving breast cancer risk information, which ideally should produce no detrimental effects. This study investigates risk perception, the proportion wishing to know their 10-year risk and whether subsequent screening attendance is affected. METHODS: Fifty thousand women attending the NHS Breast Screening Programme completed a risk assessment questionnaire. Ten-year breast cancer risks were estimated using a validated algorithm (Tyrer-Cuzick) adjusted for visually assessed mammographic density. Women at high risk (⩾8%) and low risk (<1%) were invited for face-to-face or telephone risk feedback and counselling. RESULTS: Of those invited to receive risk feedback, more high-risk women, 500 out of 673 (74.3%), opted to receive a consultation than low-risk women, 106 out of 193 (54.9%) (P<0.001). Women at high risk were significantly more likely to perceive their risk as high (P<0.001) and to attend their subsequent mammogram (94.4%) compared with low-risk women (84.2%; P=0.04) and all attendees (84.3%; ⩽0.0001). CONCLUSIONS: Population-based assessment of breast cancer risk is feasible. The majority of women wished to receive risk information. Perception of general population breast cancer risk is poor. There were no apparent adverse effects on screening attendance for high-risk women whose subsequent screening attendance was increased.

Evaluation of Kinect 3D Sensor for Healthcare Imaging.

Microsoft Kinect is a three-dimensional (3D) sensor originally designed for gaming that has received growing interest as a cost-effective and safe device for healthcare imaging. Recent applications of Kinect in health monitoring, screening, rehabilitation, assistance systems, and intervention support are reviewed here. The suitability of available technologies for healthcare imaging applications is assessed. The performance of Kinect I, based on structured light technology, is compared with that of the more recent Kinect II, which uses time-of-flight measurement, under conditions relevant to healthcare applications. The accuracy, precision, and resolution of 3D images generated with Kinect I and Kinect II are evaluated using flat cardboard models representing different skin colors (pale, medium, and dark) at distances ranging from 0.5 to 1.2 m and measurement angles of up to 75°. Both sensors demonstrated high accuracy (majority of measurements <2 mm) and precision (mean point to plane error <2 mm) at an average resolution of at least 390 points per cm2. Kinect I is capable of imaging at shorter measurement distances, but Kinect II enables structures angled at over 60° to be evaluated. Kinect II showed significantly higher precision and Kinect I showed significantly higher resolution (both p < 0.001). The choice of object color can influence measurement range and precision. Although Kinect is not a medical imaging device, both sensor generations show performance adequate for a range of healthcare imaging applications. Kinect I is more appropriate for short-range imaging and Kinect II is more appropriate for imaging highly curved surfaces such as the face or breast.

Mammographic density adds accuracy to both the Tyrer-Cuzick and Gail breast cancer risk models in a prospective UK screening cohort.

INTRODUCTION: The Predicting Risk of Cancer at Screening study in Manchester, UK, is a prospective study of breast cancer risk estimation. It was designed to assess whether mammographic density may help in refinement of breast cancer risk estimation using either the Gail model (Breast Cancer Risk Assessment Tool) or the Tyrer-Cuzick model (International Breast Intervention Study model). METHODS: Mammographic density was measured at entry as a percentage visual assessment, adjusted for age and body mass index. Tyrer-Cuzick and Gail 10-year risks were based on a questionnaire completed contemporaneously. Breast cancers were identified at the entry screen or shortly thereafter. The contribution of density to risk models was assessed using odds ratios (ORs) with profile likelihood confidence intervals (CIs) and area under the receiver operating characteristic curve (AUC). The calibration of predicted ORs was estimated as a percentage [(observed vs expected (O/E)] from logistic regression. RESULTS: The analysis included 50,628 women aged 47-73 years who were recruited between October 2009 and September 2013. Of these, 697 had breast cancer diagnosed after enrolment. Median follow-up was 3.2 years. Breast density [interquartile range odds ratio (IQR-OR) 1.48, 95 % CI 1.34-1.63, AUC 0.59] was a slightly stronger univariate risk factor than the Tyrer-Cuzick model [IQR-OR 1.36 (95 % CI 1.25-1.48), O/E 60 % (95 % CI 44-74), AUC 0.57] or the Gail model [IQR-OR 1.22 (95 % CI 1.12-1.33), O/E 46 % (95 % CI 26-65 %), AUC 0.55]. It continued to add information after allowing for Tyrer-Cuzick [IQR-OR 1.47 (95 % CI 1.33-1.62), combined AUC 0.61] or Gail [IQR-OR 1.45 (95 % CI 1.32-1.60), combined AUC 0.59]. CONCLUSIONS: Breast density may be usefully combined with the Tyrer-Cuzick model or the Gail model.

Accuracy of Digital Breast Tomosynthesis for Depicting Breast Cancer Subgroups in a UK Retrospective Reading Study (TOMMY Trial).

PURPOSE: To compare the diagnostic performance of two-dimensional (2D) mammography, 2D mammography plus digital breast tomosynthesis (DBT), and synthetic 2D mammography plus DBT in depicting malignant radiographic features. MATERIALS AND METHODS: In this multicenter, multireader, retrospective reading study (the TOMMY trial), after written informed consent was obtained, 8869 women (age range, 29-85 years; mean, 56 years) were recruited from July 2011 to March 2013 in an ethically approved study. From these women, a reading dataset of 7060 cases was randomly allocated for independent blinded review of (a) 2D mammography images, (b) 2D mammography plus DBT images, and (c) synthetic 2D mammography plus DBT images. Reviewers had no access to results of previous examinations. Overall sensitivities and specificities were calculated for younger women and those with dense breasts. RESULTS: Overall sensitivity was 87% for 2D mammography, 89% for 2D mammography plus DBT, and 88% for synthetic 2D mammography plus DBT. The addition of DBT was associated with a 34% increase in the odds of depicting cancer (odds ratio [OR] = 1.34, P = .06); however, this level did not achieve significance. For patients aged 50-59 years old, sensitivity was significantly higher (P = .01) for 2D mammography plus DBT than it was for 2D mammography. For those with breast density of 50% or more, sensitivity was 86% for 2D mammography compared with 93% for 2D mammography plus DBT (P = .03). Specificity was 57% for 2D mammography, 70% for 2D mammography plus DBT, and 72% for synthetic 2D mammography plusmDBT. Specificity was significantly higher than 2D mammography (P < .001in both cases) and was observed for all subgroups (P < .001 for all cases). CONCLUSION: The addition of DBT increased the sensitivity of 2D mammography in patients with dense breasts and the specificity of 2D mammography for all subgroups. The use of synthetic 2D DBT demonstrated performance similar to that of standard 2D mammography with DBT. DBT is of potential benefit to screening programs, particularly in younger women with dense breasts. (©) RSNA, 2015.

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.

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.

Automatic assessment of mammographic density using a deep transfer learning method.

Purpose: Mammographic breast density is one of the strongest risk factors for cancer. Density assessed by radiologists using visual analogue scales has been shown to provide better risk predictions than other methods. Our purpose is to build automated models using deep learning and train on radiologist scores to make accurate and consistent predictions. Approach: We used a dataset of almost 160,000 mammograms, each with two independent density scores made by expert medical practitioners. We used two pretrained deep networks and adapted them to produce feature vectors, which were then used for both linear and nonlinear regression to make density predictions. We also simulated an "optimal method," which allowed us to compare the quality of our results with a simulated upper bound on performance. Results: Our deep learning method produced estimates with a root mean squared error (RMSE) of 8.79 ± 0.21 . The model estimates of cancer risk perform at a similar level to human experts, within uncertainty bounds. We made comparisons between different model variants and demonstrated the high level of consistency of the model predictions. Our modeled "optimal method" produced image predictions with a RMSE of between 7.98 and 8.90 for cranial caudal images. Conclusion: We demonstrated a deep learning framework based upon a transfer learning approach to make density estimates based on radiologists' visual scores. Our approach requires modest computational resources and has the potential to be trained with limited quantities of data.

The effect of variable labels on deep learning models trained to predict breast density.

Purpose. High breast density is associated with reduced efficacy of mammographic screening and increased risk of developing breast cancer. Accurate and reliable automated density estimates can be used for direct risk prediction and passing density related information to further predictive models. Expert reader assessments of density show a strong relationship to cancer risk but also inter-reader variation. The effect of label variability on model performance is important when considering how to utilise automated methods for both research and clinical purposes.Methods. We utilise subsets of images with density labels from the same 13 readers and 12 reader pairs, and train a deep transfer learning model which is used to assess how label variability affects the mapping from representation to prediction. We then create two end-to-end models: one that is trained on averaged labels across the reader pairs and the second that is trained using individual reader scores, with a novel alteration to the objective function. The combination of these two end-to-end models allows us to investigate the effect of label variability on the model representation formed.Results. We show that the trained mappings from representations to labels are altered considerably by the variability of reader scores. Training on labels with distribution variation removed causes the Spearman rank correlation coefficients to rise from 0.751 ± 0.002 to either 0.815 ± 0.026 when averaging across readers or 0.844 ± 0.002 when averaging across images. However, when we train different models to investigate the representation effect we see little difference, with Spearman rank correlation coefficients of 0.846 ± 0.006 and 0.850 ± 0.006 showing no statistically significant difference in the quality of the model representation with regard to density prediction.Conclusions. We show that the mapping between representation and mammographic density prediction is significantly affected by label variability. However, the effect of the label variability on the model representation is limited.

Mammographic breast density and breast cancer risk in the Saudi population: a case-control study using visual and automated methods.

OBJECTIVE: This study aims to establish risk of breast cancer based on breast density among Saudi women and to compare cancer prediction using different breast density methods. METHODS: 1140 pseudonymised screening mammograms from Saudi females were retrospectively collected. Breast density was assessed using Breast Imaging Reporting and Data System (BI-RADS) density categories and visual analogue scale (VAS) of 285 cases and 855 controls matched on age and body mass index. In a subset of 160 cases and 480 controls density was estimated by two automated methods, Volpara Density™ and predicted VAS (pVAS). Odds ratios (ORs) between the highest and second categories in BI-RADS and Volpara density grades, and highest vs lowest quartiles in VAS, pVAS and Volpara Density™, were estimated using conditional logistic regression. RESULTS: For BI-RADS, the OR was 6.69 (95% CI 2.79-16.06) in the highest vs second category and OR = 4.78 (95% CI 3.01-7.58) in the highest vs lowest quartile for VAS. In the subset, VAS was the strongest predictor OR = 7.54 (95% CI 3.86-14.74), followed by pVAS using raw images OR = 5.38 (95% CI 2.68-10.77) and Volpara Density ™ OR = 3.55, (95% CI 1.86-6.75) for highest vs lowest quartiles. The matched concordance index for VAS was 0.70 (95% CI 0.65-0.75) demonstrating better discrimination between cases and controls than all other methods. CONCLUSION: Increased mammographic density was strongly associated with risk of breast cancer among Saudi women. Radiologists' visual assessment of breast density is superior to automated methods. However, pVAS and Volpara Density ™ also significantly predicted breast cancer risk based on breast density. ADVANCES IN KNOWLEDGE: Our study established an association between breast density and breast cancer in a Saudi population and compared the performance of automated methods. This provides a stepping-stone towards personalised screening using automated breast density methods.

Single reading with computer-aided detection for screening mammography.

BACKGROUND: The sensitivity of screening mammography for the detection of small breast cancers is higher when the mammogram is read by two readers rather than by a single reader. We conducted a trial to determine whether the performance of a single reader using a computer-aided detection system would match the performance achieved by two readers. METHODS: The trial was designed as an equivalence trial, with matched-pair comparisons between the cancer-detection rates achieved by single reading with computer-aided detection and those achieved by double reading. We randomly assigned 31,057 women undergoing routine screening by film mammography at three centers in England to double reading, single reading with computer-aided detection, or both double reading and single reading with computer-aided detection, at a ratio of 1:1:28. The primary outcome measures were the proportion of cancers detected according to regimen and the recall rates within the group receiving both reading regimens. RESULTS: The proportion of cancers detected was 199 of 227 (87.7%) for double reading and 198 of 227 (87.2%) for single reading with computer-aided detection (P=0.89). The overall recall rates were 3.4% for double reading and 3.9% for single reading with computer-aided detection; the difference between the rates was small but significant (P<0.001). The estimated sensitivity, specificity, and positive predictive value for single reading with computer-aided detection were 87.2%, 96.9%, and 18.0%, respectively. The corresponding values for double reading were 87.7%, 97.4%, and 21.1%. There were no significant differences between the pathological attributes of tumors detected by single reading with computer-aided detection alone and those of tumors detected by double reading alone. CONCLUSIONS: Single reading with computer-aided detection could be an alternative to double reading and could improve the rate of detection of cancer from screening mammograms read by a single reader. (ClinicalTrials.gov number, NCT00450359.)

Mammographic features of breast cancers at single reading with computer-aided detection and at double reading in a large multicenter prospective trial of computer-aided detection: CADET II.

PURPOSE: To evaluate the mammographic features of breast cancer that favor lesion detection with single reading and computer-aided detection (CAD) or with double reading. MATERIALS AND METHODS: The Computer Aided Detection Evaluation Trial II study was approved by the ethics committee, and all participants provided written informed consent. A total of 31,057 women were recruited from three screening centers between September 2006 and August 2007. They were randomly allocated to the double reading group, the single reading with CAD group, or the double reading and single reading with CAD group at a ratio of 1:1:28, respectively. In this study, cancers in the women whose mammograms were read with both single reading with CAD and double reading were retrospectively reviewed. The original mammograms were obtained for each case and reviewed by two of three experienced breast radiologists in consensus. The method of detection was noted. The size and predominant mammographic feature of the cancer were recorded, as was the breast density. CAD marking data were reviewed to determine if the cancer had been correctly marked. RESULTS: A total of 227 cancers were detected in 28,204 women. A total of 170 cases were recalled with both reading regimens. Lesion types were masses (66%), microcalcifications (25%), parenchymal deformities (6%), and asymmetric densities (3%). The ability of the reading regimens to correctly prompt the reader to recall cases varied significantly by lesion type (P < .001). More parenchymal deformities were recalled with double reading, whereas more asymmetric densities were recalled with single reading with CAD. There was no difference in the ability of either reading regimen to prompt the reader to correctly recall masses or microcalcifications. CAD correctly prompted 100% of microcalcifications, 87% of mass lesions, 80% of asymmetric densities, and 50% of parenchymal deformities. CAD correctly marked 93% of spiculated masses compared with 80% of ill-defined masses (P = .054). There was a significant trend for cancers detected with double reading to occur only in women with a denser mammographic background pattern (P = .02). Size had no effect on lesion detection. CONCLUSION: Readers using either single reading with CAD or double reading need to be aware of the strengths and weaknesses of reading regimens to avoid missing the more challenging cancer cases.

Is there a safety-net effect with computer-aided detection?

Computer-aided detection (CAD) systems are used to aid readers interpreting screening mammograms. An expert reader searches the image initially unaided and then once again with the aid of CAD, which prompts automatically detected suspicious regions. This could lead to a "safety-net" effect, where the initial unaided search of the image is adversely affected by the fact that it is preliminary to an additional search with CAD and may, therefore, be less thorough. To investigate the existence of such an effect, we created a visual search experiment for nonexpert observers mirroring breast screening with CAD. Each observer searched 100 images for microcalcification clusters within synthetic images in both prompted (CAD) and unprompted (no-CAD) conditions. Fifty-two participants were recruited for the study, 48 of whom had their eye movements tracked in real-time; the other 4 participants could not be accurately calibrated, so only behavioral data were collected. In the CAD condition, before prompts were displayed, image coverage was significantly lower than coverage in the no-CAD condition ( t 47 = 5.29 , p < 0.0001 ). Observer sensitivity was significantly greater for targets marked by CAD than the same targets in the no-CAD condition ( t 51 = 6.56 , p < 0.001 ). For targets not marked by CAD, there was no significant difference in observer sensitivity in the CAD condition compared with the same targets in the no-CAD condition ( t 51 = 0.54 , p = 0.59 ). These results suggest that the initial search may be influenced by the subsequent availability of CAD; if so, cross-sectional CAD efficacy studies should account for the effect when estimating benefit.