Breast Cancer Risk Prediction Research Articles

Abstract 3458: DNA methylation-predicted blood protein levels and breast cancer risk

Abstract Background: Blood DNA methylation (DNAm) profiles can be used to predict a variety of human physiologic traits that are associated with breast cancer incidence and plasma concentrations of over 100 proteins, including those involved in inflammation, coagulation, metabolism, and growth. DNAm predictions of circulating proteins, here referred to as “DNAm proteins,” may be useful in providing molecular insights into breast cancer etiology and could be a valuable resource for personalizing breast cancer risk prediction. Methods: The Sister Study is a nationwide, prospective cohort of 50,884 women with a family history of breast cancer designed to identify novel environmental and biological risk factors for breast cancer. Epigenome-wide DNAm data were previously generated using the HumanMethylation450 or MethylationEPIC arrays on baseline blood samples collected from two case-cohort samples, for a combined sample of 4,479 women. This sample included 2,151 (48%) women who were selected because they were diagnosed with breast cancer (1,673 invasive cancer and 478 ductal carcinomas in situ) in the 15 years (mean: 7.9 years) following blood draw. Blood levels of 109 DNAm proteins were estimated from DNAm profiles using the method developed by Gadd et al. (eLife, 2022). Weighted Cox regression models were used to estimate associations between these DNAm proteins and incident breast cancer; associations were considered statistically significant at a False Discovery Rate (q) ≤ 0.10. Results: In age and platform adjusted models, higher breast cancer rates were associated with higher DNAm-predicted blood concentrations of RARRES2, IGFBP4, and CCL21 and lower predicted concentrations of F7, SELL, CXCL9, CD48, and IL19. The strongest associations were observed for RARRES2 (per 1-SD increase, HR: 1.24, 95% CI: 1.10, 1.40, P= 6 × 10−4, q= 0.03) and F7 (per 1-SD increase, HR: 0.82, 95% CI: 0.74, 0.91, P= 3 × 10−4, q= 0.03). Estimates were similar after adjustment for known breast cancer risk factors and excluding those diagnosed within two years of their blood draw. Associations appeared stronger for invasive breast cancer. Conclusions: Using novel DNAm-based estimates of blood proteins, we identified associations between circulating proteins and breast cancer incidence. IGFBP4 and RARRES2 have previously been reported to be higher in newly-diagnosed breast cancer patients. We also identify new potential protein biomarkers related to immune response (CCL21, SELL, CXCL9, CD48, and IL19). Citation Format: Jacob K. Kresovich, Brett Reid, Doratha A. Byrd, Katie M. O'Brien, Zongli Xu, Clarice R. Weinberg, Dale P. Sandler, Jack A. Taylor. DNA methylation-predicted blood protein levels and breast cancer risk [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 3458.

The Prognostic Quality of Risk Prediction Models to Assess the Individual Breast Cancer Risk in Women: An Overview of Reviews

Purpose. Breast cancer is the most common cancer among women globally, with an incidence of approximately two million cases in 2018. Organised age-based breast cancer screening programs were established worldwide to detect breast cancer earlier and to reduce mortality. Currently, there is substantial anticipation regarding risk-adjusted screening programs, considering various risk factors in addition to age. The present study investigated the discriminatory accuracy of breast cancer risk prediction models and whether they suit risk-based screening programs. Methods. Following the PICO scheme, we conducted an overview of reviews and systematically searched four databases. All methodological steps, including the literature selection, data extraction and synthesis, and the quality appraisal were conducted following the 4-eyes principle. For the quality assessment, the AMSTAR 2 tool was used. Results. We included eight systematic reviews out of 833 hits based on the prespecified inclusion criteria. The eight systematic reviews comprised ninety-nine primary studies that were also considered for the data analysis. Three systematic reviews were assessed as having a high risk of bias, while the others were rated with a moderate or low risk of bias. Most identified breast cancer risk prediction models showed a low prognostic quality. Adding breast density and genetic information as risk factors only moderately improved the models’ discriminatory accuracy. Conclusion. All breast cancer risk prediction models published to date show a limited ability to predict the individual breast cancer risk in women. Hence, it is too early to implement them in national breast cancer screening programs. Relevant randomised controlled trials about the benefit-harm ratio of risk-adjusted breast cancer screening programs compared to conventional age-based programs need to be awaited.

Polygenic risk score-based prediction of breast cancer risk in Taiwanese women with dense breast using a retrospective cohort study

Mammographic screening has contributed to a significant reduction in breast cancer mortality. Several studies have highlighted the correlation between breast density, as detected through mammography, and a higher likelihood of developing breast cancer. A polygenic risk score (PRS) is a numerical score that is calculated based on an individual's genetic information. This study aims to explore the potential roles of PRS as candidate markers for breast cancer development and investigate the genetic profiles associated with clinical characteristics in Asian females with dense breasts. This is a retrospective cohort study integrated breast cancer screening, population genotyping, and cancer registry database. The PRSs of the study cohort were estimated using genotyping data of 77 single nucleotide polymorphisms based on the PGS000001 Catalog. A subgroup analysis was conducted for females without breast symptoms. Breast cancer patients constituted a higher proportion of individuals in PRS Q4 (37.8% vs. 24.8% in controls). Among dense breast patients with no symptoms, the high PRS group (Q4) consistently showed a significantly elevated breast cancer risk compared to the low PRS group (Q1–Q3) in both univariate (OR = 2.25, 95% CI 1.43–3.50, P < 0.001) and multivariate analyses (OR: 2.23; 95% CI 1.41–3.48, P < 0.001). The study was extended to predict breast cancer risk using common low-penetrance risk variants in a PRS model, which could be integrated into personalized screening strategies for Taiwanese females with dense breasts without prominent symptoms.

Cigarette smoking and mammographic breast density in post-menopausal women from the EPIC Florence cohort.

Cigarette smoking has been recognized as a risk factor for breast cancer (BC) also if the biological mechanism remains poorly understood. High mammographic breast density (MBD) is associated with BC risk and many BC risk factors, such as genetic, anthropometric, reproductive and lifestyle factors and age, are also able to modulate MBD. The aim of the present study was to prospectively explore, in post-menopausal women, the association between smoking habits and MBD, assessed using an automated software, considering duration and intensity of smoking. The analysis was carried out in 3,774 women enrolled in the European Prospective Investigation into Cancer and Nutrition (EPIC) Florence cohort in 1993-98, participating in the 2004-06 follow up (FU) and with at least one full-field digital mammography (FFDM) performed after FU. For each woman, detailed information on smoking habits, anthropometry, lifestyle and reproductive history was collected at enrollment and at FU. Smoking information at baseline and at FU was integrated. The fully automated Volpara™ software was used to obtain total breast volume (cm3), absolute breast dense volume (DV, cm3) and volumetric percent density (VPD, %) from the first available FFDM (average 5.3 years from FU). Multivariable linear regression models were applied to evaluate the associations between smoking habits and VPD or DV. An inverse association between smoking exposure and VPD emerged (Diff% -7.96%, p <0.0001 for current smokers and -3.92%, p 0.01 for former smokers, compared with non-smokers). An inverse dose-response relationship with number of cigarettes/day, years of smoking duration and lifetime smoking exposure (pack-years) and a direct association with time since smoking cessation among former smokers emerged. Similar associations, with an attenuated effect, emerged when DV was considered as the outcome variable. This longitudinal study confirms the inverse association between active smoking, a known risk factor for BC, and MBD among post-menopausal women. The inclusion of smoking habits in the existing BC risk prediction models could be evaluated in future studies.

AsymMirai: Interpretable Mammography-based Deep Learning Model for 1-5-year Breast Cancer Risk Prediction.

Background Mirai, a state-of-the-art deep learning-based algorithm for predicting short-term breast cancer risk, outperforms standard clinical risk models. However, Mirai is a black box, risking overreliance on the algorithm and incorrect diagnoses. Purpose To identify whether bilateral dissimilarity underpins Mirai's reasoning process; create a simplified, intelligible model, AsymMirai, using bilateral dissimilarity; and determine if AsymMirai may approximate Mirai's performance in 1-5-year breast cancer risk prediction. Materials and Methods This retrospective study involved mammograms obtained from patients in the EMory BrEast imaging Dataset, known as EMBED, from January 2013 to December 2020. To approximate 1-5-year breast cancer risk predictions from Mirai, another deep learning-based model, AsymMirai, was built with an interpretable module: local bilateral dissimilarity (localized differences between left and right breast tissue). Pearson correlation coefficients were computed between the risk scores of Mirai and those of AsymMirai. Subgroup analysis was performed in patients for whom AsymMirai's year-over-year reasoning was consistent. AsymMirai and Mirai risk scores were compared using the area under the receiver operating characteristic curve (AUC), and 95% CIs were calculated using the DeLong method. Results Screening mammograms (n = 210 067) from 81 824 patients (mean age, 59.4 years ± 11.4 [SD]) were included in the study. Deep learning-extracted bilateral dissimilarity produced similar risk scores to those of Mirai (1-year risk prediction, r = 0.6832; 4-5-year prediction, r = 0.6988) and achieved similar performance as Mirai. For AsymMirai, the 1-year breast cancer risk AUC was 0.79 (95% CI: 0.73, 0.85) (Mirai, 0.84; 95% CI: 0.79, 0.89; P = .002), and the 5-year risk AUC was 0.66 (95% CI: 0.63, 0.69) (Mirai, 0.71; 95% CI: 0.68, 0.74; P < .001). In a subgroup of 183 patients for whom AsymMirai repeatedly highlighted the same tissue over time, AsymMirai achieved a 3-year AUC of 0.92 (95% CI: 0.86, 0.97). Conclusion Localized bilateral dissimilarity, an imaging marker for breast cancer risk, approximated the predictive power of Mirai and was a key to Mirai's reasoning. © RSNA, 2024 Supplemental material is available for this article See also the editorial by Freitas in this issue.

An advanced machine learning method for simultaneous breast cancer risk prediction and risk ranking in Chinese population: A prospective cohort and modeling study.

Breast cancer (BC) risk-stratification tools for Asian women that are highly accurate and can provide improved interpretation ability are lacking. We aimed to develop risk-stratification models to predict long- and short-term BC risk among Chinese women and to simultaneously rank potential non-experimental risk factors. The Breast Cancer Cohort Study in Chinese Women, a large ongoing prospective dynamic cohort study, includes 122,058 women aged 25-70 years old from the eastern part of China. We developed multiple machine-learning risk prediction models using parametric models (penalized logistic regression, bootstrap, and ensemble learning), which were the short-term ensemble penalized logistic regression (EPLR) risk prediction model and the ensemble penalized long-term (EPLT) risk prediction model to estimate BC risk. The models were assessed based on calibration and discrimination, and following this assessment, they were externally validated in new study participants from 2017 to 2020. The AUC values of the short-term EPLR risk prediction model were 0.800 for the internal validation and 0.751 for the external validation set. For the long-term EPLT risk prediction model, the area under the receiver operating characteristic curve was 0.692 and 0.760 in internal and external validations, respectively. The net reclassification improvement index of the EPLT relative to the Gail and the Han Chinese Breast Cancer Prediction Model (HCBCP) models for external validation was 0.193 and 0.233, respectively, indicating that the EPLT model has higher classification accuracy. We developed the EPLR and EPLT models to screen populations with a high risk of developing BC. These can serve as useful tools to aid in risk-stratified screening and BC prevention.

Early life body size and puberty markers as predictors of breast cancer risk later in life: A neural network analysis.

The early life factors of birthweight, child weight, height, body mass index (BMI) and pubertal timing are associated with risks of breast cancer. However, the predictive value of these factors in relation to breast cancer is largely unknown. Therefore, using a machine learning approach, we examined whether birthweight, childhood weights, heights, BMIs, and pubertal timing individually and in combination were predictive of breast cancer. We used information on birthweight, childhood height and weight, and pubertal timing assessed by the onset of the growth spurt (OGS) from 164,216 girls born 1930-1996 from the Copenhagen School Health Records Register. Of these, 10,002 women were diagnosed with breast cancer during 1977-2019 according to a nationwide breast cancer database. We developed a feed-forward neural network, which was trained and tested on early life body size measures individually and in various combinations. Evaluation metrics were examined to identify the best performing model. The highest area under the receiver operating curve (AUC) was achieved in a model that included birthweight, childhood heights, weights and age at OGS (AUC = 0.600). A model based on childhood heights and weights had a comparable AUC value (AUC = 0.598), whereas a model including only childhood heights had the lowest AUC value (AUC = 0.572). The sensitivity of the models ranged from 0.698 to 0.760 while the precision ranged from 0.071 to 0.076. We found that the best performing network was based on birthweight, childhood weights, heights and age at OGS as the input features. Nonetheless, this performance was only slightly better than the model including childhood heights and weights. Further, although the performance of our networks was relatively low, it was similar to those from previous studies including well-established risk factors. As such, our results suggest that childhood body size may add additional value to breast cancer prediction models.

Abstract B080: Advancing Breast Cancer Risk Prediction: Introducing the Gabbi Risk Assessment Model (GRAM)

Abstract Introduction: Advanced breast cancer risk models hold promise in implementing more effective and accurate screening strategies, facilitating early detection while minimizing screening-related harm compared to current guidelines. However, integrating machine learning risk models into clinical practice requires essential steps to be taken, including continual refinement to enhance their accuracy, validation across diverse populations, and demonstrating their ability to optimize clinical workflows. Gabbi Risk Assessment Model (GRAM), an innovative machine learning model, is a binary classifier based on medical claims developed to provide personalized breast cancer risk prediction. GRAM efficiently estimates risk and utilizes potentially missing risk factor information. Furthermore, it offers easy integration into EMR systems, thus optimizing its practical application. Methods: Construction of the GRAM utilized a medical claims dataset of 500k members from IQVIA carefully sampled from the entire IQVIA population (IQVIA acquires ~94% of the US claims data) with 4 years of history. The information extracted from the dataset included demographics, and medical history in the form of diagnostic and treatment codes, laboratory tests, and medications. The dataset consisted of 250k women aged 18-70 with a breast cancer diagnosis and a random sample of 250k women aged 18-70 without a known breast cancer diagnosis. For the selected cohorts, a 4-year history before the first breast cancer diagnosis was obtained. The control group was carefully chosen to possess similar statistical properties as the group with breast cancer. All women had a minimum enrollment period of 4 years before diagnosis with the same insurance provider. The GRAM was developed with 4 main steps: 1) member selection - members with breast cancer based on selection criteria defined above were in the test group, members without in the control group; 2) feature engineering to ensure that GRAM learns from medical history before a first diagnosis; 3) model development &amp; testing - training/fine tuning using the ensemble methods and neural network, followed by hyperparameter tuning using a held out validation dataset; and 4) model performance on the held out validation set and identification of key risk factors contributing to the breast cancer risk. Results and Conclusion: By addressing these challenges and harnessing the potential of machine learning, GRAM has the capacity to revolutionize breast cancer risk estimation. Its integration into clinical practice can lead to more targeted and efficient screening strategies, resulting in earlier detection and improved patient outcomes. Preliminary results have shown that the GRAM surpasses the current standard of care models (GAIL and Tyrer Cusick) by at least 20% in sensitivity, and AUC ROC with no decline in specificity. GRAM achieved 91% AUC ROC (balanced data), 86% sensitivity, and 80% specificity on a held out dataset. As research progresses, continuous refinement and validation, the GRAM model holds promise to make significant contributions in the fight against breast cancer and identifying the correlated risk factors. Citation Format: Priyanshi Jain, Shaleen K Theiler, Kaitlin Christine. Advancing Breast Cancer Risk Prediction: Introducing the Gabbi Risk Assessment Model (GRAM) [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Advances in Breast Cancer Research; 2023 Oct 19-22; San Diego, California. Philadelphia (PA): AACR; Cancer Res 2024;84(3 Suppl_1):Abstract nr B080.

Uncertainty in Breast Cancer Risk Prediction: A Conformal Prediction Study of Race Stratification.

The use of Artificial Intelligence (AI) in medicine has attracted a great deal of attention in the medical literature, but less is known about how to assess the uncertainty of individual predictions in clinical applications. This paper demonstrates the use of Conformal Prediction (CP) to provide insight on racial stratification of uncertainty quantification for breast cancer risk prediction. The results presented here show that CP methods provide important information about the diminished quality of predictions for individuals of minority racial backgrounds.

Risk Modeling of Time-Varying Covariates Using an Ensemble of Survival Trees: Predicting Future Cancer Events

The challenge of survival prediction is ubiquitous in medicine, but only a handful of methods are available for survival prediction based on time-varying data. Here we propose a novel method for this problem, using a random forest of survival trees for left-truncated and right-censored data. We demonstrate the advantage of our method on prediction of breast cancer and prostate gland cancer risk among healthy individuals by analyzing routine laboratory measurements, vital signs and age. We analyze electronic medical records of 20,317 healthy individuals who underwent routine checkups and identified those who later developed cancer. In cross-validation, our method predicted future prostate and breast cancers six months before diagnosis with an area under the ROC curve of 0.62±0.05 and 0.6±0.03 respectively, outperforming standard random forest, random survival forest, cox-regression model, dynamic deep-hit and a single survival tree. Our work proposes a new framework for survival risk prediction in time-varying data and our results suggest that computational analysis of data on healthy individuals can improve the detection of those at risk of future cancer development.

Association between Family History of Breast Cancer and Breast Density in Saudi Premenopausal Women Participating in Mammography Screening.

Mammographic density and family history of breast cancer (FHBC) are well-established independent factors affecting breast cancer risk; however, the association between these two risk factors in premenopausal-screened women remains unclear. The aim of this study is to investigate the relationship between mammographic density and FHBC among Saudi premenopausal-screened women. A total of 446 eligible participants were included in the study. Mammographic density was assessed qualitatively using the Breast Imaging Reporting and Data System (BIRADS 4th edition). Logistic regression models were built to investigate the relationship between mammographic density and FHBC. Women with a family history of breast cancer demonstrated an 87% greater chance of having dense tissue than women without a family history of breast cancer (95% CI: 1.14-3.08; p = 0.01). Having a positive family history for breast cancer in mothers was significantly associated with dense tissue (adjusted odds ratio (OR): 5.6; 95% CI: 1.3-24.1; p = 0.02). Dense breast tissue in Saudi premenopausal women undergoing screening may be linked to FHBC. If this conclusion is replicated in larger studies, then breast cancer risk prediction models must carefully consider these breast cancer risk factors.

Polygenic risk score model for renal cell carcinoma in the Korean population and relationship with lifestyle-associated factors

BackgroundThe polygenic risk score (PRS) is used to predict the risk of developing common complex diseases or cancers using genetic markers. Although PRS is used in clinical practice to predict breast cancer risk, it is more accurate for Europeans than for non-Europeans because of the sample size of training genome-wide association studies (GWAS). To address this disparity, we constructed a PRS model for predicting the risk of renal cell carcinoma (RCC) in the Korean population.ResultsUsing GWAS analysis, we identified 43 Korean-specific variants and calculated the PRS. Subsequent to plotting receiver operating characteristic (ROC) curves, we selected the 31 best-performing variants to construct an optimal PRS model. The resultant PRS model with 31 variants demonstrated a prediction rate of 77.4%. The pathway analysis indicated that the identified non-coding variants are involved in regulating the expression of genes related to cancer initiation and progression. Notably, favorable lifestyle habits, such as avoiding tobacco and alcohol, mitigated the risk of RCC across PRS strata expressing genetic risk.ConclusionA Korean-specific PRS model was established to predict the risk of RCC in the underrepresented Korean population. Our findings suggest that lifestyle-associated factors influencing RCC risk are associated with acquired risk factors indirectly through epigenetic modification, even among individuals in the higher PRS category.

Addition of polygenic risk score to a risk calculator for prediction of breast cancer in US Black women

BackgroundPrevious work in European ancestry populations has shown that adding a polygenic risk score (PRS) to breast cancer risk prediction models based on epidemiologic factors results in better discriminatory performance as measured by the AUC (area under the curve). Following publication of the first PRS to perform well in women of African ancestry (AA-PRS), we conducted an external validation of the AA-PRS and then evaluated the addition of the AA-PRS to a risk calculator for incident breast cancer in Black women based on epidemiologic factors (BWHS model).MethodsData from the Black Women’s Health Study, an ongoing prospective cohort study of 59,000 US Black women followed by biennial questionnaire since 1995, were used to calculate AUCs and 95% confidence intervals (CIs) for discriminatory accuracy of the BWHS model, the AA-PRS alone, and a new model that combined them. Analyses were based on data from 922 women with invasive breast cancer and 1844 age-matched controls.ResultsAUCs were 0.577 (95% CI 0.556–0.598) for the BWHS model and 0.584 (95% CI 0.563–0.605) for the AA-PRS. For a model that combined estimates from the questionnaire-based BWHS model with the PRS, the AUC increased to 0.623 (95% CI 0.603–0.644).ConclusionsThis combined model represents a step forward for personalized breast cancer preventive care for US Black women, as its performance metrics are similar to those from models in other populations. Use of this new model may mitigate exacerbation of breast cancer disparities if and when it becomes feasible to include a PRS in routine health care decision-making.

The impact of circulating protein levels identified by affinity proteomics on short-term, overall breast cancer risk

ObjectiveCurrent breast cancer risk prediction scores and algorithms can potentially be further improved by including molecular markers. To this end, we studied the association of circulating plasma proteins using Proximity Extension Assay (PEA) with incident breast cancer risk.SubjectsIn this study, we included 1577 women participating in the prospective KARMA mammographic screening cohort.ResultsIn a targeted panel of 164 proteins, we found 8 candidates nominally significantly associated with short-term breast cancer risk (P < 0.05). Similarly, in an exploratory panel consisting of 2204 proteins, 115 were found nominally significantly associated (P < 0.05). However, none of the identified protein levels remained significant after adjustment for multiple testing. This lack of statistically significant findings was not due to limited power, but attributable to the small effect sizes observed even for nominally significant proteins. Similarly, adding plasma protein levels to established risk factors did not improve breast cancer risk prediction accuracy.ConclusionsOur results indicate that the levels of the studied plasma proteins captured by the PEA method are unlikely to offer additional benefits for risk prediction of short-term overall breast cancer risk but could provide interesting insights into the biological basis of breast cancer in the future.

Association between allostatic load and breast cancer risk: a cohort study

BackgroundAllostatic load (AL) reflects the collective load of chronic stress during lifetime. Previous studies have shown that higher AL is associated with poor clinical outcomes among breast cancer patients. However, the relationship between AL and breast cancer risk is still unclear.MethodsTo fill the gap, we analyzed the association between AL and the development of breast cancer in 181,455 women identified from the UK Biobank.ResultsDuring the follow-up from 2006 to 2020, 5,701 women were diagnosed with incident breast cancer. Significantly higher AL was observed among incident breast cancer cases than all study participants (mean: 2.77 vs. 2.63, P < 0.01). Univariate Cox regression analysis indicated the risk of breast cancer was increased by 5% per one AL unit increase (hazard ratio (HR) = 1.05, 95% confidence interval (CI) 1.04, 1.07). In multivariate analyses, after adjusting demographics, family history of breast cancer, reproductive factors, socioeconomic status, lifestyle factors, and breast cancer polygenic risk score (PRS), the significant association remained (HR = 1.05, 95%CI 1.03, 1.07). The significant relationship was further confirmed in the categorical analysis. Compared with women in the low AL group (AL: 0 ~ 2), those in the high AL group (AL: 3 ~ 11) had a 1.17-fold increased risk of breast cancer (HR = 1.17, 95%CI 1.11, 1.24). Finally, in the stratified analysis, joint effects on the risk of breast cancer were observed between the AL and selected known breast cancer risk factors, including age, family history of breast cancer, PRS, income, physical activity, and alcohol consumption.ConclusionIn summary, those findings have demonstrated that higher AL was associated with an increased breast cancer risk in women. This association is likely independent of known breast cancer risk factors. Thus, the AL could be a valuable biomarker to help breast cancer risk prediction and stratification.

Computer-extracted global radiomic features can predict the radiologists' first impression about the abnormality of a screening mammogram.

Radiologists can detect the gist of abnormal based on their rapid initial impression on a mammogram (ie, global gist signal [GGS]). This study explores (1) whether global radiomic (ie, computer-extracted) features can predict the GGS; and if so, (ii) what features are the most important drivers of the signals. The GGS of cases in two extreme conditions was considered: when observers detect a very strong gist (high-gist) and when the gist of abnormal was not/poorly perceived (low-gist). Gist signals/scores from 13 observers reading 4191 craniocaudal mammograms were collected. As gist is a noisy signal, the gist scores from all observers were averaged and assigned to each image. The high-gist and low-gist categories contained all images in the fourth and first quartiles, respectively. One hundred thirty handcrafted global radiomic features (GRFs) per mammogram were extracted and utilized to construct eight separate machine learning random forest classifiers (All, Normal, Cancer, Prior-1, Prior-2, Missed, Prior-Visible, and Prior-Invisible) for characterizing high-gist from low-gist images. The models were trained and validated using the 10-fold cross-validation approach. The models' performances were evaluated by the area under receiver operating characteristic curve (AUC). Important features for each model were identified through a scree test. The Prior-Visible model achieved the highest AUC of 0.84 followed by the Prior-Invisible (0.83), Normal (0.82), Prior-1 (0.81), All (0.79), Prior-2 (0.77), Missed (0.75), and Cancer model (0.69). Cluster shade, standard deviation, skewness, kurtosis, and range were identified to be the most important features. Our findings suggest that GRFs can accurately classify high- from low-gist images. Global mammographic radiomic features can accurately predict high- from low-gist images with five features identified to be valuable in describing high-gist images. These are critical in providing better understanding of the mammographic image characteristics that drive the strength of the GGSs which could be exploited to advance breast cancer (BC) screening and risk prediction, enabling early detection and treatment of BC thereby further reducing BC-related deaths.

Comparison of machine learning models on breast cancer risk prediction challenge

Breast cancer is a malignant tumor that poses a serious risk to women's life and wellbeing. To make matters worse, this cancer is less symptomatic in its early stages. It is not easily diagnosed through traditional means. The topic of this essay is to investigate machine learning for the determination of breast cancer.. The methods based on machine learning are as followed: automated nuclear section segmentation model, BCRecommender System, DNNs, and computer-aided diagnosis model (CADM). The methods studied are all based on the BreCaHad dataset and use a comparative metric.To measure the performance of each model, the accuracy, F1-score, specificity, and precision are used. The result shows that the approaches based on machine learning work well in diagnosing breast carcinoma, with high accuracy. Most of them have a percentage over 90% in accuracy and some of them are even higher than 95%. However, some of the models work poorly, such as layer 1 of BCRecommender with 61.06% accuracy and EfficientNetB0 with 72.96%. With every aspect taken into consideration, computer-aided diagnosis (CADM: Using combined features of HOG, WPD, ResNet as well as PCA + SVM) has the greatest advantage in diagnosing BC.

Clinical implications of incorporating genetic and non-genetic risk factors in CanRisk-based breast cancer risk prediction

BackgroundBreast cancer (BC) risk prediction models consider cancer family history (FH) and germline pathogenic variants (PVs) in risk genes. It remains elusive to what extent complementation with polygenic risk score (PRS) and non-genetic risk factor (NGRFs) data affects individual intensified breast surveillance (IBS) recommendations according to European guidelines. MethodsFor 425 cancer-free women with cancer FH (mean age 40·6 years, range 21–74), recruited in France, Germany and the Netherlands, germline PV status, NGRFs, and a 306 variant-based PRS (PRS306) were assessed to calculate estimated lifetime risks (eLTR) and estimated 10-year risks (e10YR) using CanRisk. The proportions of women changing country-specific European risk categories for IBS recommendations, i.e. ≥20 % and ≥30 % eLTR, or ≥5 % e10YR were determined. FindingsOf the women with non-informative PV status, including PRS306 and NGRFs changed clinical recommendations for 31·0 %, (57/184, 20 % eLTR), 15·8 % (29/184, 30 % eLTR) and 22·4 % (41/183, 5 % e10YR), respectively whereas of the women tested negative for a PV observed in their family, clinical recommendations changed for 16·7 % (25/150), 1·3 % (2/150) and 9·5 % (14/147). No change was observed for 82 women with PVs in high-risk genes (BRCA1/2, PALB2). Combined consideration of eLTRs and e10YRs identified BRCA1/2 PV carriers benefitting from IBS <30 years, and women tested non-informative/negative for whom IBS may be postponed. InterpretationFor women who tested non-informative/negative, PRS and NGRFs have a considerable impact on IBS recommendations. Combined consideration of eLTRs and e10YRs allows personalizing IBS starting age. FundingHorizon 2020, German Cancer Aid, Federal Ministry of Education and Research, Köln Fortune.

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.

A Systematic Review and Critical Assessment of Breast Cancer Risk Prediction Tools Incorporating a Polygenic Risk Score for the General Population.

Single nucleotide polymorphisms (SNPs) in the form of a polygenic risk score (PRS) have emerged as a promising factor that could improve the predictive performance of breast cancer (BC) risk prediction tools. This study aims to appraise and critically assess the current evidence on these tools. Studies were identified using Medline, EMBASE and the Cochrane Library up to November 2022 and were included if they described the development and/ or validation of a BC risk prediction model using a PRS for women of the general population and if they reported a measure of predictive performance. We identified 37 articles, of which 29 combined genetic and non-genetic risk factors using seven different risk prediction tools. Most models (55.0%) were developed on populations from European ancestry and performed better than those developed on populations from other ancestry groups. Regardless of the number of SNPs in each PRS, models combining a PRS with genetic and non-genetic risk factors generally had better discriminatory accuracy (AUC from 0.52 to 0.77) than those using a PRS alone (AUC from 0.48 to 0.68). The overall risk of bias was considered low in most studies. BC risk prediction tools combining a PRS with genetic and non-genetic risk factors provided better discriminative accuracy than either used alone. Further studies are needed to cross-compare their clinical utility and readiness for implementation in public health practices.