Promoting Artificial Intelligence for Global Breast Cancer Risk Prediction and Screening in Adult Women: A Scoping Review

Background: Artificial intelligence (AI) algorithms can be applied in breast cancer risk prediction and prevention by using patient history, scans, imaging information, and analysis of specific genes for cancer classification to reduce overdiagnosis and overtreatment. This scoping review aimed to identify the barriers encountered in applying innovative AI techniques and models in developing breast cancer risk prediction scores and promoting screening behaviors among adult females. Findings may inform and guide future global recommendations for AI application in breast cancer prevention and care for female populations. Methods: The PRISMA-SCR (Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews) was used as a reference checklist throughout this study. The Arksey and O’Malley methodology was used as a framework to guide this review. The framework methodology consisted of five steps: (1) Identify research questions; (2) Search for relevant studies; (3) Selection of studies relevant to the research questions; (4) Chart the data; (5) Collate, summarize, and report the results. Results: In the field of breast cancer risk detection and prevention, the following AI techniques and models have been applied: Machine and Deep Learning Model (ML-DL model) (n = 1), Academic Algorithms (n = 2), Breast Cancer Surveillance Consortium (BCSC), Clinical 5-Year Risk Prediction Model (n = 2), deep-learning computer vision AI algorithms (n = 2), AI-based thermal imaging solution (Thermalytix) (n = 1), RealRisks (n = 2), Breast Cancer Risk NAVIgation (n = 1), MammoRisk (ML-Based Tool) (n = 1), Various MLModels (n = 1), and various machine/deep learning, decision aids, and commercial algorithms (n = 7). In the 11 included studies, a total of 39 barriers to AI applications in breast cancer risk prediction and screening efforts were identified. The most common barriers in the application of innovative AI tools for breast cancer prediction and improved screening rates included lack of external validity and limited generalizability (n = 6), as AI was used in studies with either a small sample size or datasets with missing data. Many studies (n = 5) also encountered selection bias due to exclusion of certain populations based on characteristics such as race/ethnicity, family history, or past medical history. Several recommendations for future research should be considered. AI models need to include a broader spectrum and more complete predictive variables for risk assessment. Investigating long-term outcomes with improved follow-up periods is critical to assess the impacts of AI on clinical decisions beyond just the immediate outcomes. Utilizing AI to improve communication strategies at both a local and organizational level can assist in informed decision-making and compliance, especially in populations with limited literacy levels. Conclusions: The use of AI in patient education and as an adjunctive tool for providers is still early in its incorporation, and future research should explore the implementation of AI-driven resources to enhance understanding and decision-making regarding breast cancer screening, especially in vulnerable populations with limited literacy.


Background
Breast cancer incidence rates among women have slowly increased per year by 0.5% [1,2].It is the most diagnosed cancer worldwide, surpassing even lung cancer, accounting for 31% of estimated newly diagnosed cancer cases and 15% of estimated deaths [2,3].In 2020, breast cancer accounted for an estimated 2.3 million cases and 685,000 deaths [3].Mainly, breast cancer has posed significant global health challenges, with notable disparities in survival rates among socioeconomically disadvantaged women [4,5].The incidence rates vary widely among countries, with developed nations like the UK and the USA witnessing high rates due in part to an increased prevalence of risk factors and "more extensive use of mammography screening since the 1980s" [3].Disparities in health outcomes further complicate the global burden of cancer.For instance, non-Hispanic Black women have a higher mortality rate regarding breast cancer compared to their non-Hispanic White counterparts, and this effect might be more pronounced due to specific social determinants of health such as race, socioeconomic status, and healthcare access [6,7].The situation is exacerbated in developing countries, where globalization and economic growth are predicted to significantly increase breast cancer incidence by 2040 [8].In India, urban areas report the highest incidence in the 40-49 age group, contrasting with rural areas where the peak is between 65 and 69 years [8].
Screening remains a pivotal strategy in early cancer detection [3].The WHO and the American Cancer Society have set guidelines for mammography-based screenings, emphasizing their importance for women in specific age groups [8][9][10].The US Preventive Services Task Forces (USPSTF) recommends that women between 50 and 74 years old receive a mammogram every two years, while women between 40 and 49 years old should make an individualized decision [11,12].Although breast cancer screening aims at early detection, intrinsic limitations do exist, such as false-positive detections leading to overdiagnosis, unnecessary costs, and negative mental and health well-being [13,14].The recent COVID-19 pandemic has further strained global cancer care and contributed to disruptions leading to potential delays in breast cancer detection, with countries such as Canada projecting significant increases in advanced-stage diagnoses and related deaths due to screening pauses [10].As the world grapples with these challenges, the primary goal remains clear: improving cancer screening behaviors through evidence-based strategies to reduce the global cancer burden [9].One of these strategies is the application of innovative artificial intelligence (AI) models and techniques to predict factors that contribute to informed decision-making about breast cancer screening in at-risk women [15].
AI applications within society are highly prevalent and are beginning to grow substantially within the healthcare field [15,16].Specifically, radiology and pathology specialties are witnessing the introduction of digital workflows and AI, which offer promising prospects in the field of precision medicine.AI is a broad term that illustrates the concept of "mimicking human intelligence using computers" [17].Computer programmers create an algorithm, and eventually, the computers can use specific data provided by programmers to make decisions [18].AI systems and techniques have rapidly evolved over the last 20 years, transitioning from machine learning (ML) to deep learning (DL), to the inclusion of advanced pathways for imaging analysis by allowing healthcare providers to analyze spatial and contextual information from images through multiple layers and convolutional operations.When it comes to daily application of AI systems, radiologists are more effectively managing workflows and detecting suspicious lesions more accurately.Hence, certain AI systems are exceeding human capabilities in predicting long-term breast cancer risk through the development of risk scores tailored for early detection of the disease and adequate intervention.
More exposure to new information improves the ability to interpret data and make decisions [18].Many cancer screening programs, such as breast cancer, focus on a "one size fits all" approach while prone to inter-observer variability, making patient selection and risk stratification challenging [17,18].In addition, overdiagnosis and false positives, as previously mentioned, are concerns within the cancer screening process, which could lead to unnecessary treatment and harm to patients [19,20].AI techniques and models can be applied in cancer prevention and management by using patient history, scans, imaging information, and analysis of specific genes for cancer classification to reduce overdiagnosis and overtreatment [21,22].This approach can help personalize medicine and benefit all patients, providers, and the healthcare system by providing risk assessment, early cancer detection, diagnosis and classification of cancers, treatment response prediction and efficacy, and helping radiologists process a large amount of data quickly [17,18].Additionally, there is a need to identify and understand the current circumstances of AI's application on breast cancer screening and prevention among adults, primarily female adults, for more effective cancer care prevention and recommendations for its future use [23].
This scoping review aims to (1) compare the major outcomes from the application of the different AI models in risk score development and screening rates changes; (2) identify the barriers encountered in applying innovative AI models and techniques in promoting breast cancer screening behaviors and predicting the risk of developing breast cancer among adult females; and (3) highlight recommendations for the adoption, adaptation, and practical implementation of such tools for breast cancer risk score development and incorporation in breast cancer screening efforts.Findings from this review can inform and guide future global recommendations for AI application in breast cancer prevention and care for female populations.

Methods
The PRISMA-SCR (Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews) was used as a reference checklist throughout this study [24].The Arksey and O'Malley methodology was used as a framework to guide this review [25].The framework methodology consisted of 5 steps: (1) Identify research questions; (2) Search for relevant studies; (3) Selection of studies relevant to the research questions; (4) Chart the data; and (5) Collate, summarize, and report the results.

Step 1: Identify Research Questions
The two research questions for the scoping review were: (1) What are the barriers encountered in the application of innovative AI techniques and models in promoting breast cancer screening and predicting breast cancer risk among adult females worldwide?And (2) what are global future recommendations for AI application in breast cancer prediction and early detection for female populations?

Step 2: Search for Relevant Studies
Keywords and mesh terms were developed by a research librarian (MK) experienced with scoping review protocols to allow for the replication of the methodology used for future reviews and other studies relevant to the topic at hand (Supplementary File S1).Search terms included AI, ML, clinical decision aid, computational intelligence, machine computer reasoning, breast cancer, breast neoplasm, malignant tumor of the breast, screening, pre-screening, early detection, model prediction, breast cancer risk, and risk score.The Rayyan platform was used to condense all studies generated from searching four electronic databases (PubMed, Embase, Web of Science, and Cochrane Library) [26].The review of the literature was conducted over a two-month period from September 2023 to November of 2023.Screening of the articles for inclusion was carried out by primary author (LS) and co-authors (DL, SB, KL, EM, NG, JX, RM, GS).

Inclusion Criteria
Included articles were peer-reviewed studies that were published in English between 2013 and 2023 that (1) examined machine learning and artificial intelligence software and models designed to predict breast cancer risk and/or promote breast cancer screening measures in adult women globally, and (2) explored the role of artificial intelligence and/or machine learning in improving breast cancer screening rates and early detection measures in adult women.AI software and models encompassed all AI techniques such as machine learning, deep learning, robotics, data mining, and reasoning that were specifically designed to predict breast cancer risk in adult women based on social determinants of health, genetic and environmental factors, and other components rendering these women at-risk of developing the disease at one point in their life.Studies were also included if these AI models and techniques were used to influence screening behavior to improve breast cancer screening rates and impact of early detection and prevention efforts in the at-risk female population at a global level.

Exclusion Criteria
Studies were excluded if they (1) addressed cancers other than breast cancer, (2) were not focused on AI, the application of an innovative AI model, technique, or methodology, (3) targeted both male and female patients, (4) included female patients under the age of 18, and (5) were not written in English.Finally, studies that were published as abstracts or used a systematic, scoping, or narrative review methodology were excluded.

Step 3: Selection of Studies Relevant to the Research Questions
Initial article screening, extraction from the relevant databases, and Rayyan page construction were performed by the lead author (LS).Co-authors (DL, SB, KL, EM, NG, JX, RM, GS) conducted a secondary screening of titles and abstracts in pairs (KL and GS; RM and JX; DL and NG; SB and EM).Consensus on disagreements was reached via discussion involving the initial reviewer (LS).
Co-authors (DL, SB, KL, EM, NG, JX, RM, GS) extracted, summarized, and tabulated the data from all relevant studies.Senior author (LS) reviewed all tabulated data to resolve any discrepancies.Summary tables included one evidence table describing study characteristics (Table 1).Table 2 summarized the barriers encountered in the application of innovative AI techniques and models in promoting breast cancer screening and/or predicting breast cancer risk among adult females globally.Table 3 provides future directions and recommendations in building more effective models for increased accuracy in breast cancer risk prediction and early detection of the disease through the promotion of screening behaviors in adult women.Basic qualitative content analysis was carried out to identify similar themes in recommendations for the advancement of AI models and techniques for breast cancer risk prediction and increased effective screening measures across studies.Selection bias by excluding certain populations (i.e., race/ethnicity, family history, past medical history) 3.
Financial barriers with screening services 4.
Screening risk might impact future decision making, leading to reduced services and follow up 5.
Uncertainty of breast cancer risk models (overestimation or underestimation) 6.
Not well defined when women should begin screening 7.
Negative • Small sample size of only six patients, likely introduced bias into the type of data variables that would be found (or missing) in the downloaded files, and therefore might limit the generalizability of findings Need to look at long-term outcomes with longer follow-ups 3.
Need to increase external validity and decrease hidden biases by increasing sample size, diversifying population (i.e., race/ethnicities), and including external datasets and population-based studies 4.
Need for personalized risk assessment/screening 5.
Validating models with other populations 6.
Communication  • Include a more diverse demographic in the study to improve the generalizability of the findings • Utilize external datasets for validation of the machine learning models • Explore the inclusion of more comprehensive predictor variables that might influence breast cancer risk

Steps 4 and 5: Data Charting, Collation, Summarization, and Reporting of Results
Study characteristics were tabulated for primary author, year of publication, study design, country, sample size, study population, study purpose, type of AI model or technique applied to breast cancer screening/risk prediction, and major outcomes (Table 1).Common limitations and challenges in the application of AI techniques and models were highlighted across the included studies (Table 2).For Table 3, the three phases of qualitative content analysis for the results of primary qualitative research described by Elo and Kyngas (2008) were applied: (i) preparation, (ii) organizing, and (iii) reporting [38].In the preparation phase, the unit of analysis is selected, which in our case was relevant lessons learned from each of the included studies in the application of AI techniques and models.This is followed by the organizing phase which encompasses data coding, grouping, categorization, and abstraction of lessons learned across studies for theme identification.The final phase, reporting, consists of sharing the results from the analysis process through tabulated categories.
Content analysis allows the description of the phenomenon in a conceptual form.For the purpose of our paper, deductive analysis was carried out since the resulting structure of the qualitative analysis was operationalized based on previous knowledge in the included studies.Additionally, a deductive approach allowed us to compare theme categories at different time periods of the published studies [38].This methodology has been widely used in the initial assessment of innovative approaches in healthcare studies [38] and aided in the identification of recurrent themes in recommendations for future advancements in the application of AI to prevent and screen for breast cancer.

Results
The initial study extraction yielded 5814 results from PubMed (n = 3054), EMBASE (n = 1455), Web of Science (n = 1245), and Cochrane (n = 60).A total of 2730 duplicate studies were excluded (n = 1226 from PubMed, n = 1344 from Embase, n = 103 from Web of Science, and n= 57 from Cochrane).A total of 3084 studies were screened for eligibility by review of their abstracts.A total of 3070 articles were excluded due to focus on breast cancer diagnosis, treatment, malignancy detection, tumors, or breast density rather than on breast cancer risk detection or screening initiation (n = 1954), lack of artificial intelligence application (n = 592), wrong population (n = 373), wrong study design (n = 144), and publication in a language other than English (n = 7).Fourteen studies were initially selected for full text review and were sourced from PubMed (n = 7), EMBASE (n = 3), Web of Science (n = 1), and Cochrane (n = 3).Upon full article review, three studies were excluded due to being published as abstracts without full texts.

Major Outcomes
The ML-DL model used by Akselrod-Ballin et al. predicted breast malignancy, identified false negative findings from previous mammograms, and outperformed existing clinically based risk models.When combined with clinical risk models, these ML-DL models improved predictive performance.Meanwhile, a variety of academic and commercial algorithms demonstrated significantly higher sensitivity in predicting breast cancer risk compared to traditional radiology assessments and higher discrimination than the BCSC model (n = 2).Additionally, the combined use of AI algorithms and the BCSC clinical risk model marked a significant increase in predictive accuracy (n = 2).This combination also reduced over-screening and under-screening of patients.They also showed improved performance when trained for shorter periods (n = 2).Furthermore, an ML model using clinical data showed improved model performance over time and helped identify novel clinically relevant patterns.The AI-based thermal imaging solution, Thermalytix, was effective in breast cancer screening.Decision aids such as RealRisks improved accuracy in breast cancer risk perception.MammoRisk, an ML-based tool, influenced changes in risk scores and screening decisions based on polygenic risk scores (PRS).The application of various ML models, including logistic regression and neural networks, provided diverse insights contributing to breast cancer risk detection and prevention.In summary, these AI techniques and models have significantly contributed to enhancing the accuracy of breast cancer risk detection and mammography screening, demonstrating the potential for improved early detection and patient-specific screening strategies (Table 1).

Common Barriers in AI Applications for Breast Cancer Prediction and Prevention
In the 11 included studies, a total of 39 barriers to AI applications in breast cancer prediction and prevention were identified (Table 2).The most common barriers in the application of innovative AI techniques and models to promote breast cancer screening behavior and improve breast cancer risk detection included lack of external validity and limited generalizability (n = 6), as AI was used in studies with either a small sample size or datasets with missing data.Many studies (n = 5) also encountered selection bias due to the exclusion of certain populations based on characteristics such as race/ethnicity, family history, or past medical history.In regard to challenges encountered by healthcare providers in the application of such tools and models, financial limitations (n = 2) and negative attitudes of female patients toward screening services (n = 2) additionally impacted the role of AI in breast cancer screening.The breast cancer risk models commonly demonstrated high levels of uncertainty (n = 3) and could not estimate when women should begin screening (n = 1).They also faced limitations with technical aspects, such as graphic demand (challenges in producing high-quality quantitative image feature analysis-based prediction models) (n = 1), long-term outcome prediction (n = 1), precise localization (n = 1), and differentiation between calcification and mass (n = 1).After receiving the results, patients had only a limited understanding of them, in part due to their complexity (n = 1).Finally, screening risk with underestimation could impact patient future decision making, and decreasing follow up (n = 1) (Table 2).

Lessons Learned and Future Directions
There are many considerations and future directions for the application of AI techniques and models for breast cancer screening (Table 3).First, AI models urgently need to include a broader spectrum and more complete predictive variables for risk assessment.Hence, there is a need to invest in generating diverse datasets to enhance the practicality and validity of AI models for breast cancer screening.Second, investigating long-term outcomes with improved follow-up periods is critical in assessing the impacts of AI on clinical decisions beyond just the immediate outcomes.Third, to enhance external validity, there are avenues for improvement, such as addressing issues with incomplete variable datasets and small sample sizes that could impact the accuracy of findings, along with including diverse population groups to avoid selection bias and ensure the generalizability of AI models.Fourth, personalized risk assessments and screenings are essential for cancer prevention strategies and can be improved by expanding datasets to include a broader spectrum of predictive variables for risk assessment.Fifth, to increase general applicability, the models should be validated with other populations to address potential biases related to race/ethnicity, family history, or past medical history that might arise in dataset selection.Finally, utilizing AI to improve communication strategies at both a local and organizational level can assist in informed decision-making and compliance, especially in populations with limited literacy levels.A concise set of variables such as benefits, harms, and statistics need to be present and clear to provide the opportunity for informed decision-making on breast cancer screening and improve patient-provider communication on the role of AI models in breast cancer risk prediction [27][28][29][30][31][32][33][34][35][36][37].

Discussion
The purpose of this scoping review was to identify the obstacles encountered when applying current innovative AI techniques and models to foster breast cancer screening behaviors and improve breast cancer risk prediction among adult females.As AI techniques and models continuously evolve in nature and scope, they are more equipped to predict breast cancer risk in women accurately [39][40][41].Romanov et al.'s model received an AUC of 0.747 when predicting cancer-free mammograms from women who went on to develop breast cancer with high predictive power, while the Mirai model maintained its accuracy across seven different populations across five countries [39,40].By incorporating diverse demographics into AI algorithms, AI offers the opportunity to individualize care and reduce healthcare disparities, such as racial and socioeconomic bias, and allow healthcare to become more equitable [40,42].Its application in early breast cancer risk detection has shown advantages towards breast cancer screening adherence by encouraging short-term and long-term actions among women [37].For instance, women who receive high-risk estimates for breast cancer could potentially be motivated to seek a physician early to begin screening and take preventative actions, like hormone therapy replacement or chemoprevention, before breast cancer arises [37].Some AI models have been designed to also identify those women at high risk for poor psychological resilience after breast cancer diagnosis, to provide early resources to women most in need to improve mental health and quality of life in the future [43].In addition, AI can serve as a feasible and affordable option, especially in promoting healthcare access in underserved and resource-poor populations [31,37].At the organizational level, implementing AI models can reduce the burden on the healthcare system, as demonstrated by Ng et al.'s study, which noted a 45% workload reduction while still enhancing breast cancer detection [44].Overall, the incorporation of AI alongside physicians has been shown to significantly reduce diagnostic time and enhance diagnostic accuracy, ultimately providing efficiency within the workplace [45][46][47].
It is crucial to address the major barriers that limit the worldwide implementation of AI models for improving breast cancer screening rates and early detection through risk scores.One significant barrier identified from this review is the limited generalizability of AI models due to small sample sizes or incomplete variable datasets.These limitations are frequently reported and stem from the challenges associated with gathering large, diverse, and comprehensive datasets that accurately reflect the broader population [48].The issue underscores the need for additional funding to support the collection of data that accurately represents diverse populations globally, ensuring inclusivity in risk assessment models and their application across larger and more diverse sample sizes [49,50].Moreover, integrating social determinants of health (SDOH) into developed risk scores is imperative to ensure these tools are more inclusive and accurately represent the populations at risk [51].SDOH encompasses a range of factors, such as socioeconomic status, education, neighborhood and physical environment, employment, and social support networks, as well as access to healthcare [52].By incorporating these factors into AI-driven risk assessments, models can provide a more nuanced and comprehensive evaluation of an individual's risk for developing breast cancer [53].This approach not only enhances the precision of risk scores but also addresses the disparities in healthcare access and outcomes among different demographic groups, particularly underserved and minority populations, by taking into consideration the underlying factors contributing to increased breast cancer risk in such racial and ethnic groups [51][52][53].In addition to challenges related to data diversity, model generalizability, and the integration of comprehensive risk factors, this review found that financial limitations and technical challenges hinder the potential of AI and ML tools to revolutionize breast cancer screening and early detection.Overcoming these obstacles requires concerted efforts to secure additional funding for staff training on effective application of AI models, foster collaborative research initiatives, and develop methodologies for integrating SDOH into AI models, thereby ensuring that these innovative tools can benefit a wider range of populations globally and contribute to the reduction of breast cancer morbidity and mortality [51][52][53].
It has been established throughout this review that the incorporation of AI into breast cancer screening is a promising tool for early detection and improved outcomes, but it also highlights the need for a multi-level approach when discussing ML to enhance general applicability and validity across numerous sociodemographic groups.Some literature has shown AI-based models to be accurate among diverse datasets; however, there remains a need for training these models on more robust, diverse datasets [39].The existing literature trains these models on large datasets, with hundreds to thousands of patients, yet these often come from a single study site and/or community [40,41].Research around AI and breast cancer screening needs to invest in generating more diverse datasets to elevate these proof-of-concept models by improving their practicality and reducing data bias.With this said, generating a diverse dataset can be challenging.Shams et al. investigated diversity and inclusion within AI research and found that studies point out an under-representation of minority groups in sampling during model training/testing, that there is less attention on equity and justice in AI design and development in general, and there is a general difficulty in measuring diversity within an algorithm [54].An obvious solution to overcome this is to share data across institutions, but this becomes highly implausible due to patient privacy policies.To circumvent this, researchers could share their models while data remains local to the study site, the concept of federated learning, to further develop their algorithms [55].Until minority groups are considered in the design, development, and implementation of AI systems, these groups are potentially not receiving any benefit from such technologies [54,55].
Moreover, underserved minority groups often face barriers to informed decisionmaking due to limited healthcare resources and lower health literacy rates [56,57].Consequently, lower health literacy rates are associated with numerous poor health and behavioral outcomes [58].Notably, individuals with inadequate self-reported health literacy have been found less likely to be adherent to mammography guidelines and have been associated with increased cancer fatalism [59,60].This highlights a clear deficit in how health-related information is presented or communicated [61].AI can play a pivotal role in developing an improved decision-making aid on breast cancer screening in underserved communities.The release of more widely available AI platforms, such as OpenAI's ChatGPT and Google's Bard, has increased public consciousness about AI, with one study finding 80% of Americans willing to use AI-power tools in their health management, underscoring its potential uses in public health across a broad community [62].Some of these conversational chatbots have already been investigated in cancer screening, prevention, and management with some success [63,64].Beyond chatbots, AI can be applied in redesigning existing patient education materials to different reading levels for patients [65].The use of AI in patient education and as an adjunctive tool for providers is still early in its incorporation, and future research should explore the implementation of AI-driven resources to enhance understanding and decision-making regarding breast cancer screening, especially in vulnerable populations with limited literacy [61][62][63][64][65].
If these concerns are not addressed within the domain of breast cancer screening development, late-stage diagnosis continues to be a considerable burden for patients and their providers.To add to this, breast cancer screening has declined due to the COVID-19 pandemic in the US [23].Women from racial/ethnic minorities (i.e., American Indian/Alaskan Native, Asian/Pacific Islander, Hispanic) and rural areas particularly are seeing great declines in screening test rates [23].This recently observed trend may consequently result in delayed identification and late-stage disease diagnosis.AI can potentially step in by improving patient education and risk assessment/stratification, as well as contribute to a more automated, cost-effective approach that would hopefully enhance existing and develop new screening and diagnostic approaches as it relates to breast cancer, benefiting both healthcare providers and at-risk women [19][20][21][22][23].

Limitations
Findings from this review should be interpreted in the context of study limitations.Although a comprehensive search across four databases was carried out for article selection that are relevant to our inclusion criteria, this review did not include tracing of reference lists, manual searches of journals, or grey literature.Additionally, this review only focused on breast cancer screening and risk prediction measures, and excluded breast cancer diagnostic measures.Broader reviews are recommended to account for other sources of literature and extend to diagnostic and management measures rather than focusing solely on preventive measures.Second, artificial intelligence in healthcare is a rapidly evolving field, so it is possible that some studies were not included due to the unintentional omission of search terms.Collaboration with a research librarian for a thorough development of mesh terms to include technical keywords relevant to machine learning and artificial intelligence has likely mitigated this concern.Third, since this is a scoping review, a formal assessment of the quality of the included studies was beyond the scope of this paper.Future systematic reviews should apply a validated checklist from the AI field to adequately assess the application and limitations of the diverse AI tools in predicting breast cancer and promoting screening behavior.

Conclusions
This scoping review describes efforts to apply innovative AI techniques and models to improve breast cancer screening rates and enhance the accuracy of breast cancer risk prediction scores.Results may contribute to a broader understanding of the limitations of these tools in breast cancer screening and breast cancer prevention measures, particularly in developing countries with limited affordability and quality of such innovative resources.This study can inform future AI healthcare specialists on more effective ways to improve the global reach and sustainability of these tools in underserved female communities who are at-risk of developing breast cancer.

Figure 1 .Figure 1 .
Figure 1.PRISMA Flow Diagram of the Study Selection Process.Reasons for record exclusion (**) were as follo • Wrong Outcome including focusing on breast cancer diagnosis, breast cancer treatment, detection of malignancy, tumors, or breast density, not focusing on breast cancer risk detection, screening initiation, mammography (n=1954) • No application of artificial intelligence tools (n=592) • Wrong Population (n=373) • Wrong Study Designs including systematic review, scoping review, narrative review, and meta-analysis (n=144) • Published in a Language Other Than English (n=7)

Table 2 .
Common Barriers in AI application for Breast Cancer Prediction and Prevention.
•Retrospective ascertainment of BCSC clinical risk model inputs for family history and prior breast biopsies could have led to underestimation of BCSC performance • Most algorithms have not been trained to predict long term outcomes

Table 3 .
Limited predictor variables due to reliance on available data in the PLCO dataset • Lack of external validation to demonstrate the generalizability of the models • Potential bias due to excluding certain demographic groups based on the data available in the PLCO dataset Lessons Learned and Future Recommendation in AI application for breast cancer screening and risk prediction efforts.
•Image-only deep learning models can be more accurate than traditional risk factor models, especially when risk factors are not available in the patient history, thus