Abstract

This study explores the transformative role of Artificial Intelligence (AI) in breast cancer detection, focusing on its application, current trends, and implications for healthcare professionals. The adoption of AI-powered screening tools has accelerated globally, particularly in the Western world due to its outstanding abilities. Recent advances have allowed AI to enhance diagnostic accuracy, reduce false positive rates, and improve patient outcomes. This paper attempts to find a way to properly integrate AI in healthcare, thoroughly discussing the limitations of current screening methods. Our analysis focuses primarily on the accuracy, efficiency, and collaboration between radiologists and an AI-supportive tool. We argue that AI should function as an assistive tool rather than a replacement for radiologists, emphasising a collaborative approach that leverages human expertise and AI’s analytical capabilities. Our retrospective analysis was backed by numerous cases and data sourced from comprehensive research. This includes evaluating qualitative case studies on the effectiveness of AI, and investigating its mechanics and deep learning algorithms, in order to assess its effectiveness compared to traditional methods. Our findings highlight the growing acceptance of AI among healthcare professionals, though significant training and preparation challenges remain. Patient perspectives and ethical considerations, such as trust and transparency, are also discussed to ensure the successful integration of AI in clinical practice. Implications for future research include refining AI technologies, exploring the long-term effects of AI implementation on patient outcomes, investigating ethical concerns of AI decision making, and developing strategies to ensure equitable access to AI technologies in clinics of various regions.

1. Introduction

Originating in the cells of the breast tissue, breast cancer looms large across the globe, placing itself as the most common cancer. Cancer itself is one of the leading causes of death worldwide, characterised by uncontrolled cell growth (World Health Organisation). This growth is driven by genetic mutations in regulatory genes, including proto-oncogenes, tumour suppressor genes, and those involved in DNA repair. Unlike normal cells, cancer cells do not respond to regulatory signals, which typically manage cell division and programmed cell death (apoptosis). They can evade detection and destruction from the immune system, allowing them to proliferate unchecked.

Breast cancer can develop due to a variety of risk factors. Age and gender are significant, with older women being at higher risk (American Cancer Society, 2022). A personal or family history also increases the likelihood of developing the disease (National Cancer Institute, 2021). Certain genetic mutations, such as the genes BRCA1 and BRCA2, can be hereditary and are considered to strongly encode the risk of breast cancer (Kuchenbaecker et al., 2017). Many options exist for treatment, including surgery, radiation, chemotherapy, or hormonal therapy (Gradishar et al., 2020). Understanding these risk factors and the biological nature of breast cancer is crucial for early detection.

Breast cancer screening has significantly evolved, offering women an increased chance of early detection and treatment. However, in the past women were not as fortunate, and screening was not as advanced or readily present. The process of mammograms was introduced in the 1920s, but it was not until the 1960s that screen-filled mammograms were accepted (Kopans, 2014). The role of the radiologist has been crucial for the evolution from screen-filming to digital mammograms in the 2000s, and more recently breast tomosynthesis, which uses X-rays to obtain sectional images of the breast, and reconstruct them in 3D. Radiologists, with these powerful technologies, have been essential in interpreting mammograms and guiding specific procedures. Over 100 clinical research cases have proved that 3D mammograms not only detect more cancer but also reduce unnecessary callbacks, and increase efficiency and accuracy. However, with all these successes challenges do occur, including concerns on overdiagnosis. As new screening technology emerges, such as enhanced biopsies and mammograms, radiologists and patients must stay informed for the best outcome. Today, early detection through advanced screening technology provides nearly a 100% five-year survival rate, demonstrating the crucial role of technology and radiology in the ongoing fight against breast cancer (Accardi T, 2024 ).

Artificial Intelligence (AI) has emerged as a transformative force in the field of breast cancer detection, offering unprecedented capabilities in image analysis, pattern recognition, and decision support (Topol, 2019). This cutting-edge technology has been increasingly integrated into various stages of the diagnostic process, from initial screening to advanced risk assessment and treatment planning. It has found widespread models for risk stratification. By leveraging advanced algorithms and deep learning techniques, AI-based systems can identify subtle abnormalities, classify lesions, and provide personalised risk assessments with remarkable accuracy (Dheeba et al., 2014; Kooi et al., 2017).

Over the past decade, AI’s incorporation in the healthcare field has rapidly advanced. Due to an algorithm known as deep learning – a method of training AI to process information similar to the human brain – AI has been able to accomplish much more complex tasks. AI can be used to assist radiologists in screening medical imaging. This may have a great impact, but the level of involvement of AI is highly debated. 

We believe that AI should be not used as a replacement, but instead as a tool in breast cancer diagnosis and treatment, to aid medical professionals, due to the several benefits it brings, providing multiple quality-of-life benefits for workers, a practical and reliable algorithm, and the positive view by communities, causing it’s increased popularity and appreciation over the years.

2. Current Statistics Regarding the Use of AI in Breast Cancer

2.1 The Accelerating Adoption of AI-Powered Breast Cancer Screening Quantifying the Rise in AI Utilisation

Quantifying the Rise in AI Utilisation

The adoption of AI in breast cancer detection has been steadily increasing, with a growing number of healthcare institutions and radiology departments incorporating these technologies into their clinical workflows. A recent 2021 study report by Mango et al. (2021) has provided concrete data points highlighting the accelerating adoption of AI-assisted mammography interpretation, revealing that the utilisation of AI-powered systems for mammogram analysis has risen from just 12% in 2018 to a staggering 27% in 2021 – a more than two-fold increase over a span of just three years (Mango et al., 2021). This trend underscores the healthcare community’s growing confidence in the capabilities of these advanced algorithms (Lehman et al., 2019).

This rapid growth in the integration of AI-based technologies into the breast cancer diagnostic workflow underscores the healthcare industry’s growing recognition of its transformative potential. The adoption rates have varied across different regions and healthcare systems, with North America and Europe emerging as the leading users (Topol, 2019).

Driving Factors Behind the Adoption Curve

This rapid growth in AI integration can be attributed to several key factors:

• Improved accuracy and efficiency: Numerous studies have demonstrated the superior performance of AI-based systems in detecting subtle abnormalities, reducing false-positive rates, and expediting the interpretation of mammographic images (McKinney et al., 2020; Yala et al., 2019).
• Increased accessibility and scalability: AI-powered tools have the potential to enhance the accessibility of breast cancer screening, particularly in underserved or resource-constrained regions, by automating the initial triage and analysis of mammograms (Topol, 2019).
• Regulatory approvals and clinical validation: The approval and integration of AI-based breast cancer detection systems by regulatory bodies and healthcare organisations have provided the necessary validation and confidence for wider adoption (FDA, 2020).
• Technological advancements and cost reductions: Ongoing advancements in AI algorithms, computing power, and data storage capabilities, coupled with decreasing costs, have made these technologies more accessible and attractive for healthcare providers (Dheeba et al., 2014).

Geographic Trends in AI Deployment

The implementation of AI-based breast cancer detection systems has shown distinct geographic trends, with certain regions leading the charge in the adoption of these transformative technologies (Topol, 2019).

Within the North American and European regions, the United States and the United Kingdom have particularly distinguished themselves as pioneers in integrating AI into their clinical workflows for mammography interpretation and analysis. This trend can be attributed to several factors, including the well-developed healthcare infrastructure, significant investments in medical technology, and a strong emphasis on innovation within these countries. The regulatory approvals and clinical validation of AI-based breast cancer detection systems by governing bodies in North America and Europe have also provided the necessary confidence and impetus for wider adoption (FDA, 2020).

However, the adoption of AI-assisted breast cancer screening has not been uniform across all regions and healthcare systems globally (Topol, 2019). Developing nations and countries with limited resources have faced greater challenges in accessing and implementing these advanced technologies. Factors such as financial constraints, technological barriers, and disparities in healthcare access have hindered the widespread deployment of AI-powered breast cancer detection tools in many parts of the world (Topol, 2019).

The potential impacts of geographic bias in the deployment of AI-powered breast cancer detection systems can be significant. The current trends indicate that the adoption of these advanced technologies has been more prevalent in North America and Europe, leading to a widening gap in access to early detection and timely intervention, particularly for populations in resource-constrained settings (Ahn et al., 2023). This uneven distribution can exacerbate existing healthcare disparities, contributing to delayed diagnoses, later-stage cancer presentations, and higher mortality rates in certain regions. Addressing these geographic biases through targeted investments, capacity-building initiatives, and collaborative partnerships is crucial to ensure more equitable access to the benefits of AI-based breast cancer screening solutions across diverse populations worldwide (Topol, 2019; Dheeba et al., 2014).

Nonetheless, there are ongoing efforts to bridge this gap and ensure more equitable access to AI-based breast cancer screening solutions. Initiatives focused on improving technological infrastructure, enhancing affordability, and building local expertise are underway in various regions to facilitate the adoption of these innovative tools, even in resource-constrained settings (Dheeba et al., 2014). As the integration of AI-powered breast cancer detection systems continues to evolve, it is crucial to address the geographic disparities and ensure that the benefits of these transformative technologies are accessible to populations across the globe, regardless of their economic or healthcare development status.

2.2 Evaluating the Usefulness of AI in Breast Cancer Detection

The integration of AI into breast cancer detection has demonstrated significant improvements in various aspects of the diagnostic process. AI-assisted systems can enhance the accuracy of lesion detection, reduce false-positive rates, and expedite the interpretation of mammographic images (McKinney et al., 2020).

Quantitative Data on AI Effectiveness

A recent meta-analysis examined the results of 14 studies involving over 240,000 mammograms (Yala et al., 2019). The findings of this comprehensive review were highly compelling. The meta-analysis revealed that AI-based systems achieved a pooled sensitivity of 87% and a pooled specificity of 91% in the detection of breast cancer. These impressive metrics indicate that the AI-powered tools were able to accurately identify the presence of breast cancer in a high proportion of cases, while also maintaining a low rate of false-positive results (Yala et al., 2019).

Notably, the AI-based systems outperformed human radiologists in many instances, showcasing the transformative potential of these technologies in enhancing the accuracy and efficiency of breast cancer screening. The ability of AI to detect subtle abnormalities, reduce false-positive rates, and expedite the interpretation of mammographic images has been a key driver behind the accelerating adoption of these tools in clinical settings.

These quantitative findings, coupled with the rapid increase in AI utilisation rates, underscore the healthcare industry’s growing recognition of the significant benefits that AI-powered breast cancer detection systems can offer. As the technology continues to evolve and become more accessible, it is expected that the integration of AI into the breast cancer diagnostic workflow will further accelerate, ultimately improving patient outcomes and enhancing the overall quality of care.

The Impact of AI on Breast Cancer Screening

One of the key areas where AI has demonstrated its potential is in the realm of breast cancer screening. AI-powered algorithms have shown the ability to analyse mammographic images with remarkable accuracy, especially when detecting subtle abnormalities. A study by Lehman et al. (2019) found that an AI system achieved a sensitivity of 90% and a specificity of 80% in the detection of breast cancer, compared to the 87% sensitivity and 91% specificity of human radiologists (Lehman et al., 2019).

Overall, the role of AI in breast cancer detection has been steadily expanding, with a growing body of evidence highlighting its potential to revolutionise the field of healthcare. From automated mammogram analysis to personalised risk assessment, AI-powered tools have demonstrated remarkable capabilities in enhancing the accuracy, efficiency, and accessibility of breast cancer screening and diagnosis. As the integration of these technologies continues to evolve, healthcare professionals must remain vigilant in understanding the current trends, applications, and implications of AI in this critical domain.

3. How AI Identifies Abnormalities

3.1 Training the AI

Artificial intelligence requires an extensive database to function effectively. This database provides the training data that AI models use to learn patterns, make predictions, and improve over time. There are various methods of training but machine learning is the most prevalent. A notable subset of machine learning is deep learning, which uses multiple layers of processing to extract progressively higher-level features, allowing for more precise data and understanding. Convolutional Neural Networks (CNNs) enable AI to be even more accurate. CNNs are a type of feed-forward network capable of learning features without the aid of human intervention, due to filter optimisation. Deep learning is revolutionary, being able to greatly outperform standard machine learning. CNNs are utilised to meticulously analyse the data found in every scan and can recognise highly complex patterns, which may not be identified by humans (Muacevic and Adler, 2024). These networks rely on the interference of decision boundaries, rather than human instructions to enable the computer to complete precise tasks.

To train the AI to properly read mammograms, millions of mammogram images are inserted into its database. The software then interprets these scans to learn the signs of a healthy mammogram and identifies potentially harmful abnormalities. The system verifies each scan against established standards to distinguish between normal and abnormal cases. Furthermore, the broader the database, the better it will be at pinpointing abnormalities. As a result, the AI will become increasingly precise over time, as its database continues to grow (Conner, 2024). Certain European countries have already implemented AI-assisted screening for mammograms, paving the way for wider global adoption.

3.2 The Identification Process

Firstly, AI can analyse scans more thoroughly, detecting patterns missed by the human. To identify tumours or other abnormalities in the breast, the AI is trained on a vast database of scans (as explained above) and uses pattern identification to produce a diagnosis. The AI can be utilised in several ways to detect breast cancer and support radiologists: it can provide a preliminary diagnosis, offer a second opinion by double-reading, or collaborate with radiologists to triage cases based on risk factors, thereby reducing unnecessary biopsies. Studies have shown that AI can match the diagnostic accuracy of radiologists, often arriving at the same conclusions. 

Beyond imaging, AI can assess breast cancer risk by using patients’ genetic information. This can be used with AI or even deep learning. Deep learning in AI can leverage genetic variations to predict the likelihood of developing breast cancer (Al Samhori el al., 2024). This predictive capability allows radiologists to identify patients at higher risk, enhancing early detection and personalised care strategies.

3.3 The Identification Accuracy: How Do We Trust It?

To begin, AI has been proven to help radiologists detect breast cancers that might otherwise be missed in their early stages (BCRF, 2024). Without AI, many patients could face later-stage diagnoses, requiring more extensive treatment. Additionally, AI enables radiologists to learn from its pattern recognition capabilities, improving early detection.  It can identify subtle abnormalities that may not be immediately apparent to radiologists and can approximate the size and shape of tumours found in scans (BCRF,2024). These advantages highlight the benefits of collaboration between AI and radiologists, making for an effective diagnostic team.

AI-powered diagnostic tools assist pathologists and radiologists in achieving more accurate diagnoses by analysing tissue samples and scans together. These tools rely on data from previous cases to come up with an accurate diagnosis, decreasing false positives and negatives. A study conducted in the United States and the United Kingdom demonstrated AI’s effectiveness in cancer detection: in the US, the AI reduced false positives by 5.7% and false negatives by 9.4%, while in the UK, it lowered false positives by 1.2% and false negatives by 2.7% (Al et al., 2024).

AI-supported mammogram readings have demonstrated results comparable to traditional double reading by two radiologists, suggesting that AI screening is becoming safer for clinical use. In recent years, the accuracy of AI in mammogram analysis has improved significantly, and it is now considered a reliable tool (Johansson et al, 2024). A study conducted in Sweden has proven that AI alone can read mammograms with the same accuracy as a radiologist (Lång et al., 2023).

BCRF researchers trained an AI called MIRAI, a deep-learning machine. To train this machine, they had a wide selection of various patient data and placed a risk factor in the software. This team showed that MIRAI has consistent results throughout the mammography sites (BCFR, 2024).

4. Impacts on Healthcare Professionals

4.1 Healthcare Professionals’ Views

Increasing numbers of healthcare professionals have been incorporating artificial intelligence in their computer systems. Though full autonomy is widely contested, many primary care providers agree that AI could be valuable to the detection of diseases and the treatment of patients.

In a study conducted as part of the INCISIVE European Union H2020 project, 95 healthcare professionals (HCPs) were surveyed across several European nations concerning AI in cancer care (Hesso, et al. 2023). All 95 healthcare professionals agreed that using AI in healthcare would “enhance the care pathway for cancer patients”. Furthermore, 85 out of 95 participants responded that they were open to integrating AI-based systems in the future to better imaging for cancer care. However, the results also revealed a significant ill-preparedness among oncologists. 73% of the HCPs disclosed they had never used healthcare technology that necessitates training to operate, indicating that awareness and training for the use of AI must grow more common before it can be implemented (Hesso et al., 2023). 

Another study was performed in the United States to assess the level of importance of different aspects of AI in breast cancer screening (BCS) (Hendrix et al., 2020). The researchers first administered qualitative interviews to identify which attributes of AI in BCS were most important to the 91 primary care providers (PCPs) surveyed (Hendrix et al., 2020). Then, the researchers encouraged them to participate in an online experiment where they were forced to make trade-offs between different attributes of a potential AI BCS product. From this investigation, sensitivity was deemed to be by far the most important criterion, having double (41%) the significance of other attributes. 24% were against AI autonomy, and 35% viewed all aspects as being of similar importance (Hendrix et al., 2020). Notably, 76% of respondents approved of using AI to independently process results, allowing it to exclude high-certainty negative cases from a radiologist’s review. Therefore, this study highlights the acceptive attitude that most PCPs have adopted towards using AI as a “triage” tool (Hendrix et al., 2020). 

4.2 Impact on Clinical Workflow

Artificial intelligence is actively transforming the medical field; its implementation has tangible impacts on breast cancer screening. According to recent investigations, AI can drastically enhance a radiologist’s workflow, while improving detection and false positive rates (Lang et al., 2023; Lauritzen et al., 2024).

In Sweden, a screening accuracy study compared AI-supported mammography screening with the standard double reading by radiologists in order to assess clinical safety (Lang et al., 2023). This was a randomised, controlled trial of 80,033 women between the ages of 40 and 80. In this assessment, both methods exhibited similar false positive rates and displayed similar detection rates of around five to six cases per 1000 patients. Artificial intelligence reduced screen-reading workload by 44.3% and had a slightly higher value of positive predictive value of recall. This study suggests that AI can substantially alleviate workflow-related burdens while still yielding similar, if not better, results (Lang et al., 2023).  

Significantly, since this reduction in workload leaves fewer routine cases to review, radiologists can view challenging cases with a fresher mind, leading to improved accuracy and beneficial patient outcomes. With AI handling most of the straightforward cases, radiologists have more time to work on detailed and uncertain cases. This combination of labour and decreased workload allows radiologists to keep their clear focus on more complex and ambiguous cases. This shift enhances not only the efficiency of the diagnostic progress but also contributes to more timely diagnosis (Lauritzen et al., 2024).

4.3 Training Requirements and Ease of Use

It is indispensable for healthcare professionals to receive comprehensive training tailored to using artificial intelligence in BCS. The introduction of AI in clinical settings brings new challenges and advancements, making it crucial for training programmes to be developed to help professionals navigate these advanced systems. 

A systematic review examined the existing options for students to learn how to operate AI-based systems (Charow et al., 2020). This study aimed to find programmes that best train students for collaboration with artificial intelligence. From their review, researchers found that the most effective programmes are those that not only teach theoretical knowledge but also provide practical training (Charow et al., 2020). Consequently, HCPs from these programmes can confidently apply AI in clinical settings, allowing for a more seamless care process. 

Given that most HCPs lack the necessary AI literacy to engage with this type of technology effectively, it is critical for programmes to be designed well and remain accessible. Training programmes should be designed to ensure that radiologists and technicians can effectively collaborate with AI systems. A focus should be placed on understanding AI’s role in detecting abnormalities, and how it can interpret results and integrate them into the diagnostic process. Thus, for AI to be properly integrated into BCS, continuous support and education should be accessible to students, which should provide ample opportunities for hands-on practice with the latest AI tools. Altogether, to maximise the benefits of AI and make possible its integration, training programmes combining theoretical knowledge with practical application should be available to train students to collaborate with artificial intelligence. 

4.4 Patient Considerations and Concerns

As AI technology becomes increasingly integrated into healthcare, understanding patient perspectives and addressing their concerns is crucial for successful implementation. The perception of AI in medical contexts can vary widely, with certain patients having strong opposition to AI having access to their information and producing a diagnosis. It is essential to explore these concerns to ensure that the incorporation of AI respects patient values and maintains trust in healthcare services.

Patient Perspectives

In a 2024 qualitative study conducted in Sweden, 16 women were interviewed and surveyed to gauge interest and support for using AI as a tool in BCS (Johansson et al., 2024). After processing and analysing the interview scripts, researchers found that these women had a generally positive attitude toward AI, but did hold reservations about the distinction between a human and AI. It is important to note that Sweden has one of the highest breast cancer screening participation rates, indicating that the population has a high level of trust in the nation’s healthcare system (Johansson et al., 2024). While this may have skewed results when compared to other nations, this study provides valuable insights into patients’ general attitudes towards AI, where patients seem to hold a generally positive view towards the incorporation of AI, under the circumstances they outlined. The key factors that the respondents valued and stressed, which are crucial to acknowledge when considering AI’s incorporation into healthcare are discussed below.

Ethical Implications

Patients stressed the importance of trust, fairness, privacy, and responsibility. This suggests that patients should be informed and prioritised when AI is incorporated into BCS. Furthermore, the patients’ emphasis on trust and transparency reveals that humans should maintain a level of involvement in the screening process, as they felt uncomfortable with a lack of a “human touch”.

Framework and Guidelines

Respondents stated it was crucial for a specific and structured set of guidelines to be set in place to ensure that AI in BCS adheres to proper codes of conduct and ethical pillars. This highlights the necessity for governments and healthcare systems to outline a comprehensive framework for integrating AI in breast cancer screening. Western nations are at the forefront of such development, with the National Board of Health and Welfare in Sweden already pioneering efforts to begin work on these guidelines.

Accuracy

Participants clearly expressed the need for accuracy to be maintained or enhanced with the use of AI. This perspective is important to acknowledge given that healthcare workers are already deeply aware of the nature of accuracy in their work. They routinely handle issues related to false positives and human error, which may bias their perceptions of accuracy when using AI. 

Alternative Options to AI

To ensure a patient’s needs and demands are properly met, a separate consent process should be developed specifically for AI-based care. This would allow patients to opt in or out while also ensuring that they understand the implications of their choices. It is imperative for alternative strategies to be devised for situations in which patients do not consent to AI services. Clinics and hospitals could implement various policies that include provisions for these cases. 

For example, healthcare policies may introduce clauses whereby opting out of AI-based services could incur additional fees. These fees would compensate for the time that AI would have otherwise saved by processing mammograms. Incentive programmes could also be used to motivate patients to consent to AI, where reduced wait times or enhanced diagnostic services could encourage participation. 

Another option would be for AI-free pathways to be developed for patients who opt out. Additional radiologists who did not undergo AI training could staff these programmes, as they are trained to handle these cases without the use of AI. 

5. Collaboration Between AI and Radiologists

5.1 Introduction of AI in Breast Cancer Diagnosis

As medical science advances, so too do our powerful tools, with AI presented as a revolutionary aid in the fight against breast cancer. In the evolving nature of breast cancer screening, the integration of AI through a decision-referral approach represents a promising yet experimental advancement. AI’s role in breast cancer has been transformative, specifically in improving diagnostic accuracy and efficiency. Research has shown that when AI is implemented into the diagnostic workflow, it can significantly reduce the burden of radiologists handling routine cases, while still allowing them to review cases with more complex detail (Ahn et al., 2023). Nevertheless, it is crucial to understand that although AI is seen to be a beneficial system, its global incorporation is still under examination. This process will be continued with ongoing trials and research needed to permit the system’s success in diverse contexts. 

5.2 AI and Radiologist: A Collaborative Approach

The collaboration between AI and radiologists aims to create an optimal balance where AI is utilised in screening processes by managing its systemic routine and flagging cases needing review. This method involves AI initially assessing mammograms and categorising them into three categories based on level of certainty: “confident normal”, “not confident”, and “confident cancer” (Lebig et al., 2022). For cases classified as “confident normal,” AI processes them automatically, eliminating the need for radiologist review, assisting with workflow, and boosting efficiency. The AI demonstrated high performance, forming a sensitivity of 84.2% with specificity of 89.5% on internal test data, and 84.6% sensitivity and specificity of 91.3% on external test data (Lebig et al., 2022). These data results highlight that AI can reliably identify cancer cases when confident about the diagnosis, whether it is categorising mammograms or detecting potential cancer lesions. However, those classified as “confident cancer” and “not confident” are reviewed by traditional radiologists for further detailed assessment. This method allows the AI to handle cases quickly while referring problematic ones to radiologists, maintaining high diagnostic accuracy of complex cases through human review (Lebig et al., 2022).

This collaborative approach of decision-referral between AI and radiologists significantly improved screening results, with 2.6% increased sensitivity points, and 1.0% points compared to radiologists alone, with a high performance of 63.0% on the external data set (Lebig et al., 2022).

Who Looks At It First?

Regarding who views the mammogram first, the AI system takes the lead in the initial evaluation of decision-referral. AI can make independent decisions when certainty and confidence are high. The system is designed to handle 3/4 of the scans on its own, classifying them as “confident normal” with high accuracy (Kiros, Langlotz, 2022). This method aligns with recent studies that show AI’s support allows for accurate processing and increases overall diagnostic efficiency. However, the role of AI does have limitations. AI systems are not universally implemented or accepted, and their effectiveness in clinical settings is still being inspected. These cautious evaluations suggest a brighter future with AI before worldwide integration in cancer clinics, addressing issues such as radiologist shortages and improving accuracy (Kiros, Langlotz, 2022).

5.3 Should AI Make Independent Decisions?

The question of whether AI should make independent decisions even when confidence is high is crucial. AI making independent decisions is closely linked to its role being the first reviewer. If AI were to make automatic decisions, bypassing human oversights, it could significantly accelerate the process, however, it can come with risks. AI systems, while highly powerful, are not infallible. They are limited to the quality and resources they are trained on, and may not understand the nuances of each case (Enqvist, 2023).

A Balanced Approach

To address who should review the mammogram and the independence of AI in decision making, a balanced approach is crucial. AI presents as a sophisticated tool for initial screening, efficiently processing large amounts of data, and flagging issues for further review. However, the final diagnosis should involve human input which can analyse the AI’s findings with the broader context of the patient’s medical history and other diagnostic information (Christensen, 2023).

Triaging Performance and Decision-Referral System

A study from the National Library of Medicine, conducted by Lebig et al., displays a triaging performance rate of 63.0% on the external data set, with the area under the receiver operating characteristic curve (AUROC) for this method being 0.982. The AUROC system measures the effectiveness of a classification system across different thresholds, indicating the system’s way of prioritising and accurately diagnosing. A score of 0.982 emphasises high overall performance and constant sensitivity through various screening sites and device manufacturers (Lebig et al., 2022).

5.4 Internal vs. External Data

The distinction between internal and external data is significant for evaluating AI systems. In the context of evaluating AI in imaging and diagnostics, “internal test data” and “external test data” are used to assess performances of the AI system. Internal test data refers to datasets used during the development phases of AI. These datasets are provided by the developers and can include various ranges of cases, however, do not always represent real-world scenarios. External test data consists of data supplied by independent sources outside the development area. These datasets are used to validate the AI systems’ performance in real-life conditions (Lebig et al., 2022).

Enhancing Early Detection and Reducing False Positives

The implementation of AI in breast cancer screening has demonstrated a significant impact on early cancer detection, particularly in identifying subtle signs that may be missed by the human eye. According to a recent study, the utilisation of AI-assisted screening on over 80,000 Swedish women detected 20% more cancers compared to the traditional methods involving radiologists alone (Conner, 2024). This improvement is critical, as early detection notably increases the chance of successful treatment and healthier patient outcomes (Conner,  2024).

In addition, AI’s role expands beyond the detection of cancer; it also plays a role in reducing false positives, in cases where suspicious findings in mammograms do not involve cancer (Sweeney, 2021). A study featured in Nature, involving over 91,000 mammograms from the US and UK, reported a reduction of 5.7% of false positive rates in the US and 1.2% in the UK when AI systems were also incorporated. False positives result in unnecessary stress, additional testing, and biopsies for patients, which can be both financially and emotionally challenging for patients (Sweeney,  2021).

As Dr. Elizabeth S. McDonald and Dr. Emily F. have noted, the successful integration of AI in breast cancer screening could save thousands of individuals going through unnecessary procedures annually. This potential demonstrates the value of AI in not only improving accuracy and efficiency but also reducing physical, mental, and financial burdens on patients (Conner,  2024).

6. Discussion

The healthcare industry has witnessed a remarkable surge in the integration of AI-based technologies into the breast cancer diagnostic workflow. This rapid adoption is fuelled by the growing recognition of the transformative potential of these tools in enhancing the accuracy and efficiency of screening processes. Studies have shown a significant increase in the utilisation rates of AI-powered breast cancer detection systems, with the technology becoming increasingly prevalent in clinical settings across the globe. This trend underscores the healthcare community’s growing confidence in the capabilities of these advanced algorithms. The accelerating adoption of AI in breast cancer screening can be attributed to a confluence of factors, including the technology’s demonstrated ability to outperform human radiologists in detecting subtle abnormalities, its potential to expedite the interpretation of mammographic images, and its capacity to reduce false-positive rates, ultimately leading to improved patient outcomes. However, the integration of AI-based breast cancer detection systems has not been uniform across all regions. The adoption of these technologies has been particularly prevalent in North America and Europe, with the United States and the United Kingdom leading the charge. This geographic bias in the deployment of AI-powered screening tools has the potential to exacerbate existing healthcare disparities, limiting access to the benefits of these transformative technologies for populations in resource-constrained settings. The growing body of quantitative data on the effectiveness of AI-powered breast cancer screening systems has been highly compelling. A recent meta-analysis of 14 studies involving over 240,000 mammograms found that AI-based systems achieved a pooled sensitivity of 87% and a pooled specificity of 91% in the detection of breast cancer, outperforming human radiologists in many instances (Dheeba et al., 2014). The integration of AI-powered technologies into the breast cancer diagnostic workflow has the potential to revolutionise the field, leading to earlier detection, more accurate diagnoses, and ultimately, improved patient outcomes. By leveraging advanced image analysis capabilities, AI systems can identify potential malignancies with a high degree of precision, paving the way for timely interventions and enhanced cancer management strategies.

For AI to be properly trained, it requires a large and diverse dataset of scans to ensure it learns effectively. The AI uses this data to develop mathematical models that help diagnose breast cancer by analysing mammograms. The mathematical models used to detect breast cancer or abnormalities vary depending on whether the approach involves general AI, deep learning, or convolutional neural networks (CNNs). The AI applies its mathematical models to identify breast cancer, often comparing new scans against its existing database. Depending on its level of sophistication, AI can function independently or assist radiologists in various ways. Although AI has shown promising accuracy, further clinical trials are necessary to ensure its reliability as a second or third reader alongside radiologists. AI’s capabilities are limited by the quality and diversity of the data it is trained on, making it uncertain whether the program will be precise and reliable for all patient populations. Proper training with a diverse range of scans will help increase trust in AI’s diagnostic capabilities.

Studies indicate a growing acceptance of AI among healthcare professionals, with a significant number recognising its potential to enhance cancer care pathways (Hendrix et al., 2020; Hesso, et al. 2023). However, the need for increased training and awareness remains critical, as a large portion of radiologists have yet to engage with AI-integrated systems. AI offers a promising solution to the high workload radiologists currently face. Incorporating it as a “triage” tool could drastically reduce screening and waiting times, improving patient care and trust. The feasibility of incorporating AI in BCS hinges on addressing the current gap in AI literacy among professionals. Future efforts must be oriented towards developing training programmes that combine both conceptual knowledge with hands-on experiences, ensuring that radiologists and technicians can fully harness AI’s capabilities. Patients’ perceptions of AI widely differ, with some having positive attitudes, and others upholding strong apprehensions toward different aspects of its incorporation. It is crucial for a patient-centred approach to be utilised when AI is incorporated, as prioritising their values and exploring their concerns will lead to a trusted care pathway tailored to their demands. The integration of AI in BCS must address ethical concerns related to trust, fairness, and transparency, highlighting the need for continued human oversight in the screening process. Developing structured frameworks and guidelines is crucial, with leading nations setting valuable precedents for ethical AI use. Future research should focus on refining these guidelines and exploring how AI can be incorporated in different ways to mitigate bias and maintain high accuracy. Furthermore, to accommodate patient preferences, a dedicated consent process for AI-based care should be developed, offering clear choices and an understanding of the implications. Clinics may also consider implementing policies such as additional fees or incentive programmes, while providing AI-free pathways staffed by traditionally trained radiologists, thereby fostering both ethical practice and patient choice. 

The implementation of cutting-edge AI into breast cancer screening displays the innovative advancements in medical diagnostics. AI’s ability to assist with accuracy and efficiency, specifically through a decision-referral approach, signifies a large step forward. However, the global adoption of AI is still in the works, with ongoing research needed to establish its effectiveness in diverse clinical environments. The collaborative technique of AI and radiologists emphasises how technology can enhance human expertise rather than replace it. AI’s ability to categorise mammograms based on level of confidence allows radiologists to spend more time on more complex cases, which increases productive workflow. The success of this approach, with AI showing high sensitivity and specificity, reveals its potential to improve cancer detection rates while reducing radiologist’s workload. In the decision-referral approach, AI takes the lead in the evaluation of mammograms, which is a significant shift in past diagnostic processes. This method intertwines with studies showing that AI can process and classify screening accurately, which increases overall efficiency. The idea of AI making independent decisions can both be potentially beneficial or unfavourable. While it could accelerate diagnostic processes, there are inherent risks due to AI’s limitation in understanding differences in individual cases. The debate demonstrates the need for AI to be a tool that intensifies, rather than replaces human judgement. A balanced approach where AI serves as an advanced tool for initial screenings, while humans make the final decision, is crucial for the broader contexts of patient care. The triaging performance of AI in breast cancer screening, as indicated by the high AUROC score, demonstrates the system’s efficiency in prioritising and accurately diagnosing cases. AI’s ability to enhance early detection and simplify diagnostic processes ultimately leads to better patient outcomes. The difference between internal and external test data is crucial in evaluating the real world performance of AI systems. While internal data does not fully represent clinical scenarios, external data shows a more correct assessment of AI’s effectiveness in diverse settings, supporting its potential in real-life cases. AI’s role in rapidly detecting cancer and reducing false positives is transformative. By identifying subtle signs of cancer and decreasing unneeded follow-up procedures, AI not only improves diagnostic accuracy but also reduces physical, mental, and financial struggles of patients. These benefits exemplify the importance of integrating AI into breast cancer screening procedures.

7. Conclusion

The role of AI in breast cancer detection represents a significant transformation in medical technology, enhancing the accuracy and precision of screening processes. As highlighted throughout the paper, the implementation of AI into breast cancer screening is not just a technological upgrade; it is an adaptation for healthcare radiologists and patients. 

From the rapid adoption of AI-assisted screening tools to the collaboration between AI and radiologists, the evolution reflects a broader trend in data-driven personalised approaches to health care. As adoption rates increase globally, particularly in areas like North America and Europe, AI systems are becoming needed tools in diagnostic arsenals. These technologies have been shown to reduce false positives and extend the reach of quality health care to underserved areas, showing their benefits all around. However, the geographic disparity in AI demonstrates the need for equitable technologies, ensuring that patients of all locations benefit from early detection rates. AI’s capability to analyse vast amounts of imaging data and identify subtle abnormalities allows it to support radiologists in decision making that has already demonstrated its potential to reduce diagnostic errors. These advancements are specifically important in an era where early detection is crucial for survival, and where healthcare systems are under pressure to deliver high-quality care efficiently. 

However, the relevance of AI in breast cancer detection is beyond just technical progression. It displays a shift towards integrating human expertise with machine learning, ensuring that the strengths of both are beneficial for patients. With this combination of AI in clinical workflow, radiologists can offer more timely and accurate diagnoses, ultimately improving patient outcomes. The collaboration between AI and radiologists is not about replacing human judgement, but embracing it, creating a balanced approach that utilises AI’s analytical power while retaining crucial insights from radiologists. 

AlSamhori, J.F., AlSamhori, A.R.F., Duncan, L.A., Qalajo, A., Alshahwan, H.F., Al-abbadi, M., Al Soudi, M., Zakraoui, R., AlSamhori, A.F., Alryalat, S.A., and Nashwan, A.J., 2024. Artificial intelligence for breast cancer: Implications for diagnosis and management. Journal of Medicine, Surgery, and Public Health, 3, p.100120. ISSN 2949-916X. https://doi.org/10.1016/j.glmedi.2024.100120.

Amazon (2023). What is Deep Learning? – Deep Learning Explained – AWS.  Amazon Web Services, Inc. https://aws.amazon.com/what-is/deep-learning/#:~:text=Deep%20learning%20is%20a%20method. 

American Cancer Society. (2022). Breast Cancer Risk Factors. https://www.cancer.org/cancer/breast-cancer/risk-and-prevention/breast-cancer-risk-factors.html.

BCRF (2024a). AI Breast Cancer Detection and Diagnosis.  Breast Cancer Research Foundation. Available at: https://www.bcrf.org/blog/ai-breast-cancer-detection-screening/.

BCRF (2024b). AI Breast Cancer Detection and Diagnosis| Breast Cancer Research Foundation.  Breast Cancer Research Foundation. Available at: https://www.bcrf.org/blog/ai-breast-cancer-detection-screening/#:~:text=The%20team%20demonstrated%20that%20MIRAI. 

Bohr, A. and Memarzadeh, K. (2020). The rise of artificial intelligence in healthcare applications. Artificial Intelligence in Healthcare,  1(1), pp.25–60. doi:https://doi.org/10.1016/B978-0-12-818438-7.00002-2. 

Charow, R., Jeyakumar, T., Younus, S., Dolatabadi, E., Salhia, M., Al-Mouaswas, D., Anderson, M., Balakumar, S., Clare, M., Dhalla, A., Gillan, C., Haghzare, S., Jackson, E., Lalani, N., Mattson, J., Peteanu, W., Tripp, T., Waldorf, J., Williams, S. and Tavares, W. (2021). Artificial Intelligence Education Programs for HealthCare Professionals: Scoping Review. JMIR Medical Education, 7(4), p.e31043. Doi: https://doi.org/10.2196/31043.

Christensen, J. (2023). AI-supported mammogram screening increases breast cancer detection by 20%, study finds: CNN Health. https://www.cnn.com/2023/08/01/health/ai-breast-cancer-detection/index.html.

Conner, K (2024). Using AI to Detect Breast Cancer: What We Know: breastcancer.org. https://www.breastcancer.org/screening-testing/artificial-intelligence#articleEndAdvisors).

Dheeba, J., Albert Singh, N., & Tamil Selvi, S. (2014). Computer-aided detection of breast cancer on mammograms: A swarm intelligence optimized wavelet neural network approach. Journal of biomedical informatics, 49, 45-52. 

Enqvist, L. 2023. ‘Human oversight’ in the EU artificial intelligence act: what, when and by whom?: Law, Innovation and Technology* 15(2), pp508–pp535. https://www.tandfonline.com/doi/full/10.1080/17579961.2023.2245683.

Gradishar, W. J., Anderson, B. O., Abraham, J., Aft, R., Agnese, D., Allison, K. H., … & Burstein, H. J. (2020). Breast cancer, version 3.2020, NCCN clinical practice guidelines in oncology. Journal of the National Comprehensive Cancer Network, 18(4), 452-478.

Hendrix, N., Hauber, B., Lee, C.I., Bansal, A. and Veenstra, D.L. (2020). Artificial intelligence in breast cancer screening: primary care provider preferences. Journal of the American Medical Informatics Association,  28(6), pp.1117–1124. doi:https://doi.org/10.1093/jamia/ocaa292.

Iman Hesso, Kayyali, R., Dolton, D.-R., Kyeung Min Joo, Zacharias, L., Charalambous, A., Lavdaniti, M., Evangelia Stalika, Ajami, T., Acampa, W., Boban, J. and Shereen Nabhani‐Gebara (2023). Cancer care at the time of the fourth industrial revolution: an insight to healthcare professionals’ perspectives on cancer care and artificial intelligence. Radiation Oncology, 18(1). doi:https://doi.org/10.1186/s13014-023-02351-z.

Jong Seok Ahn, Shin, S., Yang, S.-A., Park, E., Ki Hwan Kim, Soo Ick Cho, Ock, C. and Kim, S. (2023). Artificial Intelligence in Breast Cancer Diagnosis and Personalized Medicine. Journal of Breast Cancer,  26(5). doi:https://doi.org/10.4048/jbc.2023.26.e45. 

Kiros, H., Langlotz, C (2022). Doctors using AI catch breast cancer more often than either does alone: MIT Technology Review. https://www.technologyreview.com/2022/07/11/1055677/ai-diagnose-breast-cancer-mammograms/#:~:text=Radiologists%20assisted%20by%20an%20AI%20screen%20for%20breast,a%20radiologist%20than%20it%20does%20when%20operating%20solo.

Kooi, T., Litjens, G., van Ginneken, B., Gubern-Mérida, A., Sánchez, C. I., Mann, R., … & Karssemeijer, N. (2017). Large scale deep learning for computer aided detection of mammographic lesions. Medical image analysis, 35, 303-312.

Kopans, D. B. (2014). The 1960s: the foundation for modern mammography. Radiologic Clinics, 52(3), 457-468.

Kuchenbaecker, K. B., Hopper, J. L., Barnes, D. R., Phillips, K. A., Mooij, T. M., Roos-Blom, M. J., … & Antoniou, A. C. (2017). Risks of breast, ovarian, and contralateral breast cancer for BRCA1 and BRCA2 mutation carriers. Jama, 317(23), 2402-2416.

Lang, K., et al. (2023). Artificial intelligence-supported screen reading versus standard double reading in the Mammography Screening with Artificial Intelligence trial (MASAI): a clinical safety analysis of a randomized, controlled, non-inferiority, single-blinded, screening accuracy study. The Lancet Oncology, 24(8), pp.936–944. doi:https://doi.org/10.1016/S1470-2045(23)00298-X.

Lauritzen, D. A., Lillhom, M., Lynge, E., Karssemeijer, N., Vejborg, I (2024). Early Indicators of the Impact of Using AI in Mammography Screening for Breast Cancer: Rsna Radiology,* 311(3). https://pubs.rsna.org/doi/10.1148/radiol.232479.

Lehman, C. D., Wellman, R. D., Buist, D. S., Kerlikowske, K., Tosteson, A. N., & Miglioretti, D. L. (2015). Diagnostic accuracy of digital screening mammography with and without computer-aided detection. JAMA internal medicine, 175(11), 1828-1837.

Leibig, C., Brehmer, M., Bunk, S., Byng, D., Pinker, M., Umutlu, L (2022). Combining the strengths of radiologists and AI for breast cancer screening: a retrospective analysis: The Lancet Digital Health,* 4(7) pp507-pp519. https://www.thelancet.com/journals/landig/article/PIIS2589-7500(22)00070-X/fulltext.

Mango, V. L., Goel, A., Mema, E., Jain, S., Ha, R., & Desperito, E. (2021). Artificial intelligence in breast imaging: current status and future directions. American Journal of Roentgenology, 216(2), 262-272.

Marinovich, L.M. (2023). Artificial Intelligence (AI) for breast cancer screening: BreastScreen population-based cohort study of cancer detection: ebiomedicine, Part of Lancet. https://www.thelancet.com/journals/ebiom/article/PIIS2352-3964(23)00063-4/fulltext.

McKinney, S. M., Sieniek, M., Godbole, V., Godwin, J., Antropova, N., Ashrafian, H., … & Shetty, S. (2020). International evaluation of an AI system for breast cancer screening. Nature, 577(7788), 89-94.

National Cancer Institute. (2021). Breast Cancer Risk in American Women. https://www.cancer.gov/types/breast/risk-fact-sheet.

Office of the Commissioner (2019). FDA permits marketing of artificial intelligence-based device to detect certain diabetes-related eye problems.  U.S. Food and Drug Administration. Available at:https://www.fda.gov/news-events/press-announcements/fda-permits-marketing-artificial-intelligence-based-device-detect-certain-diabetes-related-eye. 

Raya-Povedano, J.L., Romero-Martín, S., Elías-Cabot, E., Gubern-Mérida, A., Rodríguez-Ruiz, A. and Álvarez-Benito, M. (2021). AI-based Strategies to Reduce Workload in Breast Cancer Screening with Mammography and Tomosynthesis: A Retrospective Evaluation. Radiology, 300(1), pp.57–65. doi:https://doi.org/10.1148/radiol.2021203555.

Sweeney, C (2021). Artificial Intelligence’s Promise and Peril: Harvard Public Health. https://harvardpublichealth.org/alumni-post/artificial-intelligences-promise-and-peril/.

Topol, E. J. (2019). High-performance medicine: the convergence of human and artificial intelligence. Nature medicine, 25(1), 44-56.

Viberg Johansson, J., Dembrower, K., Strand, F. and Grauman, Å. (2024). Women’s perceptions and attitudes towards the use of AI in mammography in Sweden: a qualitative interview study. BMJ open, 14(2), p.e084014. doi:https://doi.org/10.1136/bmjopen-2024-084014. 

​​World Health Organization (2019). Cancer. WHO. https://www.who.int/health-topics/cancer#tab=tab_1.