Review Article

Introduction

Securing access to quality healthcare services, particularly for essential screening tests like mammograms, presents a significant challenge in developing nations. Women often encounter extensive waiting periods, sometimes extending for several meses, to undergo a mammogram. This crucial test plays a pivotal role in the early detection of breast cancer, where early diagnosis is crucial for effective treatment and enhanced survival <sup>1 ➤
1. Henriksen EL, Carlsen JF, Vejborg IM, Nielsen MB, Lauridsen CA. The efficacy of using computer-aided detection (CAD) for detection of breast cancer in mammography screening: a systematic review. Acta Radiol. 2019 Jan;60(1):13–8</sup> . Delays in diagnosis can significantly impact the survival of patients and their well-being, underscoring the importance of early detection. Compounding these challenges is the scarcity of resources and healthcare professionals, which hinders swift and efficient access to preventive care <sup>2 ➤
2. Watanabe AT, Lim V, Vu HX, Chim R, Weise E, Liu J, et al. Improved Cancer Detection Using Artificial Intelligence: a Retrospective Evaluation of Missed Cancers on Mammography. J Digit Imaging. 2019 Aug;32(4):625–37.</sup> . Such constraints underscore the pressing need for improvements in the interpretation of radiological studies and a reduction in the workload of imaging specialists. These improvements would not only optimize interdisciplinary collaboration, but also enhance patient care, particularly for critical screenings like mammograms <sup>3 ➤
3. Jairam MP, Ha R. A review of artificial intelligence in mammography. Clin Imaging. 2022 Aug;88:36–44.</sup> .

Technological advancements, particularly in artificial intelligence (AI), have significantly influenced numerous sectors, including healthcare <sup>4 ➤
4. Jones MA, Islam W, Faiz R, Chen X, Zheng B. Applying artificial intelligence technology to assist with breast cancer diagnosis and prognosis prediction. Front Oncol. 2022 Aug 31;12:980793.</sup> . Yet, in areas such as Latin America, where the healthcare infrastructure and resources may not fully support the integration of advanced diagnostic techniques based on informatic tools, the potential of these innovations seems to be underutilized. A promising application of technology is in computer-aided diagnosis (CAD) for breast cancer screening with mammograms. CAD leverages vast datasets and pattern recognition algorithms to detect anomalies within mammograms, potentially facilitating the early identification of lesions <sup>5 ➤
5. Boumaraf S, Liu X, Ferkous C, Ma X. A New Computer-Aided Diagnosis System with Modified Genetic Feature Selection for BI-RADS Classification of Breast Masses in Mammograms. BioMed Res Int. 2020 May 11;2020:1–17.</sup> . Given that breast cancer ranks among the most prevalent diseases affecting women globally, early detection is vital for enhancing patient survival rates and quality of life. Given these challenges, our study aims to examine the latest developments in artificial intelligence technology as a supplementary tool in CAD strategies for mammograms. Our goal is to foster better interdisciplinary collaboration among clinicians, radiologists, and pathologists through a state of the art (SOTA) review of the topic. By doing so, we anticipate streamlining the diagnostic workflow and elevating the efficiency of breast cancer detection and treatment processes.

1.1. Breast cancer screening

The prevalence of breast cancer among women globally is a pressing concern, with it being the most commonly diagnosed cancer and ranking fifth in terms of mortality rates in developing nations <sup>5 ➤
5. Boumaraf S, Liu X, Ferkous C, Ma X. A New Computer-Aided Diagnosis System with Modified Genetic Feature Selection for BI-RADS Classification of Breast Masses in Mammograms. BioMed Res Int. 2020 May 11;2020:1–17.</sup> . This underscores the urgent need for advancements in diagnostic methodologies to effectively and consistently identify the early signs of this disease. Early detection is essential in reducing mortality rates and catching the disease in its incipient stages, thereby broadening the options for optimal patient treatment and care <sup>2 ➤
2. Watanabe AT, Lim V, Vu HX, Chim R, Weise E, Liu J, et al. Improved Cancer Detection Using Artificial Intelligence: a Retrospective Evaluation of Missed Cancers on Mammography. J Digit Imaging. 2019 Aug;32(4):625–37.</sup> . Consequently, efforts to enhance diagnostic methods should aim not only to enhance accuracy, but also to ensure widespread availability of consistent mammogram interpretations. Several developed countries have initiated additional supplementary mammographic screening initiatives to analyze the effectiveness of early screening. In 2015, an assessment conducted by the International Agency for Research on Cancer analyzed data from 40 aggregated studies in different high-income regions of Europe, Australia, and North America, where such screening programs were implemented. The assessment findings indicated that the implementation of mammographic screening programs resulted in a 23% decrease in breast cancer mortality rates <sup>5 ➤
5. Boumaraf S, Liu X, Ferkous C, Ma X. A New Computer-Aided Diagnosis System with Modified Genetic Feature Selection for BI-RADS Classification of Breast Masses in Mammograms. BioMed Res Int. 2020 May 11;2020:1–17.</sup> .

In light of this challenge, AI and CAD systems hold promise in revolutionizing the breast cancer screening process. These technologies offer substantial advantages by potentially reducing the time required for radiologists to interpret images, thereby supporting healthcare professionals in their diagnostic responsibilities <sup>6 ➤
6. Chougrad H, Zouaki H, Alheyane O. Deep Convolutional Neural Networks for breast cancer screening. Comput Methods Programs Biomed. 2018 Apr;157:19–30.</sup> . Through the integration of AI and CAD systems into these screening protocols, the efficiency and accuracy of breast cancer detection could be significantly enhanced, leading to improved patient outcomes.

1.2 BIRADS in mammograms

To create a line of communication through which both radiologists and other physicians with different clinical specialties can understand each other when talking about the description of mammographic findings, the Breast Imaging Reporting and Data System (BI-RADS) lexicon was developed by the American College of Radiology. The descriptors in BI-RADS have the purpose of classifying the observed masses in a mammography by the radiologist, as shown in (table 1). According to the BI-RADS classification given, it will be given a distinct clinical path of action.

Each one of the categories that findings, will point towards the nature of the lesion or the clinical interpretation. Category 1 and 2 require annual screening, and category 3 a six month follow-up test. Given the large number of masses, which can follow under these categories, the need for a more effective way to evaluate images arises. There have been instances where the variability of observation between radiologists has led to a conflict of opinions in the evaluation of an image and the lesion observed, and this discrepancy could lead to an incorrect classification and skewing of the correct steps of vigilance to be taken <sup>5 ➤
5. Boumaraf S, Liu X, Ferkous C, Ma X. A New Computer-Aided Diagnosis System with Modified Genetic Feature Selection for BI-RADS Classification of Breast Masses in Mammograms. BioMed Res Int. 2020 May 11;2020:1–17.</sup> . In an effort to improve the consistency and accuracy of mammographic interpretation, CAD systems have been developed. CAD systems use algorithms to assist radiologists in detecting and classifying abnormalities in mammograms. These systems can help reduce variability among radiologists and improve the overall quality of mammographic interpretations.

1.3 Computer Aided Diagnosis

Historically, traditional CAD systems have heavily relied on human input, employing predefined parameters to detect anomalies when used in the context of medical images, particularly concerning breast cancer diagnosis. For microcalcifications, these systems analyze high-intensity pixels resembling rods, whereas for masses, they scrutinize specific attributes, such as shapes, textures, gradients, and grayscale levels, as shown in figure 1. CAD systems are categorized into two distinct groups: detection systems, responsible for identifying potential abnormalities, and diagnostic systems, which evaluate the likelihood of disease based on abnormality features <sup>5 ➤
5. Boumaraf S, Liu X, Ferkous C, Ma X. A New Computer-Aided Diagnosis System with Modified Genetic Feature Selection for BI-RADS Classification of Breast Masses in Mammograms. BioMed Res Int. 2020 May 11;2020:1–17.</sup> . However, radiologists retain the final authority in concrete diagnosis and patient management decisions.

While traditional CAD systems have achieved comparable performance to human readers in mass detection <sup>3 ➤
3. Jairam MP, Ha R. A review of artificial intelligence in mammography. Clin Imaging. 2022 Aug;88:36–44.</sup> , they are limited by this need for manually crafted features. In a notable development <sup>6 ➤
6. Chougrad H, Zouaki H, Alheyane O. Deep Convolutional Neural Networks for breast cancer screening. Comput Methods Programs Biomed. 2018 Apr;157:19–30.</sup> , successfully implemented a CAD system based on a Convolutional Neural Network (CNN) to classify benign and malignant breast masses. Their model utilized a combination of low and high-level deep features from different CNN layers, achieving a classification accuracy of 96.7%.

Another CAD system was proposed for the detection and classification of breast masses, utilizing a regional CNN architecture known as You Only Look Once (YOLO) <sup>7 ➤
7. Al-antari MA, Al-masni MA, Choi MT, Han SM, Kim TS. A fully integrated computer-aided diagnosis system for digital X-ray mammograms via deep learning detection, segmentation, and classification. Int J Med Inf. 2018 Sep;117:44–54.</sup> . YOLO is distinguished by its ability to concurrently learn Regions of Interest (ROIs) and their backgrounds. The YOLO-based CAD system is structured around four primary stages: a) preprocessing of mammogram images. b) feature extraction via multi-convolutional deep layers. c) mass detection employing a confidence model. d) breast mass classification facilitated by a fully connected neural network (FC-NN). This innovative methodology aims to bolster the efficiency and precision of breast cancer diagnosis by mitigating certain constraints inherent in conventional CAD systems. Successful application of the YOLO-based CAD system has proven to analyze mammograms with an overall accuracy of 99.7%. As mentioned previously, YOLO generates the confidence probability for each potential ROI that represents the mass position.

1.4 AI and CAD in breast cancer screening

AI and CAD are revolutionizing breast cancer screening by enhancing the accuracy and efficiency of detection. AI algorithms can analyze mammograms, identifying subtle patterns and anomalies that may indicate early stages of cancer, even before they are noticeable to the human eye. CAD systems work in tandem with radiologists, flagging areas of concern for further evaluation. This technology not only improves detection rates but also reduces the chances of false positives, leading to more effective and timely interventions. AI and CAD are transforming the landscape of breast cancer screening, offering a powerful tool in the fight against this disease.

Diagnostic accuracy of mammography is dependent on factors such as breast anatomy, tissue density, and mainly the radiologists' skill and experience. In single-readings (SR), it is estimated that 70% of missed breast cancers (BC) are due to misinterpretation, whereas 30% are attributed to lesions being overlooked. To improve the sensitivity of mammographic screenings, CAD systems have been introduced to highlight suspicious areas, including microcalcifications and masses. The U.S. Food and Drug Administration approved the first CAD system for mammography in 1998. By 2016, the utilization of digital mammography technology had risen up to 91.8% in the USA <sup>1 ➤
1. Henriksen EL, Carlsen JF, Vejborg IM, Nielsen MB, Lauridsen CA. The efficacy of using computer-aided detection (CAD) for detection of breast cancer in mammography screening: a systematic review. Acta Radiol. 2019 Jan;60(1):13–8</sup> .

1.5 Clinical application

The concept of CAD was first introduced by Winsberg in 1967. In the diagnostic process, CAD employs pattern recognition software to identify unfamiliar forms in images for consideration by physicians. Various imaging modalities, including mammography (MM), ultrasonography (USG), computerized tomography (CT), magnetic resonance imaging (MRI), and biopsy, are all utilized by CAD systems for breast cancer diagnosis. CAD shows promise to improve the efficiency and routine of radiologists by saving time between readings and maintaining consistency in lesion recognition. Machine learning is integrated into breast imaging through CAD, serving as a "second pair of eyes" and aiding in the interpretation and processing of accurate medical images. The primary goal of CAD is to reduce human error in observations and minimize false reports during image readings.

CAD has emerged as a major research focus in medical imaging and diagnostic radiology. In this paradigm, radiologists utilize computer-generated outputs as a "second opinion" in making final diagnostic decisions. CAD is a concept that acknowledges the equal roles of physicians and computers, distinct from automated computer diagnosis relying solely on algorithms. The aim is not for computers to outperform physicians, but for their performance to complement that of physicians. CAD systems have found extensive application in assisting physicians, particularly in the early detection of breast cancers through mammograms.

While the current performance of CAD systems is promising, it is not yet sufficient for them to function as standalone clinical detection and diagnosis systems. Various CAD systems designed to aid in the early detection of breast cancer typically progress through three stages: tumor detection, segmentation, and classification based on tumor shape and subtypes, utilizing deep learning models. Initial detection involves identifying the ROI using a faster CNN detector. CAD algorithms heavily depend on mammograms, and efforts have been made to establish breast cancer recognition and grouping images to enhance precision and overcome operator dependence. Despite the significant developments in CAD since the early computer era, challenges persist, in areas such as the algorithmic limitations, assembly of input data, preprocessing, processing, and system assessments.

2. Methodology

The PRISMA methodology <sup>8 ➤
8. Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. Syst Rev. 2021 Dec;10(1):89.</sup> was used, following the recommendations for reviews of literature. Diagram 1 reflects the process of the search stage for this review and includes the search queries used for each database, as shown in table 2.

3. Results

In figure 2, we present an overview of the general concepts found in this work, how they can be orderly understood privileging clinical practice and pertinence. The following sections expand on each one of the elements presented in the figure.

3.1 Clinical methods and performance of CAD

Different methods such as CNN, YOLO-based CAD systems, Full Resolution Convolutional network (FrCN), and traditional CAD systems have been employed. Various datasets including digital database for screening mammography (DDSM), INbreast, Breast Ultrasound Images Dataset (BUSI), Curated Breast Imaging Subset of Digital Database for Screening Mammography (CBIS-DDSM), and private datasets have been used for the studies.

In table 3, it is presented the performance of CAD systems in identifying breast lesions based on BI-RADS categories across different papers. Metrics reported for each algorithm are sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and overall accuracy. Results demonstrate the effectiveness of CAD systems in assisting with detection of breast lesions across various BI-RADS categories.

From a clinical perspective, the papers analyzed offer valuable insights into the advancement of CAD systems for breast cancer detection. A study by Assari et al., in 2022 <sup>9 ➤
9. Assari Z, Mahloojifar A, Ahmadinejad N. A bimodal BI-RADS-guided GoogLeNet-based CAD system for solid breast masses discrimination using transfer learning. Comput Biol Med. 2022 Mar;142:105160.</sup> , focused on solid breast mass classification, employing separate models for each imaging modality and integrating them into a single Bimodal Model (BCNN). Their achievement of 90.38% accuracy and an area under the curve (AUC) of 95.82% signifies promising progress in accurately classifying breast masses, which is pivotal for effective diagnosis and treatment planning.

Similarly, in the study by Boumaraf, S. et al., in 2020 <sup>5 ➤
5. Boumaraf S, Liu X, Ferkous C, Ma X. A New Computer-Aided Diagnosis System with Modified Genetic Feature Selection for BI-RADS Classification of Breast Masses in Mammograms. BioMed Res Int. 2020 May 11;2020:1–17.</sup> , mammographic masses were classified into four assessment categories using an average ensemble of models with an XGBoost classifier. Their comprehensive evaluation encompassing accuracy, AUC, sensitivity, specificity, F1 score, and Matthews Correlation Coefficient (MCC) across different datasets shows the robustness of their approach. This use of separate models for different imaging modalities and subsequent integration into a single model highlights the potential for multimodal approaches to enhance diagnostic accuracy. By leveraging these multiple assessment metrics, their study provides a deep understanding of CAD system performance, essential for its potential future clinical decision-making. Scientifically, both studies contribute to advancing CAD methodologies for breast cancer detection. Integration of diverse imaging modalities enables a more comprehensive assessment of breast masses, potentially reducing false-positive and false-negative diagnoses.

3.2 Advancements and hurdles in the global application of CAD

Recent advancements in CAD systems for breast cancer detection via mammograms are shown in table 4, in which it highlights the importance of screening mammography for early detection of breast cancer, as well as the challenges associated with CAD systems.

Further testing of different CAD models highlights the current status and challenges that they still have in breast cancer detection. While the reported performance of CAD systems is encouraging, with an average area under the ROC curve of 0.86 <sup>10 ➤
10. Ramadan SZ. Methods Used in Computer-Aided Diagnosis for Breast Cancer Detection Using Mammograms: A Review. J Healthc Eng. 2020 Mar 12;2020:1–21.</sup> , they are not yet deemed reliable enough for standalone global clinical use. One of the primary challenges facing CAD systems is the low contrast between normal and malignant breast tissues, particularly prominent in dense breast tissue. As discussed previously, this poses a significant obstacle to accurate lesion detection and classification.

Moreover, the reliability of CAD systems is limited by their dependence on specific datasets for training and tuning, making it challenging to generalize excellent performance results reported in the literature. Addressing the variability in breast densities presents a significant challenge in the area of mammographic analysis. Failure to appropriately calibrate detection models to the specific density of the mammogram can lead to erroneous measurements and diagnostic inaccuracies. This concern underscores the critical need for leveraging a diverse array of AI tool models, meticulously trained on an extensive library of mammographic variations.

Transfer learning stands out as a pivotal strategy in this context, allowing for the optimization of model performance even in scenarios with limited data availability. Notably <sup>6 ➤
6. Chougrad H, Zouaki H, Alheyane O. Deep Convolutional Neural Networks for breast cancer screening. Comput Methods Programs Biomed. 2018 Apr;157:19–30.</sup> , undertook the segmentation and classification of benign and malignant lesions, employing transfer learning methodologies with pre-trained models. Utilizing this information obtained from pre-existing models, there is the opportunity to improve diagnostic accuracy and address the variability in diagnosis associated with heterogeneous breast tissue densities.

Without substantial improvements, particularly through further advancing deep learning techniques and enhanced computational power, CAD systems are relegated to secondary opinion tools within clinical practice. These challenges underscore the pressing need for further research and technological advancements to enhance the reliability and effectiveness of CAD systems for breast cancer detection.

3.3 Clinical applications of CAD systems

An important area where the implementation of CAD would be most beneficial is in reducing the time required for radiologists to evaluate mammograms. For example, senior radiologists with extensive experience who can identify lesions more quickly than their less experienced counterparts, CAD systems can significantly enhance their efficiency. The CAD model provides a rapid analysis, allowing the experienced radiologist to swiftly verify the observation, substantially decreasing the time needed for each mammogram evaluation <sup>11 ➤
11. He Z, Li Y, Zeng W, Xu W, Liu J, Ma X, et al. Can a Computer-Aided Mass Diagnosis Model Based on Perceptive Features Learned From Quantitative Mammography Radiology Reports Improve Junior Radiologists’ Diagnosis Performance? An Observer Study. Front Oncol. 2021 Dec 17;11:773389.</sup> .

However, we must ensure that, even with the faster evaluation of mammograms, the accuracy of diagnoses remains as high as possible to ensure a proper approach to the lesion. Various computational models have demonstrated higher specificity and accuracy compared to human readers using BI-RADS descriptors <sup>12 ➤
12. Patel BK, Ranjbar S, Wu T, Pockaj BA, Li J, Zhang N, et al. Computer-aided diagnosis of contrast-enhanced spectral mammography: A feasibility study. Eur J Radiol. 2018 Jan;98:207–13.</sup> , highlighting the potential of CAD-CESM (Computer-aided detection-contrast enhanced spectral mammography) systems to enhance breast cancer detection specificity. Although the model's sensitivity was lower than that of human readers, future work involving margin assessments demonstrates how sensitivity of CAD-CESM can improve, thereby maintaining high diagnostic accuracy.

A comprehensive overview of the diagnostic performance of CAD systems in different imaging modalities for breast cancer is provided in table 5. It includes data from multiple studies, presenting key metrics such as accuracy, sensitivity, specificity, PPV, NPV, and AUC.

The studies compare the performance of CAD systems with that of experienced radiologists across different levels of expertise. Additionally, it highlights the efficacy of CAD in detecting malignant lesions, aiding in risk assessment and differentiating between benign and malignant breast lesions. These findings are crucial for understanding the potential of CAD as a supportive tool in mammography for opportune breast cancer detection.

Overall, the results from various studies evaluating CAD systems for breast cancer detection showed significant variation in accuracy, sensitivity, specificity, PPV, NPV, and AUC. While some systems demonstrated high accuracy and detection rates, others exhibited lower sensitivity and specificity. These varying results can be attributed to the use of different CAD models, datasets, and evaluation metrics across studies. The variation in these parameters significantly influences the reported outcomes, highlighting once again the challenge in the path toward future global application of CAD systems.

3.4 Integration of AI tools

AI tools have shown promise in breast imaging, particularly in digital mammography and digital breast tomosynthesis, offering a stand-alone method for diagnosis and potentially replacing the need for a second reader <sup>3 ➤
3. Jairam MP, Ha R. A review of artificial intelligence in mammography. Clin Imaging. 2022 Aug;88:36–44.</sup> . Integrating AI systems into routine clinical practice could help radiologists achieve performance benchmarks and improve breast cancer screening <sup>13 ➤
13. Bahl M. Detecting Breast Cancers with Mammography: Will AI Succeed Where Traditional CAD Failed? Radiology. 2019 Feb;290(2):315–6.</sup> . However, developing methods for radiologists to interpret AI decisions will be crucial for efficient radiology practices and better patient care <sup>14 ➤
14. Retson TA, Eghtedari M. Expanding Horizons: The Realities of CAD, the Promise of Artificial Intelligence, and Machine Learning’s Role in Breast Imaging beyond Screening Mammography. Diagnostics. 2023 Jun 21;13(13):2133.</sup> .

The appearance of mammographic lesions can be linked to specific histological information, pointing us towards an estimation of breast cancer stage and risk <sup>15 ➤
15. Hamidinekoo A, Denton E, Rampun A, Honnor K, Zwiggelaar R. Deep learning in mammography and breast histology, an overview and future trends. Med Image Anal. 2018 Jul;47:45–67.</sup> . Advances in technologies such as data analysis of high throughput radiomics features and AI-based deep transfer learning have contributed to the development of numerous CAD schemes, but further research is needed in the area of multimodal AI analysis <sup>4 ➤
4. Jones MA, Islam W, Faiz R, Chen X, Zheng B. Applying artificial intelligence technology to assist with breast cancer diagnosis and prognosis prediction. Front Oncol. 2022 Aug 31;12:980793.</sup> .

One study evaluated an AI CAD system that successfully marked malignancy in digital mammography, particularly detecting invasive lobular carcinoma (ILC) with high sensitivity for specific types of masses and calcifications <sup>16 ➤
16. Arce S, Vijay A, Yim E, Spiguel LR, Hanna M. Evaluation of an Artificial Intelligence System for Detection of Invasive Lobular Carcinoma on Digital Mammography. Cureus [Internet]. 2023 May 9 [cited 2023 Oct 16]; Available from: https://www.cureus.com/articles/144590-evaluation-of-an-artificial-intelligence-system-for-detection-of-invasive-lobular-carcinoma-on-digital-mammography</sup> .

While some articles did not provide specific sensitivity or specificity data (table 6), the overall trend indicates a significant potential for AI in improving breast cancer diagnosis and patient outcomes, highlighting the need for further research and development in this field.

3.5 AI tools use in breast cancer assessment

In table 7, we can observe two studies conducted by Lee S et al. <sup>17 ➤
17. Lee SE, Han K, Yoon JH, Youk JH, Kim EK. Depiction of breast cancers on digital mammograms by artificial intelligence-based computer-assisted diagnosis according to cancer characteristics. Eur Radiol. 2022 Apr 30;32(11):7400–8.</sup> to evaluate the agreement between an AI-CAD program and radiologists in assessing mammographic density and breast cancer abnormality scores. The first study compared assessments from the AI-CAD program with those from radiologists and an automated assessment program using data from 488 patients at Yongin Severance Hospital, while the second study retrospectively evaluated breast cancer abnormality scores and AI-CAD false-negative cases in 896 patients with 930 breast cancers.

Key findings from the first study include:

● Fair agreement between AI-CAD and radiologists in assessing mammographic density.
● Similar agreement between radiologists and a commercial automated density program.

Implications for AI-CAD application to digital mammograms include:

● Improved Screening Outcomes: AI-CAD enhances radiologists' specificity without compromising sensitivity, particularly beneficial for women with dense breasts or undergoing prevalent screening.
● Agreement with Radiologists: AI-CAD shows fair agreement with radiologists in assessing mammographic density, aiding in breast cancer risk assessment.
● Augmented Categorization and Cancer Detection: AI-CAD correlates with elevated abnormality scores, aiding in BI-RADS categorization, identifying additional cancers missed by radiologists.
● Need for Tailored Algorithms: AI-CAD systems should be designed and trained with population-specific algorithms.
● Limited Occurrence of Cancers: Further studies are needed to confirm AI-CAD effectiveness due to the limited occurrence of cancers in actual screening data.

In summary, AI-CAD has the potential to enhance digital mammography screening accuracy and efficacy in conjunction with radiologists, but further research is needed to validate its effectiveness in different populations.

3.6 AI in breast cancer screening reviews

The comparison of CAD use in screening mammography for early breast cancer detection shows varying effects on diagnostic accuracy and recall rates. Studies comparing single reading (read by one radiologist) (SR) with SR plus CAD showed higher sensitivity and/or cancer detection rates with CAD. However, the addition of CAD to SR generally increased the relative risk (RR) and reduced specificity, except in one study. When comparing double reading (read by two radiologists) (DR) with SR plus CAD, there were no significant changes in sensitivity or cancer detection rates. In most cases, adding a CAD system to screening mammography increased the RR, sensitivity and cancer detection rates.

The advantages of temporal analysis in detecting and classifying breast abnormalities using prior mammogram data include providing a clear benefit in identifying breast masses and microcalcifications by leveraging the comparison with prior mammogram data to aid in the detection and classification of abnormalities. However, a limitation arises when a newly developed abnormality lacks sufficient prior mammogram data for comparison, rendering the temporal analysis less effective in such cases. Additionally, while the integration of prior mammogram data holds significant potential for detecting breast abnormalities, the limited scope of large-scale studies restricts the broad clinical applicability of these findings. Therefore, while temporal analysis using prior mammogram data can be beneficial in certain cases, its effectiveness may be constrained by the availability of relevant historical data.

It is important to note that while most studies showed improvements in sensitivity and cancer detection rates with the addition of CAD, there were concerns about reduced specificity and increased RR in some cases. This indicates that while CAD systems may improve the detection of abnormalities, it may also lead to increased false positives.

4. Discussion

CAD systems show a promising future for enhancing mammography interpretation and breast cancer screening. Their integration into radiology practices could significantly improve diagnostic accuracy and recall rates. However, successful implementation requires further research to optimize CAD integration, consider radiologists' perspectives, and ensure acceptance in clinical settings.

Several challenges persist in CAD and breast cancer diagnosis, notably the low contrast between normal and malignant tissues, especially in dense breasts. Despite promising average performance, CAD methods are currently insufficient for standalone clinical use. Significant improvements through advancements in deep learning and computational power are needed to elevate CAD systems to primary diagnostic tools.

The efficacy of breast cancer detection depends on the CAD system's performance, the tested population, and radiologists' expertise. CAD can particularly benefit less experienced radiologists by aiding in the identification of tumors with microcalcifications. Understanding CAD's clinical integration and its impact on healthcare practitioners remains crucial, necessitating further research. Evaluating the costs associated with CAD systems is also imperative for optimizing their healthcare applications. The integration of prior mammogram data in AI tools shows significant potential but is currently limited by the lack of large-scale studies. Initial adoption in clinical settings may serve as a supplementary tool for a second reader. Most studies demonstrate improved recall rates, sensitivity, and cancer detection rates with CAD, indicating its potential for enhanced diagnostic accuracy.

Figure 1

PRISMA diagram of the literature review stage of the project

Conclusión

Further research, especially in organized population-based screening programs with longer follow-up times and digital mammography, is necessary to fully assess CAD efficacy. Validation of CAD algorithms with ample clinical data remains a critical area for future research and improvement.