Transformative insights: Image-based breast cancer detection and severity assessment through advanced AI techniques

1 Introduction

Breast cancer is a health issue that is highly prevalent worldwide and affects women. Early identification is crucial because the disease starts in cells and has the potential to spread to other parts of the body. When aberrant cells proliferate and form a mass called a lump, it is critical to identify the condition as soon as possible [1]. There are three main types of breast cancer, all of which have different traits. The milk ducts are the exclusive site of ductal carcinoma in situ, and adjacent tissues are not invaded. Invasive ductal carcinoma is the most common type in which cancer cells invade the surrounding breast tissues. The third type, invasive lobular carcinoma, starts in the lobules of the milk glands and spreads to the surrounding tissues [2]. The main sign of malignant cancer is the presence of masses, which can be identified by identifying lesions that have certain characteristics related to their marginal features and structural formation. Evaluation of these masses is essential for identifying whether cancerous development is potentially malignant [3]. Notably, calcifications, deposits of calcium in the breast tissue, represent the second primary signal. In mammographic images, these calcifications appear as tiny bright dots and are important diagnostic signals. To categorize cancer as benign or malignant, a thorough assessment of its physical dimensions and features is essential [4]. This entails closely examining the characteristics and structural elements of the detected masses or calcifications to determine the type of malignant development. Architectural aberrations are another important marker for breast cancer. These distortions, which are identifiable by anomalous architectural elements, represent aberrations in the normal tissue structures. Finding these distortions is critical to the diagnostic process, as it adds to a thorough comprehension of the traits and behavior of malignancy [5]. Early diagnosis and prompt treatment are essential to lower the mortality rate of breast cancer. Prompt intervention reduces the chance of unfavorable outcomes such as death, and early detection helps with a correct diagnosis. Customized cancer treatment, which adjusts regimens for each patient, requires complex analysis at several levels, primarily related to the tumor’s genetic makeup [6]. Accurate biomarker assessment is essential for making informed therapeutic decisions in breast pathology. The growing intricacy and necessity for precision in the diagnosis of histopathologic breast cancer present difficulties, which are exacerbated by a dearth of pathologists in several regions worldwide. To improve breast cancer diagnosis, computerized image analysis using histopathology appears to be a potential solution [7]. By expediting time-consuming processes, such as biomarker assessment, this technology may help pathologists make diagnoses that are more precise and timely. Proactive health measures are crucial for everyone, especially women [8,9]. The likelihood of early detection and effective treatment is greatly increased by routine screening and self-examination. For patients with breast cancer, early identification improves their quality of life by expanding their treatment options and improving their overall prognosis [10,11].

The likelihood of early detection and effective treatment is greatly increased by routine screening and self-examination. For patients with breast cancer, early identification improves their quality of life by expanding their treatment options and improving their overall prognosis [12]. Breast cancer can only be detected using X-rays for a considerable amount of time. A variety of imaging modalities are used to identify and assess breast abnormalities, assisting medical practitioners in making informed decisions regarding patient care [13]. The main objectives of breast cancer imaging are early detection, precise diagnosis, disease staging, and therapy response monitoring. Mammography, ultrasound, magnetic resonance imaging (MRI), computed tomography (CT), positron emission tomography (PET), and three-dimensional mammography (tomosynthesis) are biomedical imaging modalities [14]. Mammography is a popular method for detecting breast cancer. It entails obtaining X-ray images of the breast tissue to identify anomalies, such as microcalcifications or tumors. It works especially well in detecting breast cancer in its early stages, frequently before a physical examination can detect it [15]. Sound waves are used in ultrasound to provide finely detailed images of the breast tissue. It is frequently used to separate fluid-filled cysts from solid masses or further assess abnormalities found on mammography. Breast biopsies can be more precisely targeted and sampled using ultrasound guidance [16]. Using strong magnets and radio waves, breast MRI produces detailed images of the breast. It is frequently employed in particular contexts, including high-risk individual screening, determining the severity of illness, and gauging the effectiveness of treatment. Breast MRI is especially helpful in identifying multicentric or multifocal illnesses and tiny lesions [17]. When diagnosing breast cancer, CT scans can be helpful, particularly when assessing metastasis, the spread of the disease to other areas of the body. CT is not the primary imaging modality used to assess breast cancer. PET scans are used in breast cancer staging to identify distant metastases and can be used to evaluate the metabolic activity of tissues [18]. Tomosynthesis is a sophisticated type of mammography that captures pictures of the breast from various perspectives. This method reduces false positives, improves the detection of small tumors, and offers a more detailed image. Combining different imaging modalities frequently enables a more thorough evaluation of breast problems, assisting in precise diagnosis and therapeutic planning [19]. For those with breast cancer, early detection by imaging is essential for improving prognosis and outcomes.

This paragraph explains the necessity of sophisticated AI algorithms to improve detection and severity evaluation. Large amounts of data are produced from diverse sources in industries such as healthcare, including genetic data, electronic health records (EHRs), and medical photographs. These complicated data can be effectively analyzed and interpreted by advanced AI systems, assisting in the early detection of diseases and precise assessment of their severity [20]. Artificial intelligence (AI) systems can be trained to identify microscopic trends or abnormalities that may lead to problems. For instance, AI can help with early cancer identification in the healthcare industry, enabling prompt intervention and better results [21]. AI is vital in the age of personalized medicine. AI can provide customized treatment regimens by evaluating personal patient data, such as genetic information, lifestyle, and environmental factors. This increases the disease diagnostic accuracy (Acc.) and guarantees that each patient receives the most effective treatment, which improves outcomes [22]. Tasks that would take too much time or be impracticable for humans can now be automated owing to advanced AI capabilities. This is especially true for medical imaging, where AI can swiftly and precisely analyze hundreds of images to help radiologists diagnose ailments. AI in cybersecurity can automate the detection of possible threats, freeing human experts to concentrate on more difficult problems [23]. Real-time analysis is essential for many applications, including tracking the evolution of diseases and monitoring vital infrastructures. More sophisticated AI systems can continually analyze data streams, provide timely insights, and facilitate quick responses to changing conditions [24]. AI models may continuously learn from fresh data, particularly those that are machine learning (ML)-based. This flexibility is useful in situations in which threats or conditions change over time. For example, in cybersecurity, AI can adapt its detection algorithms based on new threat patterns [25]. In today’s connected world, the amount of data collected is growing and can often be too large for conventional analytical methods. DL and other advanced AI approaches are particularly effective in handling large amounts of data because they automatically identify relevant features and patterns. This makes them ideal for jobs such as picture identification, natural language processing, and predictive analytics [26]. Because of their capacity to handle complicated data, offer early insights, personalize interventions, automate jobs, monitor in real time, adjust to changing situations, and manage massive data effectively, advanced AI approaches are becoming increasingly necessary for detection and severity assessment [27]. These abilities support precise and effective decision-making in several crucial areas [28].

The major contributions are as follows.

• This article consolidates findings from 125 research articles, providing a comprehensive overview of the current landscape of image-based breast cancer detection and severity assessment.

• This review synthesizes the transformative role of AI, covering image processing, ML, and DL, in image-based breast cancer detection and severity assessment.

• This article explores the synergy between advanced image processing and ML, showcasing their combined power in automating medical image examination for nuanced.

2 Image-processing techniques

This section comprises three subsections. The different image-processing methods used for breast cancer analysis are described in Section 2.1. The challenges and limitations of conventional image-processing methods for breast cancer analysis are described in Section 2.2. Common preprocessing techniques to enhance the quality of breast images and their comparison are presented in Section 2.3.

3 Application of ML for breast cancer detection

ML has demonstrated considerable potential in the identification and diagnosis of breast cancer. Large databases of clinical data and medical imaging can be analyzed using ML algorithms to facilitate early diagnosis, risk assessment, and therapy planning. By helping radiologists identify worrisome spots in mammograms, ML algorithms can increase the precision of breast cancer screening. DL architectures, such as CNNs, can be trained to recognize irregularities and subtle patterns in mammograms. Between various preprocessing techniques and their outcomes, ML algorithms can segment and identify lesions in breast MRI and ultrasound images, offering more precise information for diagnosis. To help characterize the type of observed anomalies, ML models can be used to analyze the textural properties of lesions. By examining a variety of variables, including genetic information, family history, and clinical data, ML algorithms can determine an individual’s risk of developing breast cancer. By customizing treatment regimens based on each patient’s unique features, ML can maximize the efficacy of therapies.

Using the Wisconsin Diagnosis Breast Cancer (WDBC) dataset, Sharma et al. [69] presented several ML algorithms, including random forest (RF), KNN, and naïve Bayes, to categorize benign and malignant breast cancer tumors. After assessing recall, Acc., precision, and f1 score, it was discovered that KNN performed the best for classification. Using the same dataset, to classify two different types of breast cancer, benign and Malignant and Bazazeh et al. [70] used SVMs, RFs, and Bayesian networks (BNs). They also evaluated the Acc., recall, precision, and area of the response curve ([receiver-operating curve (ROC)]). The RF technique offers the best ROC performance, whereas the BN performs best in terms of recall and precision. “Using ML techniques, such as RF, KNN, naïve Bayes, SVM, and RF, and Ak [71] created a model to identify breast cancer and determine accuracy. The logistic regression (LR) model that contained all features produced the best results, with a classification Acc. of 98.1%.” In their study, Vaka et al. [72] examined and contrasted threshold, DNSS, FCM, and histo-sigmoid fuzzy clustering. The DNNS was determined to be far superior to current techniques. To identify breast cancer with high accuracy, Alarabeyyat and Alhanahnah [73] created a backpropagation neural network (BPNN), LR, and the BPNN using mammographic pictures to evaluate the number of features that are extracted. Compared with the BPNN, the LR model used a significantly larger number of characteristics. SVM-radial basis function (RBF) provided the highest overall Acc. of 96.33%. Nallamala et al. [74] suggested that the LR algorithm, SVM algorithm, and KNN algorithm classify two different kinds of breast cancer tumors, i.e., benign and malignant using mammographic images. Osareh and Shadgar [75] used SVM, KNN, and probabilistic neural network (PNN) models to distinguish between the benign and malignant tumors of the breast using fine needle aspirates of breast lesions and fine needle aspirates of breast lumps (FNAB). They also compared performance metrics such as Acc., Sens., Spec., and MCC. The proposed method assesses yield decipherability, computation time, and predictive capability. The proposed model achieved 98.1% Acc. Using the University of California, Irvine (UCI) ML library, Omondiagbe et al. [76] used RBF, artificial neural network (ANN), and naïve Bayes classifiers in conjunction with SVM to extract features and minimize their dimensions. SVM achieved a maximum classification Acc. of 98.82%, Sens. of 98.41%, Spec. of 99.07%, and area under the ROC curve of 0.9994, according to calculations of Acc., area under the ROC curve, precision, recall, Sens., Spec., and kappa statistic. The proposed method assesses yield decipherability, computation time, and predictive capability. The proposed model achieved 98.1% accuracy. Using the University of California, Irvine (UCI) ML library, Omondiagbe et al. [76] used RBF, ANN, and naïve Bayes classifiers in conjunction with SVM to extract features and minimize their dimensions. SVM achieved a maximum classification Acc. of 98.82%, Sens. of 98.41%, Spec. of 99.07%, and area under the ROC curve of 0.9994, according to calculations of Acc., area under the ROC curve, precision, recall, Sens., Spec., and kappa statistic. The MLP model performed the best in terms of Acc., precision, and recall of 99.12, 99.00, and 99.00%, respectively, after comparing the Acc., precision, and recall of the classification tests. “To automatically detect breast cancer using a ML algorithm using the Wisconsin Breast Cancer dataset, Dhahri et al. [77] adopted SVM, KNN, DT, gradient boosting classifier (GB), RF, LR, AdaBoost classifier (AB), Gaussian Naive Bayes (GNB), and linear discriminant analysis (LDA) models.” After comparing performance metrics such as Acc., recall, precision, and F1 score, it can be determined that, with appropriate configuration, the algorithms will perform identically. Using the WDBC dataset, Safdar et al. [78] compared ML algorithms to accurately predict breast cancer detection. After evaluating the AUC and ROC, it was discovered that KNN outperformed SVM. – radial based function, simple linear LR model, naïve Bayes, k-nearest neighbour, AdaBoost, Fuzzy unordered role induction technique, DT – J48, SVM, and WBCD datasets were employed by Gbenga et al. [79] in the identification of breast cancer.” After calculating the true positive rate (TPR), false positive rate (FPR), precision, F 1 score, and Acc., it was discovered that the SVM outperformed the others. Using the DDMS dataset, Hussain et al. [80] compared Bayesian, SVM, and decision-tree classifier models. The Acc., Sens., Spec., and FPR were computed. SVM had the lowest FPR, maximum Sens., Spec., and Acc. A state-of-the-art comparison of various ML models for the diagnosis and categorization of breast cancer illnesses is presented in Table 3.

4 Emergence of DL in breast cancer imaging

In the field of breast cancer imaging, DL has advanced significantly and offers promising improvements in diagnosis, treatment, and detection. The analysis of mammograms using DL models has aided the early detection of breast cancer. These models are capable of identifying minute patterns and anomalies that human radiologists can find difficult to recognize. By autonomously identifying and segmenting breast lesions in medical photographs, DL algorithms can expedite diagnosis and lower the risk of human error. Histopathological images of breast tissue can be analyzed using DL algorithms to categorize various subtypes of breast cancer. Appropriate treatment plans must be determined using this information. In the field of breast cancer imaging, DL has advanced significantly and offers promising improvements in diagnosis, treatment, and detection. The analysis of mammograms using DL models has aided the early detection of breast cancer. These models are capable of identifying minute patterns and anomalies that human radiologists can find difficult to recognize. By autonomously identifying and segmenting breast lesions in medical photographs, DL algorithms can expedite diagnosis and lower the risk of human error. Histopathological images of breast tissue can be analyzed using DL algorithms to categorize various subtypes of breast cancer. Appropriate treatment plans must be determined using this information. By lowering the false positives and negatives in breast cancer imaging, DL models seek to increase the overall diagnosis Acc. This is especially crucial for reducing pointless interventions and ensuring that patients receive timely therapy. The speed and Acc. of breast cancer screening programs can be increased by integrating DL algorithms into computer-aided detection (CAD) systems to help radiologists interpret screening mammograms. DL in breast cancer imaging has the potential to significantly enhance personalized treatment plans, early diagnosis, and early detection, all of which can improve patient outcomes. To address these issues and guarantee the responsible application of new technologies in healthcare, continued research and cooperation among physicians, data scientists, and regulatory agencies are important.

To assess the AUC, Sens., and Spec., Shen et al. [84] suggested utilizing VGG16 and ResNet 50-based image classification to classify breast cancer using mammographic images using the DDSM (CBIS-DDSM). ResNet50 performed the best. The relative values of the AUC, Sens., and Spec. were 98, 86.7, and 96.1%, respectively. Utilizing strong computer vision techniques and DL models, Mambou et al. [85] evaluated linear SVM and InceptionV3 for numerous breast cancer diagnosis strategies utilizing a research database (DMR) comprising frontal thermogram images and calculated Acc. and AUC. In comparison, SVM performs better than the others. “Using the IRMA dataset, Ismail and Sovuthy [86] created VGG16 and ResNet50 to test the diagnosis of breast cancer between two DL model networks. The optimal model, VGG16, had an Acc. of 94%, precision of 89%, and recall of 99%. The deep learning assisted-efficient adaboost algorithm (DLA-EABA) for breast cancer detection utilizing MRI images was proposed by Zheng et al. [87]. It attains 97.2% Acc., 98.3% Sens., and 96.5% Spec.” Using synthetic datasets and image datasets related to breast histopathology, Das et al. [88] presented 2-dimensional empirical wavelet transform (2-DEWT) decomposed modes by transforming a one-dimensional signal into two-dimensional images using the deep insight framework and the mean normalization technique. Using full-field digital mammography (FFDM), Bai et al. [89,89] built and compared deep and recurrent neural networks to identify malignancies before clinical indications manifest. It was found that the Acc., Sens., Spec., and precision of the deep neural networks were all higher. Optimized feature extraction and selection were carried out by Sha et al. [90] using the grasshopper optimization method. Before processing, a median filter was used to remove noise from the image and optimize image segmentation using the CNN and DDSM databases. It attained 92% Acc., 97% negative predictive value (NPV), 85% positive predictive value (PPV), 93% Spec., and 96% Sens. in the study by Lotter et al. [91] using an annotation-efficient DL approach with a DDSM pictures, new “maximum suspicion projection” (MSP) images were created from data built tool (DBT) data to detect mammography and digital breast tomosynthesis. It attained a Sens. of 97.7%, Spec. of 99.4%, and AUC of 0.963 ± 0.003 . Using the mammography MIAS database, Khuriwal and Mishra [92] created a CNN to detect breast cancer from histological pictures. The CNN approach achieved 98% Acc. for the 12 features. CNNs were proposed by Salvi and Kadam [93] for the identification of breast cancer using thermal camera sensor image datasets. Using this method, a 98% Acc. was attained. O-net, a mixture of two U-nets connected at the encoding level and disconnected at the decoding level, was developed by Rashed and El Seoud [94] to develop an automated breast cancer diagnosis system employing mammography and the CBIS-DDSM dataset. The Acc. is 95.01%. In a study conducted by Selvathi and Aarthy Poornila [95], breast cancer diagnosis utilizing medical images and the MIAS mammography dataset was compared using CNNs, sparse autoencoders (SAE), and supervised SAEs (SSAE). GoogleNet, AlexNet, VGG16, and FaceNet were examined by Wang et al. [96] to identify metastatic breast cancer from histopathological images. The Acc. rates for GoogleNet, AlexNet, VGG16, and faceNet are 98.4, 92.1, 97.9, and 96.8%, respectively. With 99.67% Acc. rate, Khuriwal and Mishra [97] suggested a CNN for breast cancer diagnosis utilizing a DL algorithm and the Wisconsin Breast Cancer Database. Using the CBIS-DDSM dataset, Ragab et al. [98] employed SVMs and deep CNNs for the diagnosis of breast cancer. SVM obtained an AUC of 94% and an Acc. of 87.2%. Khan et al. [99] used the idea of transfer learning with a standard benchmark dataset to compare GoogLeNet, VGGNet, and ResNet for the detection and classification of breast cancer. The accuracies of GoogLeNet, VGGNet, and ResNet are 93.5, 94.15, and 94.35%, respectively. Using the MIAS dataset, Saber et al. [100] compared Inception V3, ResNet50, VGG-16, VGG-19, and Inception-V2 ResNet. More power is present in the VGG16 model. The F1 score, AUC, Sens., Spec., Acc., and precision were 98.56, 97.25, 97.66, 99.5, and 98.96%, respectively. A comparison of the various DL models for breast cancer diagnosis is presented in Table 4.

5 Integration of imaging and AI in breast cancer diagnosis

The combination of AI [101] with imaging for the diagnosis of breast cancer has demonstrated great potential in terms of enhancing the precision, effectiveness, and customized nature of diagnostic procedures.

• Automated image analysis: Mammograms and tomosynthesis photographs were analyzed using AI algorithms to identify anomalies such as masses, microcalcifications, and architectural distortions [102]. Radiologists can benefit from automated analysis because it shortens interpretation times and highlights areas of concern.

• MRI and ultrasound analysis: AI was utilized to evaluate breast MRI and ultrasound images, offering further data for diagnosis. This covers the distinction between benign and malignant signs, as well as the identification of lesions and characterization of tumors [103].

• CAD: AI-driven CAD systems have been created to assist radiologists in interpreting images. By marking possible ROI, these technologies help radiologists identify anomalies that can be difficult to notice or easily missed [104].

• Quantitative imaging biomarkers: Through radiomics analysis, AI makes it easier to extract quantitative characteristics from medical images. These traits function as imaging biomarkers and offer important insights into the properties and behavior of tumors [105]. Imaging for breast cancer can be made more diagnostic and prognostic using radiomics.

• Integration with clinical data: Imaging data can be integrated with clinical, genetic, and other patient-specific data using AI [106]. This all-encompassing method improves diagnostic precision and aids in customized treatment planning.

• Risk stratification: To categorize patients into distinct risk groups, AI models can evaluate a variety of risk variables and imaging findings [107]. This will facilitate the creation of customized management plans and suitable screening intervals.

• Real-time decision support: A clinical decision support systems (CDSS) driven by AI [108] helps radiologists assess images in real time. It can provide advice based on imaging data analysis and assist medical professionals in making better judgments.

• Telemedicine and remote consultation: AI makes it easier to transmit and analyze imaging data remotely, allowing specialists and healthcare professionals to collaborate. This is especially helpful in places where access to qualified radiologists is scarce [109].

• Continuous learning and model improvement: AI models constantly learn and change in response to fresh information and new trends [110]. Over time, this flexibility helps algorithms perform better and keeps them up-to-date with the most recent advancements in imaging methods and medical understanding.

6 Addressing potential biases and ensuring fairness in AI models for breast cancer detection

As the integration of AI into breast cancer detection and diagnosis becomes more widespread, it is imperative to address potential biases that may exist within AI algorithms. Biases in AI can arise from various sources, including the data used to train the models, the design of the algorithms themselves, and the interpretation of their outcomes.

7 Comparing AI with human radiologists in breast cancer detection and diagnosis

The integration of AI into breast cancer detection and diagnosis has prompted numerous comparisons with human radiologists, aiming to evaluate performance, accuracy, and efficiency. Several studies have sought to measure these aspects, providing valuable insights:

• Performance accuracy: Numerous studies have indicated that AI algorithms can match or even exceed the performance of experienced radiologists in detecting breast cancer. For instance, a landmark study published in Nature by McKinney et al. [111] demonstrated that an AI model developed by Google Health outperformed radiologists in diagnosing breast cancer. The AI system showed a reduction in false positives by 5.7% and false negatives by 9.4% compared to human radiologists.

• Efficiency and speed: AI models provide consistent and rapid analysis of medical images, considerably reducing the time required for diagnosis. A study published by Hirsch et al. [112] revealed that AI-assisted mammograms significantly decreased interpretation time while maintaining diagnostic accuracy. This efficiency is particularly advantageous in high-volume screening programs, allowing for faster patient throughput and timely clinical decision-making.

• Detection of subtle anomalies: AI has demonstrated a remarkable ability to detect subtle anomalies and early signs of breast cancer that might be overlooked by human eyes. A study by Rodriguez-Ruiz et al. [113] in Journal of Clinical Oncology showed that AI systems achieved better sensitivity in identifying early-stage cancers, some of which were initially missed by radiologists. This capability can lead to earlier diagnosis and improved patient outcomes.

• Complementary roles: Rather than replacing radiologists, AI is often seen as a complementary tool that enhances human capabilities. Studies highlight the synergistic effect when AI assists radiologists, combining the strengths of both AI’s precision and radiologists’ clinical judgment. For example, a study by Dembrower et al. [114] found that AI support improved radiologists’ performance, aiding in more accurate and confident diagnoses.

• Reduction of cognitive load: AI algorithms can help reduce the cognitive load on radiologists by automating routine and repetitive tasks, allowing them to focus on more complex cases and clinical decision-making. This aspect was underscored in a study published by Patil et al. [115], where AI-driven CAD systems enhanced radiologists’ diagnostic capabilities by highlighting areas of interest and flagging potential issues.

• Training and continuous improvement: AI systems are capable of continuous learning and improvement, absorbing massive amounts of data over time to refine their diagnostic accuracy. In contrast, while human radiologists benefit from experience and ongoing education, they are limited by the volume of cases they can personally review. A study by Kaul et al. [116] emphasized the potential of AI to rapidly incorporate new findings and guidelines, ensuring it remains at the cutting edge of diagnostic standards.

• Bias mitigation: It is important to note that AI algorithms can inherit biases from the datasets they are trained on. Therefore, rigorous training with diverse, representative datasets is crucial to minimize disparities. Interestingly, studies such as one published by Norgeot et al. [117] suggest that when well trained, AI can help reduce certain types of human biases, providing more uniform diagnostic standards.

In summary, while AI has demonstrated substantial promise in breast cancer detection and diagnosis, it is most effective when used in conjunction with skilled radiologists. Together, they can achieve higher accuracy, improved efficiency, and better patient outcomes. Ongoing research and longitudinal studies will continue to refine these systems, ensuring that the combination of AI and human expertise maximizes diagnostic precision and effectiveness.

8 Impact on clinical decision-making

Clinical decision-making in breast cancer has been significantly affected by the combination of AI and image processing. AI algorithms can improve the Sens. and Spec. of breast cancer detection in medical images such as mammograms and MRIs [118]. This advancement facilitates early diagnosis and improves treatment outcomes. By automatically indicating any anomalies or suspicious areas in medical images, AI-based CAD [119] systems help radiologists. This can improve the diagnostic Acc. and decrease oversight. AI makes it possible to extract and analyze quantitative features from medical images through radiomics. These traits function as imaging biomarkers [120], offer more details regarding the nature of tumors. A more accurate diagnosis and prognosis can be achieved using radiomics. To estimate a patient’s chance of acquiring breast cancer, AI models can examine a variety of imaging data, medical history, and other pertinent data [121]. Customizing screening methods and choosing appropriate therapies are made easier with the use of this information. AI systems can assist in the creation of individualized treatment regimens by predicting molecular subtypes of breast cancer based on imaging characteristics. This method assists medical professionals in selecting treatments with a higher chance of working for particular subtypes. AI can be used to track how a tumor’s properties and response to treatment vary over time. These data will help medical professionals modify their treatment plans according to each patient’s reaction to therapy. AI makes it easier to combine genomic and clinical data with the imaging data. CDSS driven by AI [122] can help physicians make decisions in real time by recommending and supplying pertinent data. This helps medical practitioners make judgments faster and with greater Acc. By automating repetitive and time-consuming operations, AI allows healthcare personnel to concentrate on more intricate patient care responsibilities. Faster diagnosis and treatment initiation may result in increased efficiency [123]. AI makes it easier for specialists to analyze medical images remotely, allowing them to counsel and assist medical practitioners who are located in different locations [124]. This is particularly helpful in places where access to specialized knowledge is scarce. The use of AI and image-processing technologies in breast cancer clinical decision-making has many advantages; however, to ensure their safe and successful integration into standard clinical practice, it is critical to address issues such as data privacy, the interpretability of AI models, and ongoing validation [125]. Encouraging research, cooperation, and regulatory supervision is necessary to fully utilize these technologies for the treatment of breast cancer.

9 Challenges and future directions

Caliber and variety of training data have a significant impact on the effectiveness of AI models. Problems such as unbalanced datasets and differences in imaging methods can affect the generalizability of the models. Clinicians frequently need transparent insights into the decision-making process; hence, the absence of interpretability and explainability in AI models is a barrier [126]. It is difficult to seamlessly integrate AI technologies into current clinical operations. The adoption of AI technologies may be hampered by interoperability problems, resistance to change, and concerns regarding disruption. The effectiveness of AI algorithms may be affected by variations in image quality and format caused by the absence of uniform imaging practices throughout healthcare facilities [127]. Complicated issues surround informed permission, patient privacy, and ethical application of AI in healthcare. Legal and ethical frameworks must be developed in parallel with new technological development.

AI models can be made more robust and reliable by incorporating real-world clinical data and carefully selecting various relevant datasets [128]. Ongoing studies are carried out to create models and explanation methods for AI forecasts that are easier to understand. Gaining trust and easing the incorporation of AI into clinical workflows are needed [129] to provide user-friendly interfaces, guarantee interoperability with current systems, and provide comprehensive training for healthcare personnel, technologists, and healthcare professionals collaborate [130]. Promotion and creation of standardized imaging procedures to guarantee uniformity in data collection may improve the ability of AI models to generalize in various contexts defining precise privacy protection, ethical standards [131], and legal frameworks to control the advancement and application of AI in breast cancer diagnosis. It is imperative to guarantee patient participation in the decision-making procedures [132]. Further, there are other substantial barriers to integrate AI for breast cancer diagnosis.

• Cost-effectiveness of AI in breast cancer screening and diagnosis: the cost-effectiveness of AI technologies in breast cancer screening and diagnosis is crucial for their adoption, especially in low-resource settings. AI-driven tools can potentially reduce the workload on healthcare professionals, decrease diagnostic errors, and improve patient outcomes, thus leading to significant cost savings. By automating routine tasks, AI allows radiologists to focus on more complex cases, improving efficiency and reducing costs associated with delayed diagnoses and treatment. Moreover, the wide-scale adoption of AI can facilitate more equitable healthcare access by providing advanced diagnostic capabilities even in resource-constrained environments, ultimately leading to better healthcare delivery without disproportionately high investments.

• Regulatory landscape for AI in healthcare – the current regulatory landscape for AI in healthcare is rapidly evolving. Regulatory bodies like the fisher discriminant analysis are developing frameworks to ensure the safety, efficacy, and reliability of AI tools. However, the approval process presents challenges due to the dynamic nature of AI algorithms, which continually learn and adapt. Ensuring consistent performance and managing updates without compromising patient safety are critical issues. Additionally, there are challenges related to validating AI models across diverse patient populations and clinical settings to satisfy regulatory requirements. Emerging guidelines aim to balance innovation with patient safety, yet navigating these regulatory pathways remains complex.

• Integration with existing healthcare IT systems: integrating AI tools with existing healthcare IT systems, including EHRs, presents significant challenges in interoperability. Achieving seamless integration requires compatibility between AI applications and diverse healthcare IT infrastructures. Solutions involve developing standardized data formats, using interoperable APIs, and creating vendor-agnostic platforms to facilitate smooth data exchange. Collaborative efforts between technologists and healthcare professionals are vital to designing user-friendly interfaces and ensuring comprehensive training, thereby enhancing the acceptance and practical utility of AI tools in clinical workflows.

• Patient privacy and data security: addressing patient privacy and data security is paramount, particularly given the sensitive nature of health data. AI implementation in healthcare must comply with stringent regulations such as health insurance portability and accountability act and general data protection regulation (GDPR) to protect patient information. The risks associated with data breaches or unauthorized access necessitate robust encryption methods, secure data storage, and anonymization techniques. Additionally, establishing clear consent protocols and engaging patients in data governance practices foster transparency and trust, ensuring ethical handling of health data while leveraging AI’s capabilities.

• Impact on patient outcomes: evidence from various studies demonstrates the positive impact of AI on patient outcomes, including improved survival rates, lower recurrence rates, and increased patient satisfaction. For example, research published in peer-reviewed journals has shown that AI-assisted breast cancer screening can lead to earlier detection, enabling timely intervention and better prognosis. Patient satisfaction tends to improve when AI tools are used, as they contribute to more accurate diagnoses and personalized treatment plans. Longitudinal studies are essential to further quantify these benefits and solidify AI’s role in enhancing patient outcomes.

• Training healthcare providers: effective training for healthcare providers to use AI tools is essential for successful deployment. Barriers to user acceptance include lack of familiarity with AI technologies, concerns about job displacement, and potential technical challenges. Comprehensive training programs that focus on the benefits of AI, practical usage, and troubleshooting can alleviate these concerns. Continuing education and hands-on workshops help build confidence among healthcare providers, fostering a collaborative environment where AI enhances clinical practice rather than disrupts it.

• AI as a decision support tool: AI’s role in breast cancer detection should be viewed as a decision support tool rather than a replacement for human experts. AI can assist radiologists by identifying potential anomalies, analyzing complex imaging data, and providing quantitative insights that complement human expertise. This collaborative approach ensures that AI augments the diagnostic process, increasing accuracy and efficiency without diminishing the value of radiologists. Highlighting AI’s supportive role encourages its acceptance and integration into clinical practice.

• Technological advancements and future research: recent technological advancements in AI and imaging techniques continue to influence breast cancer detection. Innovations such as DL, advanced neural networks, and improved imaging modalities – like high-resolution MRI and 3D mammography – are paving the way for more accurate and comprehensive diagnostic tools. Ongoing research focuses on enhancing the interpretability of AI models, developing robust training datasets, and integrating multimodal data for holistic cancer diagnosis. As these technologies evolve, they hold the promise of transforming breast cancer care, making diagnosis more precise, personalized, and accessible.

Hence, AI presents significant opportunities to advance breast cancer detection and diagnosis; several challenges must be addressed to realize its full potential. Ensuring diverse and balanced training datasets, improving the transparency and explainability of AI models, facilitating seamless integration into clinical workflows, addressing privacy and security concerns, and navigating the regulatory landscape are paramount. Additionally, demonstrating cost-effectiveness, training healthcare providers, and emphasizing AI’s role as a decision support tool are essential steps toward widespread adoption. As technological advancements continue, collaborative efforts between technologists, healthcare professionals, and regulatory bodies will be crucial in shaping the future of AI in breast cancer care.

10 Conclusion

A revolutionary era in the identification and assessment of breast cancer severity has been brought about by the combination of sophisticated AI tools and image-based methods. The transition from conventional techniques to AI-powered technologies has shown the revolutionary changes in breast cancer diagnosis. These technologies have demonstrated greater skill in identifying minor abnormalities, supporting early detection, and offering insights into the severity of breast cancer through the application of ML and DL algorithms. AI integration helps radiologists plan treatments more precisely and individually while also increasing their productivity. Obstacles such as data integrity, comprehensibility, and moral implications highlight the necessity of continuous investigation and cooperation. The field will advance further with the support of explainable AI, multimodal integration, and international data-sharing initiatives, promising more accurate, effective, and patient-centered care for breast cancer patients. The combination of AI with image-based approaches has yielded transformative discoveries that represent a paradigm shift in breast cancer management. These insights have the potential to significantly improve outcomes and eventually save lives.