Breast cancer is a common disease among women worldwide.¹ Early diagnosis of this disease is crucial for effective treatment and increased patient survival. Mammographic images are a common screening method for breast cancer.^{2, 3} Breast cancer screening requires a thorough visual examination to detect any signs of a mass, and ultrasound alongside mammography increases diagnostic sensitivity.⁴ In a standard clinical examination, a radiologist reads the mammogram and classifies the findings based on the American College of Radiology’s Breast Imaging Reporting and Data System (BI-RADS) vocabulary.⁵

If an abnormal finding is shown in digital mammography, it usually requires diagnostic work, which may involve performing three-dimensional mammography along with other imaging techniques.⁶ If a lesion is suspected to be cancerous, further evaluation with a biopsy is recommended. Analyzing these images is challenging due to limited sensitivity, device or environmental noise, and the subtle differences between lesions and the background fibroglandular tissue, as well as various types of soft and hard lesions.⁷ Sometimes, human vision’s ability to process images is limited, leading to false-negative rates in mammographic images.⁸ Therefore, properly enhancing mammographic images through masks in deep learning techniques can help doctors accurately diagnose the disease.⁹

Breast cancer diagnosis for providing personalized treatment involves various dimensions. To increase accuracy and reduce false-positive and false-negative diagnoses in imaging and biopsy stages, recent advances in artificial intelligence (AI) have been used to develop systems that can assist doctors in clinical practice.^{10, 11} Therefore, in a precise diagnostic process, various factors such as image quality, radiologist and oncologist expertise, breast structure complexity, and the noise level of the imaging device used impact the accuracy of this disease’s diagnosis. Developing decision support systems with advanced and new methods capable of handling noisy data is helpful.^{12, 13} Although classical machine learning is widely used and shows great potential in analyzing medical images, challenges of inadequate data labeling and slow processing speed still exist. Quantum machine learning (QML) can have a transformative effect on computer science and health. Using quantum techniques in classical machine learning can lead to better performance, reduced noise, higher speed, and shorter response times.^15-18 Since all studies have focused on classical AI methods in analyzing breast cancer data, we decided to explore the applications of QML in analyzing breast cancer data. This is because it can help in accurately and early diagnosing breast cancer through increased processing speed and efficiency, reduced data noise, reduced error rates, and ultimately reduced treatment costs.

Traditional machine learning algorithms usually require a large number of computational resources to solve complex problems. For large problems with enormous data, the processing time increases dramatically. The devices and hardware required to implement traditional algorithms are usually available, and it is now easy to use conventional processors or even graphics processing units (GPUs) to accelerate the calculations. Traditional algorithms often struggle at large scales. Processing large amounts of data requires a considerable amount of time and resources. By using quantum properties such as synergy and interference, some problems can be solved faster. But in practice, the development of quantum algorithms for complex problems still brings challenges such as noise reduction and the need for advanced hardware. In QML, many algorithms are still being developed in quantum machine learning. Different qubits may have applications, but their widespread use for various problems is still limited, and their efficiency is not fully proven compared to traditional machine learning.

Considering the capabilities of QML and the large volume of breast cancer medical data, this study was designed to investigate the algorithms used to analyze breast cancer data in order to create a perspective and help researchers in conducting future research.

This study is a systematic review conducted in accordance with the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) 2020 guidelines.¹⁹ In evidence-based medical research, the formulation and design of research questions are critical components for doing a study and providing answers.

The systematic review identified 29 studies that applied AI models to personalized breast cancer management. Of these, 17 studies employed various deep learning methods that yielded favorable results in predicting treatment responses and patient prognosis. The application of these models has been particularly effective in improving diagnosis and personalized treatment for breast cancer patients (Figure 2). Two studies that used neural networks and clustering methods showed strong results in predicting patient survival and classifying breast cancer tumors. Meanwhile, one study successfully implemented transfer learning to predict treatment response, further contributing to the potential of AI in clinical applications. Additionally, quantum machine learning techniques, such as quantum support vector machines (QSVM) and quantum neural networks (QNN), were highlighted as being particularly promising. These approaches demonstrated enhanced performance in analyzing complex and noisy data, with potential improvements in processing speed, accuracy, and error reduction compared to traditional machine learning methods. Quantum convolutional neural networks (QCNN), for example, achieved an accuracy index of 100% in certain diagnostic imaging tasks (Figure 3). Further, studies utilizing quantum transfer learning and quantum kernel methods revealed their potential in improving the diagnostic precision of complex imaging data, as exemplified by accuracy scores exceeding 95% in several studies. Quantum techniques also facilitated rapid analysis of genomic and multi-omics data, effectively reducing noise and improving feature extraction processes.

This capability is crucial in identifying critical biomarkers for early detection and personalized planning in breast cancer.

Collectively, these results highlight the transformative potential of combining quantum computing with AI in the field of oncology. By significantly accelerating data processing speeds and improving model performance, QML could mitigate the limitations of classical machine learning, particularly in managing the vast and complex datasets associated with genomics, histopathology, and radiological imaging (Figure 4).

This systematic review suggests that AI methods, especially quantum algorithms, have great potential to improve the accuracy of breast cancer diagnosis. It was observed that most studies have used advanced AI techniques such as QNN and QSVM to increase the accuracy of breast cancer diagnosis. Many of these studies have been able to achieve high accuracy (over 90%) in the identification and classification of medical images. However, further research is needed to evaluate the performance of these models in real clinical settings. Also, collaboration between researchers and clinicians can help develop more effective solutions in this area. Breast cancer is a genetic disease influenced by an individual’s genomic structure. Genetic data are crucial sources of information for predicting cancer progression. AI systems can analyze patients’ genetic data to identify patterns related to disease progression and assist physicians in treatment decision-making. These analyses can help identify patients at high risk of disease progression and aid in developing personalized treatments. Compared to classical computing, quantum computing technology is more energy-efficient and optimized for developing advanced AI models. This means quantum computing can enhance AI advancements in areas such as deep learning, natural language processing, and computer vision with less energy consumption. QML offers significant potential for effectively analyzing biological and medical data to minimize medical errors.⁸ Therefore, leveraging QML algorithms can enhance the analysis of medical data for early disease diagnosis, leading to improved patient management, cost reduction, and better treatment outcomes.⁹

Based on the provided table and results, it is evident that QML techniques show great promise in enhancing breast cancer diagnosis. We reported a variety of studies employing QML algorithms like QSVM, QNN, and QCNN. These methods were applied across different data types, including genomic data, mammogram images, histopathological images, and magnetic resonance imaging (MRI) data, with a focus on classification, detection, and diagnosis of breast cancer. A significant number of these studies reported high accuracy, often exceeding 90%, and even reaching 100% in some instances, suggesting a potential advantage over classical machine learning methods. For example, Ghoabdi et al. achieved 100% accuracy using QSVM for classification. Similarly, studies using QCNN for detection and diagnosis have also reported accuracy scores close to 100%. The consistency of high accuracy across different QML methods and datasets suggests a strong capability for handling the complex and noisy data associated with breast cancer. Furthermore, the comparison between QML and classical AI methods highlights QML’s potential to address limitations of classical approaches. Several studies mentioned in the discussion section used classical machine learning, such as CNNs, and reported high accuracy, sensitivity, and specificity in breast cancer diagnosis. However, the research suggests that QML may offer an edge in terms of increased processing speed, reduced noise, and better feature extraction. Additionally, the use of quantum transfer learning and quantum kernel methods further improves diagnostic precision from complex imaging data. Overall, while classical methods have shown good results, the potential of QML to enhance speed, accuracy, and handle complex data points toward significant advancements in breast cancer diagnosis and personalized treatment. However, this study also acknowledges that the technology is still in early stages and requires further evaluation in clinical settings.

In addition, Charan et al. (2021) used CNN techniques on 342 mammographic images, resulting in an area under the receiver operating characteristic (ROC) curve of 0.879.⁵¹ Using CNN and CoroNet techniques on 2620 images from the North Florida Digital Imaging Database to diagnose breast cancer, Mobark et al. (2022) reported that the accuracy achieved was 94.92% and 88.67%, respectively.⁵² Meanwhile, some studies that employed QCNN techniques^{29, 49} on diagnostic breast cancer images, an accuracy index of 100% was reported, indicating that QCNN can learn faster or achieve better test accuracy with fewer training sessions.

Also, Li et al. used a CNN on mammographic images of 394 breast cancer patients to assess the risk of breast cancer. The area under the ROC curve in this study was approximately 0.834.³⁰ In a study by Liu and colleagues, a CNN was used on clinical data and mammographic images to diagnose four types of microcalcifications observed in malignancies. The area under the curve, sensitivity, and specificity indices for this model were 0.910, 91.9%, and 85.3%, respectively.⁵⁰ In another study, Fathy et al. used a CNN on 2517 mammographic images of breast cancer patients to identify breast masses, with results showing an area under the ROC curve of 0.965 for mass detection.⁵³ In a study by Altan et al. for classifying breast cancer using CNN on mammographic images, the deep learning model achieved high accuracy, sensitivity, and specificity indices, which were 92.48%, 95.30%, and 96.72%, respectively.⁵⁴ Additionally, Huang et al. designed a deep learning-based decision support system to assess breast cancer risk using CNN and text processing techniques on demographic and clinical data, along with mammographic images of 5107 breast cancer patients. The model demonstrated 100% sensitivity, 74% specificity, 81% accuracy, and an area under the curve of 0.93%.⁵⁵ Similar studies, such as the one by Mohapatra et al., used VGG-16 and AlexNet techniques on 9752 mammographic films for breast cancer mass identification. The results showed an area under the curve (0.86) and an accuracy (65%) for the AlexNet model that were better than VGG-16.⁵⁶ In other similar studies, Khuriwal et al. conducted a study to identify breast cancer in mammographic films using CNN on 4356 films with 12 features, achieving an accuracy of 98%.⁵⁷ Duggento et al. performed a study using a deep neural network like CNN on the DDSM database with 12 features to identify and diagnose breast cancer, reporting an area under the curve of 0.729.⁵⁸ Also, Malebary et al. utilized techniques such as RNN, CNN, VGG, RF, and BT on the public DDSM database to classify mammographic images of breast cancer. The area under the curve for ResNet and ResNet-VGG networks was reported to be 0.923 and 0.958, respectively, indicating superior capability in classifying mammographic images compared to other models.⁵⁹ In another study, Ahn et al. estimated breast density using CNN on 392 mammographic images, reporting a diagnostic accuracy of 96%.⁶⁰ Similarly, Chakravarthy et al. used deep learning techniques such as SVM with RBF kernel, ELM, and PSO on three databases (DDSM, MIAS, IN-breast) for breast cancer identification and reported PSO accuracy rates of 98.13%, 98.26%, and 97.19%, respectively, across the databases, showing superior image classification ability, compared to other models.⁶¹

One-way quantum features can aid machine learning through quantum superposition, allowing the machine to perform different stages of a task simultaneously and in parallel. Such a capability can significantly increase learning speed and efficiency. Although quantum machine learning has numerous advantages, research shows that in some areas, classical machine learning performs better. In this regard, Zheng et al. diagnosed breast cancer using Deep Learning assisted Efficient Adaboost Algorithm (DLA-EABA) and CNN techniques on mammographic, ultrasound, CT scan, and MRI images. The results from DLA-EABA showed an accuracy of 97.45%, sensitivity of 98.3%, and specificity of 96.4%, demonstrating higher capability than other models.⁶²

While quantum computing holds great potential, it is still in its early stages of development and faces many technical challenges. These challenges include qubit stability, quantum errors, and the need for extremely cold environments for quantum systems to function. As research progresses in both fields, the interaction between quantum computing and AI is expected to yield significant results, transforming industries such as medicine, finance, logistics, and information technology. Although evaluating the overall performance of quantum models on a large scale is difficult due to the limited size of quantum circuits, we can be hopeful that it might be reliably achievable in the near future.

Quantum computing is still in its early stages, facing technical hurdles like qubit stability and quantum errors, and requiring extremely cold environments to function. This makes it challenging to implement and evaluate QML models on a large scale. The limited size of quantum circuits makes it difficult to evaluate the overall performance of quantum models on a large scale. This makes it difficult to draw definitive conclusions about their effectiveness. Although many studies have achieved high accuracy using QML, further research is needed to evaluate these models in real clinical settings.

Instead of just lab-based experiments, future studies should focus on integrating QML models directly into clinical workflows. This means testing models in real hospitals with diverse patient populations, not just in controlled environments. This would help assess the practicality and real-world effectiveness of QML in improving breast cancer diagnosis and treatment. Rather than focusing on small-scale quantum circuits, future research should aim to scale up quantum technology to evaluate model performance on larger datasets. This is crucial to assess the reliability of QML in handling complex medical data. Future research should aim to develop more stable and error-resistant quantum systems. This means researching new ways to correct quantum errors and make quantum computers more reliable and resilient for practical applications. Future studies should promote interdisciplinary research and encourage collaboration among quantum computing experts, AI researchers, and clinical practitioners. This will ensure that research is driven by real clinical needs. Rather than sticking to standard quantum algorithms, future research should look to develop novel QML algorithms tailored to specific breast cancer data challenges. This could lead to algorithms that can analyze genomic, histopathological, and radiological data more efficiently. This will help to reveal the unique advantages and disadvantages of each approach and understand which one is best suited for particular tasks. Researchers should focus on using QML to develop personalized treatment strategies based on an individual’s genetic makeup, medical history, and other factors. This could involve analyzing complex genomic and multi-omics data with quantum techniques to identify precise biomarkers for early detection and targeted treatments. Future research should develop novel methods for managing and analyzing large, multidimensional breast cancer datasets. This would allow for the discovery of patterns that could improve diagnosis, treatment, and prognosis. In summary, while QML holds great potential for transforming breast cancer diagnosis and treatment, significant limitations need to be addressed in future research. This includes not only improving the technical aspects of quantum computing but also focusing on real-world clinical applications, interdisciplinary collaboration, and novel algorithm development.

The presence of various signs and features of breast cancer makes it challenging for physicians to diagnose; hence, accurate prediction of this disease is crucial. Breast cancer data is highly complex due to diverse biological characteristics, the abundance of genomic and proteomic patterns, and clinical factors such as patient age, treatment history, and impactful environmental risk factors.

By collecting more precise information, such as gene expression profiles, histopathological images, and treatment responses. The challenge lies in managing and analyzing these large and multidimensional datasets to identify patterns that could improve diagnosis, treatment, and prognosis. Due to the ability of quantum computers to process and represent high-dimensional spaces more efficiently than classical computers, QML offers potential advantages for handling such complex data. QML algorithms, such as QSVM or quantum principal component analysis (QPCA), can identify the most significant features of genomic or imaging data more quickly than classical methods and enhance predictive models.

Not applicable.