In the developing world, breast cancer remains the leading cause of death among women. A breast cancer diagnosis is essential to understand the risk level of cancer in the breast area. Medical professionals and doctors face difficulties such as inaccuracy and time consumption in breast cancer diagnosis due to their manual investigations, which may produce an erroneous outcome.
The main significance of breast cancer treatments is accurate diagnosis and early detection of breast cancer. In recent years, machine learning techniques and computer-aided design have been popularly applied for breast cancer diagnosis. More recently, deep learning technology was also utilized for breast cancer diagnosis, which attains high diagnostic accuracy.
Machine learning technology helps to predict and detect breast cancer more precisely. Medical imaging plays an important role in breast cancer diagnosis, as it imparts valuable information to differentiate breast cancer-s abnormal and normal condition. Machine learning-based breast cancer detection preprocesses breast cancer images, segments the images, extracts the appropriate features, and classifies breast cancer into benign or malignant.
Several imaging modalities have been applied for breast cancer detection and prediction. Such imaging modalities are histopathology images, breast X-ray images, Digital Mammographic Images (DMG), Ultrasound Sonograms (ULS), infrared thermal imaging, Magnetic Resonance Imaging (MRI) and Computerized Thermography (CT), digital breast tomosynthesis, and mass-spectrometry imaging, and their combination as multi-modality detection. Logistic Regression, K-Nearest Neighbor (KNN), Support Vector Machine, Naïve Bayes, Decision Tree, and Random Forest are machine learning classifiers utilized for breast cancer diagnosis.
Histopathology images – Microscopic examination of breast cancer tissue is histopathology. The histopathology image dataset is a well-known dataset for breast cancer diagnosis. The histopathology image dataset utilized for breast cancer detection and classification helps achieve high accuracy and efficiency.
Digital Mammographic Image (DMG) – Mammography is the second most common, widely and comfortable examination utilized imaging technique for diagnosing breast cancer. DMG owns significant characteristics such as manual nature, variability in mass appearance, and minimal signal-to-noise ratio. Screen Film Mammography (SFM), Full-Field Digital Mammogram (FFDM), and Digital Breast Tomosynthesis (DBT) are the categories of DMG.
Ultrasound Sonograms (ULS) – To obtain fast visualization and diagnosis of breast tissue, ultrasound imaging is a noninvasive modality. Ultrasound images utilize high-frequency sound waves for the inherent interpretation of breast tissues. The main benefits of ultrasound images for breast cancer are highly sensitive and better identification of mass location. Ultrasound imaging also plays an important role in breast lesion detection using a computer-aided diagnosis system.
Infrared thermal imaging – Infrared thermal imaging is a noninvasive technique beneficial for recognizing non-palpable breast cancer. When combined with a computer-aided device, infrared thermography provides highly accurate breast tumor detection.
Magnetic Resonance Imaging (MRI) – Magnetic resonance imaging is regarded as a diagnostics image modality that utilizes robust radio waves and magnetic fields to acquire the 3D image of breast tissue and display it in a perfect view for better breast cancer visualization. MRI is highly suggested for extreme risk and inherited breast cancer patents due to its high sensitivity and low specificity.
Computerized Thermography (CT) – Computed tomography is performed as a whole-body examination and utilized for breast cancer staging as the primary purpose. CT yields significant scope in clinical breast cancer analysis and provides 3-dimensional images of the breasts at a radiation dosage.
Mass-spectrometry imaging - Mass-spectrometry imaging technology helps to analyze the plasma and tissue of the patients for breast cancer detection and prognosis. Ambient mass spectrometry imaging (MSI) and liquid chromatography-mass spectrometry are trusted and repeatable methods for breast cancer diagnosis.
Logistic Regression - Logistic regression is applied to many biological studies and is well-known for predictive analyses. A logistic regression algorithm was applied to predict breast cancer by selecting appropriate features with reduced computational cost.
K-Nearest Neighbor (KNN) - KNN is a supervised learning technique that identifies the data-s label before predictions. K-nearest neighbor algorithm is applied for the prediction of breast cancer, in which its dataset is already labeled as malignant and benign. In breast cancer detection, KNN combines multiple classifiers such as Naïve Bayes, Decision Tree, and Random Forest.
Support Vector Machine - As a common machine learning algorithm, the support vector machine achieves better accuracy by increasing the margins between breast cancer classes and is exploited for early detection of breast cancer. In addition to the above machine learning algorithms, extreme learning machine, ensemble approaches, artificial neural networks, fuzzy classifiers, boosting algorithms, and deep learning classifiers are employed for breast cancer detection, prediction, and prognosis.
Data Quality and Bias: The quality of training data is paramount in machine learning. Biased or noisy data can lead to biased or inaccurate predictions. If the training data is not representative or contains errors, the models performance may suffer.
Data Privacy and Security: Medical data, including breast cancer diagnostic data, is sensitive and subject to strict privacy regulations. Ensuring the security and privacy of patient data is a significant challenge in machine learning applications.
Clinical Integration: Integrating machine learning models into clinical practice requires careful planning and validation. Ensuring that the model provides valuable assistance to healthcare professionals and does not disrupt clinical workflows is complex.
False Positives and False Negatives: Like any diagnostic tool, machine learning models can produce false positives and negatives. Balancing sensitivity and specificity is crucial, and achieving a perfect balance can be difficult.
Clinical Expertise: Machine learning models should assist rather than replace clinical expertise. Overreliance on automated diagnoses without the involvement of trained medical professionals can be problematic.
Cost and Resources: Implementing machine learning solutions in healthcare settings can be expensive. It requires resources for data collection, model development, hardware, and ongoing maintenance.
Limited Access to Technology: Access to advanced machine learning technology may not be available to all healthcare facilities, potentially creating disparities in diagnostic capabilities.
Limited and Imbalanced Data:
Obtaining a large and diverse dataset for training machine learning models can be challenging. The distribution of cancerous and non-cancerous cases in medical datasets is imbalanced and leads to biased models.
Data Heterogeneity: Medical data can be highly heterogeneous, coming from various sources and modalities, such as mammograms, ultrasounds, MRIs, and patient records. Integrating and processing diverse data types is a challenge.
Validation and Regulation: Medical devices and diagnostic tools, including machine learning models, often require regulatory approval before clinical use. Complying with regulatory requirements and conducting rigorous validation studies can be time-consuming and costly.
Continuous Learning and Adaptation: Medical knowledge and practices evolve. It must be continually updated to reflect the latest research and clinical guidelines.
Mammogram Analysis: Machine learning models can analyze mammograms to detect abnormalities such as masses, microcalcifications, and architectural distortions. These models can assist radiologists by highlighting suspicious areas for further examination.
Automated Tumor Segmentation: Machine learning algorithms can segment and delineate the boundaries of breast tumors within medical images, facilitating accurate tumor size and volume measurements.
Pathological Image Analysis: Besides medical imaging, machine learning models can assist pathologists in analyzing histopathological slides. These models can identify cancerous tissue and provide insights into tumor grading and subtyping.
Treatment Planning and Personalization: Machine learning can aid oncologists in treatment planning by predicting patient responses to different therapies. Personalized treatment recommendations can improve outcomes and reduce side effects.
Deep Learning Architectures: The exploration of advanced deep learning architectures, including CNNs, RNNs and attention mechanisms, to improve the accuracy of breast cancer detection and characterization from medical images.
Uncertainty Estimation: Developing methods to quantify uncertainty in machine learning predictions is crucial for clinical decision-making and risk assessment.
AI-Enhanced Screening Programs: Developing AI-driven screening programs that optimize the scheduling of mammograms and personalize screening recommendations based on individual risk factors.
Extraction of Quantitative Features: Advancing feature extraction methods that capture quantitative and qualitative characteristics from medical images to improve lesion detection and characterization.
Global Accessibility: Research efforts aimed at making AI-based breast cancer diagnosis accessible and affordable, particularly in resource-constrained healthcare settings and low-income regions.
Longitudinal Data Analysis: Analyzing longitudinal patient data to monitor disease progression, treatment response, and recurrence risk, facilitating personalized treatment plans.
Early Detection and Risk Stratification: Continued research into improving the accuracy of early breast cancer detection focuses on identifying high-risk individuals who would benefit from personalized screening and prevention strategies.
Personalized Treatment Recommendations: Advancing AI-driven models that recommend personalized treatment options based on individual patient characteristics, tumor biology, and treatment response predictions.
Longitudinal Analysis: Investigating machine learning to analyze longitudinal patient data to monitor disease progression, treatment outcomes, and recurrence risk, allowing timely interventions.
Federated Learning and Privacy-Preserving AI: Developing secure and privacy-preserving machine learning techniques (example: federated learning) that enable collaborative model training across healthcare institutions without sharing sensitive patient data.
Global Health Equity: Research aimed at reducing healthcare disparities related to breast cancer diagnosis, particularly in regions with limited access to advanced diagnostic tools and expertise.
