Exploring AI Approaches for Breast Cancer Detection and Diagnosis: A Review Article

Area of Science:

• Radiology and Pathology
• Medical Imaging
• Artificial Intelligence

Background:

• Deep learning, including convolutional neural networks (CNNs), Vision Transformers (ViTs), and generative adversarial networks (GANs), is transforming breast cancer diagnostics.
• AI integration into radiology and pathology workflows shows potential for enhanced lesion detection, triage, and reduced interpretive variability.

Purpose of the Study:

• To review recent advancements in AI for breast cancer diagnostics across various imaging modalities.
• To highlight the capabilities and challenges of AI in detection, classification, segmentation, and risk prediction.
• To discuss barriers to clinical adoption and propose priorities for responsible AI integration.

Main Methods:

• Synthesis of recent literature on AI applications in mammography, digital breast tomosynthesis (DBT), ultrasound, MRI, and whole-slide imaging.
• Focus on deep learning models like CNNs, ViTs, and GANs for diagnostic tasks.
• Analysis of factors influencing AI performance, generalizability, and clinical adoption.

Main Results:

• AI systems can improve lesion detection and reduce variability when integrated into clinical workflows.
• Performance and generalizability are contingent on dataset quality, heterogeneity, and model calibration, with risks of diminished performance and miscalibration on external data.
• AI models are increasingly used for segmentation and risk prediction, integrating imaging with clinicopathological and genomic data for personalized patient management.
• Generative adversarial networks (GANs) can aid in data augmentation but require careful quality control and bias monitoring.
• Key barriers to adoption include validation issues, domain shifts, reporting variability, interpretability, and regulatory concerns.

Conclusions:

• AI holds significant promise to augment, not replace, clinicians in breast cancer diagnostics.
• Priorities for responsible AI integration include multi-site validation, transparent reporting, bias mitigation, robust calibration, and continuous monitoring for safety and equity.