Breast cancer detection using deep learning: Datasets, methods, and challenges ahead

Area of Science:

• Oncology and Medical Imaging
• Artificial Intelligence in Healthcare
• Biomedical Data Science

Background:

• Breast Cancer (BC) is a leading cause of mortality in women, necessitating early detection for improved outcomes.
• Traditional analysis of radiographic and histopathological images for BC is costly and prone to errors.
• Advancements in imaging technologies like Mammography, CT, MRI, and Histopathological Imaging aid early diagnosis.

Purpose of the Study:

• To critically analyze existing research on AI for breast cancer detection and classification using various imaging modalities.
• To review Machine Learning (ML), Deep Learning (DL), and Deep Reinforcement Learning applications in BC.
• To identify publicly available datasets and discuss future research directions and limitations.

Main Methods:

• Comprehensive literature review of ML, DL, and Deep Reinforcement Learning studies for BC detection and classification.
• Analysis of research utilizing diverse imaging modalities: Mammography, Histopathology, Ultrasound, PET/CT, MRI, and Thermography.
• Review of publicly available datasets for BC imaging research.

Main Results:

• DL models demonstrate significant potential in BC identification and prognosis using imaging data.
• AI has shown promising results primarily in retrospective studies, requiring external validation.
• A critical discussion of current limitations and future prospects for AI in BC diagnostics is presented.

Conclusions:

• AI, particularly DL, shows promise for enhancing breast cancer detection and classification accuracy.
• External validation is crucial for translating AI tools into clinical decision-making support.
• Further research is needed to address limitations and explore the full potential of AI in breast cancer diagnostics.