Abstract
Artificial Intelligence (AI), particularly deep learning for medical image analysis, offers a promising tool to address global inequities in early cancer diagnosis. This article assesses the feasibility and early effectiveness data of deploying AI tools (e.g., for cervical cytology, mammography, dermatoscopy, and histopathology) in low-resource settings. We argue that for this technology to bridge—not widen—global cancer care disparities, major challenges of data bias, infrastructure dependency, clinical workflow integration, and local validation must be proactively addressed.
Proposed Structure
- Introduction: The global cancer burden and diagnostic inequity; AI as a potential equalizer.
- Exemplar Applications: AI for smartphone-based cervical screening; AI-assisted reading for lung nodules in LDCT; digital pathology with AI support for remote diagnosis.
- Evidence from Real-World Pilots: Summarizing results from feasibility studies in Africa, South Asia, etc.
- Critical Implementation Challenges:
- Data & Algorithmic Bias: Lack of diversity in training data.
- Infrastructure: Needs for stable power, internet, and hardware.
- Human-in-the-loop: Augmenting, not replacing, local health workers.
- A Framework for Ethical Deployment: Proposing a framework for inclusive data共建, appropriate technology, rigorous local evaluation, and sustainable business models.
Key References
- McKinney, S. M., et al. (2020). International evaluation of an AI system for breast cancer screening. Nature, 577(7788), 89-94.
- Asiedu, M. N., et al. (2023). A prospective evaluation of an AI-assisted smartphone-based system for cervical cancer screening in Ghana. The Lancet Digital Health, 5(5), e297-e306. (Example of a recent prospective study in a relevant journal)
- Challen, R., et al. (2019). Artificial intelligence, bias and clinical safety. BMJ Health & Care Informatics, 26(1), e100081.
