The World Health Organization (WHO) has just published the guidelines our team has been contributing to over the past years, highlighting pivotal regulatory aspects concerning artificial intelligence (AI) in the realm of health. The aim is to establish AI safety and effectiveness in healthcare and start a dialogues between various stakeholders - from developers, manufacturers, regulators, health practitioners to end-users like patients.

With the awakening move towards "data is the new code", switching the focus from model-centric to data-centric AI, anchoring AI systems in robust data foundations is paramount. This ensures not only the safety and efficacy but also timely deployment to critical care areas. As the medical data keeps growing in size and volume, while the machine learning analyses methods improve, AI-based instruments could accelerate and improve the industry. However, this calls for more regulations on AI in healthcare.

6 key areas the publication outlines

1. Transparency and documentation. Documenting the full product lifecycle and monitoring development procedures should evoke trust

2. Risk management. Matters such as 'intended use', 'ongoing learning', human interactions, training models, and cybersecurity risks need thorough attention, striving for simplicity in models

3. External data validation and clarity on the intended use of AI. This helps assure safety and facilitate regulation

4. Data quality. This is vital to ensuring systems do not amplify biases and errors

5. Privacy and data protection. The intricacies of significant complex regulations – like Europe's GDPR and the USA's HIPAA – are tackled, highlighting the importance of grasping jurisdictional scope and consent requirements

6. Collaboration. Collaborative efforts among regulatory bodies, patients, healthcare experts, industry stakeholders, and governmental allies can aid in maintaining product and service compliance throughout their lifecycle

‍

Resources:

Access the guidelines.

Read more in the WHO’s official press release.

‍