In our latest ByteSight episode, we sat down with Dr. Rohit Thanki, AI and healthcare data security expert, to explore the biggest hurdles facing AI in medicine today and how we can overcome them.
Key Challenges in AI for Healthcare
-
Data Authenticity & Accessibility: AI models rely on high-quality, representative data; yet data remains siloed, and models trained on one country’s data often fail to generalize globally.
-
Global Inequality in AI Readiness: While countries like the US surge ahead in AI adoption, many regions still struggle with data collection and infrastructure gaps, widening the digital health divide.
-
Regulatory Fragmentation: Varied, sometimes contradictory, regulations across countries (e.g., FDA in the US, MDR/AI Act in Europe, fragmented frameworks in India) pose significant hurdles for startups aiming to scale AI solutions globally.
Data Security & Medical Watermarking
Dr. Thanki spotlighted his pioneering work in medical watermarking, embedding unique, cryptographically-secure identifiers directly into medical images or data. This approach can:
- Protect patient privacy,
- Prevent insurance fraud,
- Authenticate ownership of healthcare data,
- And ensure data integrity during transfer or storage.
He emphasized integrating techniques like cryptography and biometrics to enhance both security and regulatory compliance.
Insights from India & the Middle East
Reflecting on his experience in Gujarat, Dr. Thanki shared how AI research in India often focuses on region-specific diseases like oral cancer and diabetes. However, India’s research landscape remains fragmented, lacking a unified national portal for PhD work or data sharing, limiting collaborative progress.
In contrast, he highlighted Dubai’s rapid AI adoption across government and healthcare services, particularly during the pandemic with AI-powered apps for testing and patient tracking, showcasing how proactive investment in infrastructure can accelerate healthcare AI readiness.
Harmonization & Clinical Applicability
Even with data access, harmonizing data across different formats, demographics, and collection standards is a major challenge. Dr. Thanki stressed leveraging cloud platforms (e.g., Azure, AWS) with certified data-handling tools to support startups lacking large computational resources.
Finally, he underscored the gap between AI development and real-world clinical use. Many AI models remain tested only in limited populations, risking failure when deployed in diverse settings. He called for multi-site clinical validations to ensure robust, equitable AI tools.
Final Thoughts
AI’s potential in healthcare is immense, but to truly benefit patients and doctors:
- We must ensure data is secure, diverse, and harmonized
- Focus on real clinical applicability over flashy demos
- And advocate for streamlined, globally-minded regulations.
AI should empower healthcare professionals by providing certified and trustworthy solutions that ease their workload and enhance patient care, rather than creating additional obstacles.
