Page 50 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

• Safety and Compliance:

o Data Privacy Protection: By combining data perturbation, differential privacy, and
multi-role model rewriting, the project ensures fuzzy encryption of patient privacy while
retaining critical medical information. Multimodal data is linked via unified patient IDs,
session IDs, and timelines under strict privacy protocols, ensuring coherence and
contextual awareness.
o Verification and Testing Platform: The project’s verification platform standardizes
patient creation for real, repeatable testing scenarios, evaluating AI's abilities in
comprehension, response, and healthcare support accuracy. All datasets include
detailed metadata, and accuracy, reliability, and usability are tested across mobile,
web, and smart device platforms to ensure safety, compliance, and clinical value.
o Cautious Human-AI Collaboration: Adds prudent guidance layers on top of large
model outputs, combining expert and AI collaboration, with safety and control
mechanisms in place.
• Doctor Digital Twin for Efficient Communication:

o Edge-Cloud Collaborative AI Avatar: Combining diffusion models with dynamic
neural radiance fields, the project supports highly realistic digital humans generated
from a single photo, achieving an industry-leading low Frechet distance of 17.06. The
self-developed edge-cloud rendering solution cuts concurrent service costs by 90%,
enabling the large-scale application of doctor digital humans in health management
scenarios.
o Audio Interaction with National Renowned Doctors: Trained a medical knowledge-
enhanced voice large model achieving 97% speech recognition accuracy. Proprietary
one-shot voice cloning technology reduced the cost of doctor voice customization by
95%. The end-to-end native voice model keeps duplex conversation response latency
within one second, offering a realistic interaction experience for health management.

Use Case Status: The use case is part of a larger product development

2�2 Benefits of the use case

1. Lowering the Barriers to Healthcare Access and Bridging the Digital Divide
• By leveraging large-scale visual language recognition with trillions of parameters and
advanced learning from vast medical Q&A datasets, multimodal interactions are made
accessible to users of all ages and technical backgrounds. Since its launch in June 2024,
the service has reached over 13 million users, delivered more than 78 million service
instances, and achieved a 98% user satisfaction rate.
• An open collaboration plan for intelligent agents has been introduced to reduce costs
for medical institutions and developers. In terms of current implementation, this case
integrates intelligent agents from local health commissions and hospitals, offering
medical guidance across over 1,000 healthcare facilities.

2. Optimizing Resource Allocation and Advancing Equitable Healthcare
• The system enables patients to quickly identify the most suitable medical resources based
on their symptoms and medical history.
• By collaborating with primary healthcare institutions, it expands access to high-quality
medical applications in remote and underserved areas.

Its deployment in China—a vast, healthcare ecosystem— brings an inspirational approach: AI
may dissolve spatiotemporal constraints in medical resource distribution, ensuring equitable
access to inclusive care for all individuals.

14