Kaohsiung Medical University Success Story: AI-Stack Enables Computing Power Management and Practical Implementation in Smart Healthcare

Computational Requirements and Challenges in Smart Healthcare

With the rapid proliferation of AI technology in healthcare, effectively harnessing vast medical datasets has become a core challenge in the digital transformation of medical institutions. Kaohsiung Medical University Chung Ho Memorial Hospital (KMU) has long been a pioneer in medical innovation, actively integrating AI technology to achieve outstanding results across clinical decision-making, patient care, and public health promotion. However, AI models demand substantial computational support. How to flexibly allocate and efficiently manage computing resources has become a critical issue that KMU’s AI team must overcome in driving this transformation.

Client Background and Core Challenges

Kaohsiung Medical University and its affiliated healthcare system have garnered repeated recognition in the field of smart healthcare in recent years:

• HIMSS EMRAM Level 7 Certification: The first healthcare institution in southern Taiwan to achieve this highest honour, demonstrating leadership in digital transformation.
• KMU Genie Smart Hospital Platform: Integrating multiple clinical AI models, pioneering the ‘AI-based Attending Physician System’ to enhance clinical applications and talent development.
• AI Smart Health Avatar Programme: Extending AI applications to primary chronic disease care, establishing a new model for community health.

However, the challenge underlying these innovative applications lies in simultaneously supporting multiple AI projects while maintaining data security and computational efficiency. As AI models for diverse purposes are progressively developed and deployed, Kaohsiung Medical University urgently requires a solution to address the pain points of frequent GPU resource sharing, inefficient resource scheduling, and an excessive IT management burden.

Solution: Digital Infinity AI-Stack Solution

In response to Kaohsiung Medical University’s requirements for GPU computing power and AI infrastructure management, Digital Infinity facilitated the implementation of AI-Stack to achieve the following objectives:

• Centralised computing resource management: A unified platform consolidates all GPU resources, streamlining IT management processes.
• Intelligent resource scheduling: Flexibly allocates computational resources based on project requirements, preventing resource conflicts or idleness.
• Support for large-scale AI training: Concurrently handles diverse tasks including clinical AI models, generative AI, and medical record data analysis.
• Enhanced operational efficiency: Enables AI teams to focus on model development and clinical implementation, rather than being distracted by development environment setup and cumbersome processes.

Achievements and Impact

The implementation of AI-Stack enables Kaohsiung Medical University Hospital to flexibly allocate computing power on a single platform. Whether for emergency AI, pathology AI, nursing AI assistants, or cross-hospital database analysis (diabetes, kidney disease, etc.), it receives immediate and stable computational support. Furthermore, AI-Stack’s computational resource scheduling capabilities have enhanced the efficiency of model development within KMU’s AI development team. The MLOps functionality provided by AI-Stack has also supported KMU’s AI-Specialist Physician Programme, enabling promising young physicians to focus on AI model research and development. These models can then be rapidly deployed on KMU’s Genie platform, thereby enhancing the clinical resilience of the smart hospital.

Through AI-Stack’s efficient infrastructure management and MLOps solutions, KMU has achieved outstanding results in advancing smart healthcare:

• Significantly enhanced computational efficiency: Optimised GPU resource utilisation, reducing idle time.
• Rapid AI model deployment: Shortened model development and deployment cycles, enabling swift responses to medical demands. Thirty-nine AI models have entered clinical application, including emergency diagnostics, medication decision-making, critical care, and AI-assisted nursing.
• Enhanced clinical efficacy: Reduced diagnostic times, more precise treatment decisions, and significantly lowered risks of medical errors.
• Extended community care: The AI Health Avatar programme has been successfully implemented in primary care settings, improving self-management capabilities for chronic disease patients.

Conclusion

Through the implementation of Digital Infinity’s AI-Stack solution, Kaohsiung Medical University has not only achieved optimal allocation of GPU resources but also established a comprehensive ecosystem spanning model development to clinical deployment. Digital Infinity’s AI-Stack stands as the pivotal enabler driving Kaohsiung Medical University Hospital’s intelligent healthcare transformation. From the successful deployment of 39 AI models, to the innovative implementation of an AI-assisted specialist physician system, and extending to community healthcare applications, the hospital’s success demonstrates the critical importance of appropriate infrastructure management for AI healthcare applications. This collaboration has not only elevated the quality and efficiency of healthcare services but also set a benchmark for Taiwan’s smart healthcare development. It demonstrates the profound impact achievable when technological innovation and clinical needs are seamlessly integrated.