Medical Imaging Compute Allocation for Low-Latency AI Workloads
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing medical imaging systems face inefficiencies in managing local compute instance (LCI) resources, leading to underutilization and increased costs due to suboptimal task allocation, especially when network connections fail or demand varies across different imaging modalities.
Innovation Solution
A resource allocation system dynamically manages tasks across a network of medical imaging systems, prioritizing local compute instances (LCIs) for latency-sensitive operations and redistributing tasks using a centralized controller, incorporating AI models and fallback strategies to optimize resource use and reduce processing times.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If tasks are performed at remote cloud compute instances, then computational power and processing capability are improved, but data transmission time and network dependency increase
Solution Approach 1:
The patent segments processing tasks into two categories: latency-sensitive tasks performed locally at LCIs and computationally intensive tasks performed remotely at CCI. This segmentation allows the system to optimize for both speed (local) and power (remote) simultaneously, resolving the contradiction between transmission time and computational capability.
Solution Approach 2:
The edge compute instance acts as an intermediary between the imaging system and cloud compute instances. It receives data locally, performs immediate processing of time-sensitive tasks, and only transmits non-time-critical data to the cloud, thereby reducing overall data transmission time while maintaining access to cloud computational power.
2Loss of time
If tasks are performed at local compute instances, then data transmission time is reduced, but resource underutilization and costs increase
Solution Approach 1:
The system creates a multi-functional architecture where LCIs handle time-sensitive processing and CCI handles computationally intensive tasks. This universal resource pool allows optimal utilization of both local and cloud resources, preventing underutilization while maintaining low latency performance.
Solution Approach 2:
The system dynamically assigns tasks to appropriate compute instances based on task characteristics, available resources, and current system state. This dynamic allocation ensures that LCIs are not overloaded beyond their capacity while CCI resources are fully utilized, optimizing overall resource productivity.
3Adaptability or versatility
If AI models are deployed at cloud compute instances, then model capabilities and processing intelligence are improved, but network dependency and system availability risks increase
Solution Approach 1:
The patent implements local quality by deploying simplified or fallback versions of AI models at edge compute instances, while maintaining full-capacity AI models at cloud instances. This allows the system to maintain basic AI functionality locally (improving reliability) while accessing advanced cloud-based AI capabilities when available (maintaining adaptability).
Solution Approach 2:
The system prepares fallback AI models and processing capabilities in advance at local edge instances. When network connectivity is available, the system uses cloud AI models; when connectivity fails, it seamlessly transitions to pre-prepared local fallback models, ensuring continuous operation and maintaining system availability.
4Power
If more computational hardware is invested, then processing capability and AI model performance are improved, but device complexity and environmental impact increase
Solution Approach 1:
The patent merges local edge computing resources with remote cloud computing resources into a unified hybrid system. This combination allows the imaging system to leverage existing local hardware for immediate processing while accessing additional cloud computational power when needed, avoiding the need to invest in expensive, complex local high-performance hardware.
Data Source
AI summary
Methods and systems are provided for increasing an efficiency of use of computing resources of a plurality of connected imaging systems, based on adaptive prioritization of tasks within and between compute instances and centralized task assignment. When a scan is not being performed using an imaging system, computing resources of the imaging system may be operated in a cooperative mode, where the computing resources may be advantageously used to perform processing tasks for other imaging systems of the connected imaging systems. Computing resources may be allocated based on compute time, data transmission time, and availability of compute instance. In particular, cloud-based AI models used by an imaging system may be integrated more efficiently. For example, local versions of such models may be cached, for operation in the event of a failure of external network connections.