As a young boy in hospital with appendicitis, I was fascinated when the doctor pressed an ultrasound probe against my stomach. He was looking at something on the screen that I couldn’t see. This early fascination with imaging technologies has never left me and is now being rekindled by the integration of artificial intelligence (AI) into ultrasound systems.
Thanks to advanced Computer-on-Modules with Intel Core processor technology, AI models can now run directly on the ultrasound device. They make use of both the integrated graphics unit (iGPU) and the new integrated neural processing units (iNPU). Each of these units fulfills specific requirements and offers distinct advantages.
| The iGPU unit: |
The iNPU unit: |
- Ideal for general graphics and image processing tasks
|
- Specialized for neural networks and deep learning workloads
|
|
|
- Increases efficiency and reduces latency
|
- Accelerates both conventional and AI-based algorithms
|
- Offloads specialized workloads from the CPU and GPU
|
The combination of these technologies enables complex image analyses to be performed in just milliseconds. The interplay between iGPU and iNPU is crucial: While the iGPU handles general image processing tasks, the iNPU optimizes neural networks and deep learning applications.
This synergy allows doctors to detect even the smallest changes in tissue in real time – an especially critical factor in cancer diagnostics. Surgeons can receive immediate indications of abnormalities during procedures, increasing both safety and precision.
Advances in ultrasound technology, particularly through AI integration, represent a turning point in medical diagnostics. What was once a vision of the future is becoming reality: running AI directly at the edge. We are at the dawn of an exciting new era in medical imaging, and I look forward to the breakthroughs still to come.
That said, not all AI is created equal. If you’d like to learn more about what kind of AI you actually need, I recommend my colleague Florian's blog: https://blog.congatec.com/en/who-needs-muscles-when-you-have-brains
And if you’re wondering how to best evaluate AI performance, my colleague Max has some excellent tips for you: https://blog.congatec.com/en/decoding-ai-hardware-performance-usable-metrics-beyond-tops
