CHALLENGE
Towards faster image processing
Grundium designs and manufactures microscope slide scanners for healthcare and veterinary use. To produce images of pathology samples, the scanner captures thousands of individual captures and merges them into a single large image. Every pixel is processed numerous times, requiring parallel computing in the device.
Fast image processing is essential for delivering positive user experience. Customers appreciate when images are generated quickly after inserting a slide into the device. The speed of image generation depends on two main factors: how quickly images are captured, and how fast all the image data can be processed.
SOLUTION
Optimizing algorithms with AI
Grundium has made significant improvements to the overall speed of its microscope slide scanners. However, image processing remained the primary bottleneck. Grundium partnered with Unikie to focus on optimizing parallel processing.
With AI-powered performance optimization, we were able to make numerous small optimizations to the image contrast enhancement algorithm.
Some of the optimizations performed:
- Move parameters from global to constant memory
- Improve memory access coalescing
- Use fused multiply-add instructions
- Enable loop unrolling
- Precompute bounds checks
- Precomputed repeated calculations
- Tune launch configurations
- Balance computation and register usage
RESULTS
AI optimization delivers 2× code efficiency and 3× faster development
As a result of the collaboration, we were able to achieve 1.99x speedup in the image contrast enhancement algorithm. For Grundium’s customers, this results in added value and better user experience for everyone using the scanners.
With AI-assisted optimization, Grundium can also accelerate the optimization process and bring products to market faster.
Humans remain in the loop, but with AI handling most of the work, algorithm optimizations can be completed in 3x faster compared to manual coding. This means shorter time-to-market and quicker product updates throughout their lifecycle.
Watch Unikie’s webinar AI-assisted Software Engineering to learn more about the case.
