New algorithm turns 2-14 second videos into all-in-focus microscopic urinalysis images
Researchers replace time-consuming multi-focus imaging with a quick, hand-focus video and deep learning.
Microscopic urinalysis is a standard diagnostic test that traditionally requires capturing multiple images at different focal planes because urine samples on glass slides have a multi-layer structure—cells at various depths are not all visible in a single lens focus. This multi-image process is time-consuming and limits automation.
To solve this, researchers from (authors: Nema, Aggrawal, Goswami, Gupta, Agarwal) propose recording a short video (2–14 seconds) while the technician manually rotates the focus knob. A novel reconstruction algorithm then synthesizes an all-in-focus image from the video frames. Finally, a deep learning model detects and classifies urine sediments on that single image. Proof-of-concept tests on 14 videos from a real lab environment demonstrate the pipeline's viability, paving the way for faster, automated urinalysis.
- Replaces multi-image focal plane stacks with a single 2-14 second video recorded during manual focus adjustment.
- Novel reconstruction algorithm creates an all-in-focus image from video frames, enabling deep learning-based sediment detection.
- Validated on 14 real-world videos from a laboratory technician, showing practical feasibility for automated urinalysis.
Why It Matters
Speeds up routine urinalysis, reduces technician effort, and enables reliable automation in diagnostic labs.