Multi-element microscope optimization by a learned sensing network with composite physical layers.

Area of Science:

• Digital imaging
• Machine learning
• Microscopy

Background:

• Digital microscopes are increasingly used for automated image analysis without human intervention.
• Optimizing microscope settings is crucial for enhancing automated interpretation tasks.

Purpose of the Study:

• To jointly optimize microscope settings and classification networks for improved automated image analysis.
• To investigate the interplay between programmable illumination and pupil transmission for automated tasks.

Main Methods:

• Developed a "learned sensing" approach to jointly optimize multiple microscope settings and a classification network.
• Experimentally imaged blood smears for automated malaria parasite detection.
• Compared multi-element learned sensing with single-element optimization.

Main Results:

• Multi-element learned sensing outperformed single-element optimization for automated malaria parasite detection.
• Low-resolution (20X-comparable) images achieved classification performance matching high-resolution (100X-comparable) imagery.
• The optimized settings provided sufficient contrast for machine learning networks.

Conclusions:

• Joint optimization of microscope settings and machine learning networks enhances automated image analysis.
• Learned sensing offers a path toward accurate automation over large fields-of-view.
• Low-resolution imaging can be effective for automated microscopy tasks.