Enhancing Nanoparticle Detection in Interferometric Scattering (iSCAT) Microscopy Using a Mask R-CNN.

Area of Science:

• Optical Microscopy
• Nanotechnology
• Machine Learning

Background:

• Interferometric scattering microscopy (iSCAT) images nano-objects like nanoparticles and viruses.
• High substrate roughness causes background scattering, mimicking nano-objects in iSCAT images.
• Traditional algorithms struggle with accurate particle detection in rough iSCAT imaging.

Purpose of the Study:

• To enhance particle detection accuracy in iSCAT experiments with significant background scattering.
• To develop a machine learning approach for improved object identification in iSCAT.
• To reduce false positives caused by background noise in iSCAT data.

Main Methods:

• Implemented a mask region-based convolutional neural network (mask R-CNN) for particle detection.
• Generated labeled datasets using experimental background images and simulated particle signals.
• Employed transfer learning for efficient mask R-CNN training on limited computational resources.

Main Results:

• Mask R-CNN trained with experimental backgrounds significantly reduced false positives.
• The inclusion of representative backgrounds improved signal differentiation between particles and noise.
• The machine learning approach outperformed traditional Haar-like feature detection in noisy iSCAT data.

Conclusions:

• Supervised machine learning, specifically mask R-CNN, offers a robust solution for iSCAT particle detection challenges.
• A workflow for creating labeled datasets with experimental backgrounds and simulated signals is presented.
• This methodology empowers researchers to improve image processing in iSCAT experiments with challenging backgrounds.