Quality control in structured illumination-based super-resolution FRET imaging via machine learning.

Area of Science:

• Biophysics
• Microscopy
• Computational Biology

Background:

• Structured illumination microscopy (SIM) enhances molecular observation in living cells.
• Reconstruction artifacts in SIM, due to low signal-to-noise ratio (SNR) or out-of-focus data, compromise subsequent Förster resonance energy transfer (FRET) analysis.
• Existing SIM quality metrics do not integrate both SNR and focus information, hindering effective raw data classification for FRET.

Purpose of the Study:

• To introduce SFQC, an ensemble machine learning algorithm for quality control of structured illumination-based super-resolution Förster resonance energy transfer microscopy (SISR-FRETM) raw data.
• To evaluate SISR-FRETM data quality based on both signal-to-noise ratio (SNR) and focus metrics.
• To provide an accurate and efficient method for selecting reliable images in quantitative FRET microscopy.

Main Methods:

• Developed an ensemble machine learning algorithm (SFQC) integrating SNR and focus quality metrics.
• Extracted features from raw SIM data using both SNR and focus quality metrics.
• Trained and ensembled four different classifiers to ensure robust quality control.

Main Results:

• SFQC achieved high F1-scores: 0.93 for focus detection and 0.95 for SNR detection, outperforming conventional SIM quality metrics.
• SFQC demonstrated the fastest processing time compared to existing metrics.
• The algorithm can generate focus error maps for localized error identification and masking.

Conclusions:

• SFQC offers an accurate and rapid solution for quality control in SR quantitative FRET imaging microscopy.
• The algorithm effectively addresses the limitations of current metrics by combining SNR and focus assessments.
• SFQC reduces manual effort in large-scale microscopy image quality control, enhancing research efficiency.