Multi-marker Similarity Enables Reduced-Reference and Interpretable Image Quality Assessment in Optical Microscopy

Area of Science:

• Optical Microscopy and Biomedical Imaging

Background:

• Optical microscopy is crucial for biological and biomedical research, requiring robust image quality assessment (IQA) for data validation.
• Current full-reference IQA methods often rely on pixel-wise comparisons, lacking perceptual agreement and comprehensive feature analysis.
• Existing methods struggle with interpretability and overlook global image quality aspects, necessitating improved assessment techniques.

Purpose of the Study:

• To develop an advanced multi-marker similarity method for reliable and interpretable optical microscopy image quality assessment.
• To address limitations of current pixel-wise comparison methods by focusing on standard quality markers.
• To enable automatic evaluation of large biomedical datasets and facilitate reduced-reference IQA implementations.

Main Methods:

• Developed a novel multi-marker similarity approach comparing standard quality markers: resolution, signal-to-noise ratio, contrast, and high-frequency components.
• Calculated individual similarity scores for each marker against a ground truth.
• Combined marker scores into an overall similarity estimate for comprehensive IQA.

Main Results:

• The multi-marker similarity method provides a full-reference IQA estimate with enhanced global feature extraction and artifact detection.
• The approach yields reliable and interpretable image quality rankings.
• Enabled reduced-reference implementations, using a single field of view as a benchmark for multiple measurements.

Conclusions:

• The multi-marker similarity method offers a significant advancement in optical microscopy image quality assessment.
• This technique ensures reliable evaluation of experimental results and facilitates automatic analysis of large datasets.
• The method's focus on quality markers rather than direct image distances improves interpretability and applicability in biomedical studies.