Artifact detection in fluorescence microscopy using convolutional autoencoder.

Area of Science:

• Microscopy
• Image Analysis
• Artificial Intelligence

Background:

• Robust artifact detection is crucial for accurate fluorescence microscopy image analysis.
• Automated methods reduce time, cost, and bias while enhancing reproducibility.
• Current AI methods often require large training datasets for known artifact types.

Purpose of the Study:

• To develop an automated method for detecting previously unseen artifacts in microscopy images.
• To eliminate the need for training sets of artifact-laden images.
• To improve the reliability of image analysis in large or time-sensitive datasets.

Main Methods:

• Trained a convolutional autoencoder on artifact-free images using surface-based intensity distribution analysis (sFIDA) technology.
• Detected artifacts by measuring discrepancies between input and reconstructed images.
• Validated the model across multiple datasets and artifact types.

Main Results:

• Achieved an average accuracy of 95.5% in classifying artifacts across different datasets.
• Successfully detected novel artifacts with variations in cause, structure, size, and intensity.
• Demonstrated the model's effectiveness without requiring a specific artifact training set.

Conclusions:

• Convolutional autoencoders offer a lightweight yet effective solution for automated artifact detection in microscopy.
• The proposed method's independence from artifact-specific training enables broad applicability across various microscopy techniques.
• This approach enhances analytical accuracy and reproducibility in fluorescence microscopy image analysis.