Automatic Tissue Detection for Mass Spectrometry Imaging.

Area of Science:

• Biomedical Imaging
• Computational Pathology
• Analytical Chemistry

Background:

• Mass spectrometry imaging (MSI) enables spatial mapping of biomolecules in tissues.
• Manual tissue delineation in MSI is time-consuming and subjective.
• Automated methods are needed to streamline MSI data preprocessing.

Purpose of the Study:

• To present an end-to-end method for automatic tissue detection in mass spectrometry images (MSIs).
• To develop a robust model for accurate tissue segmentation in MSI data.
• To reduce manual intervention in the MSI analysis workflow.

Main Methods:

• Utilized paired MSI and histological images from the same tissue section.
• Annotated histological tissue masks using QuPath software.
• Mapped histological masks to MSI space via affine transformation using landmarks.
• Developed metabolite-independent MSI representations (total ion current, root-mean-square, Shannon entropy).
• Trained a convolutional neural network (CNN) for tissue detection.

Main Results:

• Achieved high cross-validation performance: accuracy (0.953), precision (0.939), recall (0.923), and Sørensen-Dice (0.930).
• Validated the model on unseen test data from diverse studies, yielding accuracy (0.945), precision (0.965), recall (0.915), and Sørensen-Dice (0.935).
• Demonstrated model robustness across various tissue types, organisms, and spatial resolutions.

Conclusions:

• The proposed CNN-based method effectively automates tissue detection in MSIs.
• This approach significantly enhances the efficiency and objectivity of MSI data preprocessing.
• The developed technique is applicable to diverse MSI datasets, facilitating broader research applications.