ROIforMSI

ROIforMSI integrates mass spectrometry imaging (MSI) and hematoxylin and eosin (H&E) histology to define regions of interest (ROIs) by optimizing spatial segmentation and selecting the number of clusters (#Clusters) via cross-modal consistency.

Key Features:

• Multimodal Fusion Strategy: Integrates MSI mass spectral data with H&E-derived histomorphological feature spectra for joint segmentation.
• Deep Learning-Based Algorithm: Uses deep learning to extract histomorphological feature spectra from H&E images for downstream clustering.
• Cross-Modal Consistency: Performs clustering on both MSI pixels and histology-derived features and assesses consistency between modalities to validate segmentations.
• Objective #Clusters Selection: Identifies the most probable number of clusters (#Clusters) by maximizing consistency between MSI and histology segmentations.
• ROI Generation: Generates ROIs from clusters that exhibit maximal cross-modal consistency to ensure biological relevance.

Scientific Applications:

• Spatial lipidomics, metabolomics, and proteomics: Automates generation of reliable ROIs for downstream spatial omics analyses.
• Validation on biological specimens: Applied to mouse kidney and renal tumor specimens to produce multimodally corroborated ROIs aligned with ground truth.

Methodology:

Deep learning extracts histomorphological feature spectra from H&E images; MSI pixels and histology-derived features undergo clustering to produce spatial segmentations; cross-modal consistency between MSI and histology segmentations is evaluated to determine and select the optimal #Clusters; ROIs are generated from clusters with maximal cross-modal consistency.