IMC-Denoise

IMC-Denoise enhances Imaging Mass Cytometry (IMC) images by removing hot pixels and shot noise to improve signal-to-noise ratio and enable more accurate cell-scale phenotyping and spatial analyses.

Key Features:

• Content-aware automated pipeline: Performs automated, content-aware denoising tailored to IMC images.
• Dual-method approach: Combines a differential intensity map-based restoration (DIMR) algorithm with a self-supervised deep learning algorithm (DeepSNiF).
• Differential intensity map-based restoration (DIMR): Removes hot pixels using a differential intensity map-based restoration algorithm.
• DeepSNiF: Employs a self-supervised deep learning algorithm to filter shot noise.
• Adaptive hot-pixel and background removal: Provides adaptive removal of both hot pixels and background noise across channels.
• Quantified performance gains: Reports an 87% reduction in noise levels, a 5.6-fold increase in contrast-to-noise ratio, and approximately a twofold improvement in F1 score for background noise removal.
• Applicability across pathologies: Demonstrated performance across datasets derived from multiple pathologies.
• Performance on challenging samples: Effective on technically challenging samples including human bone marrow.
• Improved downstream cell-scale analyses: Enhances manual gating and automated phenotyping at the cell scale and supports spatial and density assessments of targeted cell groups.
• Validation: Analyses and phenotyping improvements verified through manual annotations.

Scientific Applications:

• Cell phenotyping: Enables more accurate automated and manual cell phenotyping at single-cell resolution in IMC datasets.
• Spatial analysis: Supports spatial assessments of cell populations within tissue microenvironments.
• Density analysis: Facilitates density-based analyses of targeted cell groups.
• Analysis of complex tissues: Applicable to characterization of complex tissue microenvironments across multiple pathologies.
• Challenging sample analysis: Improves data quality for technically challenging specimens such as human bone marrow.
• Downstream mass cytometry applications: Enhances reliability of downstream mass cytometric analyses that depend on high-quality IMC images.

Methodology:

The pipeline combines a differential intensity map-based restoration (DIMR) algorithm for hot-pixel removal with a self-supervised deep learning algorithm (DeepSNiF) to filter shot noise in an automated, content-aware workflow.