De-MSI: A Deep Learning-Based Data Denoising Method to Enhance Mass Spectrometry Imaging by Leveraging the Chemical...

Area of Science:

• Biomedical imaging
• Computational biology
• Analytical chemistry

Background:

• Mass spectrometry imaging (MSI) visualizes ion distribution in biosamples.
• Data denoising improves MSI quality but faces challenges due to complex noise and lack of ground truth.
• Existing denoising methods struggle with inherent MSI data complexities.

Purpose of the Study:

• Introduce De-MSI, a novel deep learning method for MSI data denoising without ground truth.
• To enhance the quality of MSI data for improved biosample analysis.
• To provide a reliable computational alternative to experimental noise reduction methods.

Main Methods:

• Developed De-MSI, a deep learning model for MSI denoising.
• Leveraged prior mass spectrometry knowledge from noisy MSI data to create training datasets.
• Trained a deep neural network to remove noise and improve image quality.

Main Results:

• De-MSI demonstrated superior performance in visual inspection of MALDI-acquired mouse fetus data.
• Achieved a mean PSNR of 18.93 and SSIM of 0.74 across ion images.
• Successfully enhanced high-resolution (5 μm) mouse brain MSI data and denoised rat brain DESI data.

Conclusions:

• De-MSI effectively denoises MSI data without requiring ground truth.
• The method shows adaptability across different ionization techniques (MALDI, DESI) and sample types.
• De-MSI is a promising tool for efficient MSI data analysis and interpretation.