Efficient Denoising of Shot-Noise in Mass Spectrometry Images by PCA-Assisted Self-Supervised Deep Learning.

Mass spectrometry imaging (MSI) often suffers from inherent noise due to signal distribution across numerous pixels and low ion counts, leading to shot noise. This can compromise the accurate interpretation, especially for trace molecules. Recent advances in self-supervised deep learning denoising have demonstrated significant potential for enhancing data quality. In this Letter, we propose an optimized approach for using the Noise2Void (N2 V) algorithm for MSI denoising by applying a principal component analysis (PCA) preprocessing step. By rotating the data along its principal components prior to denoising, our method, Principal Component-Assisted Noise2Void (PCA-n2v), outperforms direct N2 V implementations and other state-of-the-art denoising techniques. The limitations of PCA-n2v are also evaluated using a synthetic MSI data set, revealing that bleedthrough artifacts may arise in images with extremely low signal-to-noise ratios. To facilitate adoption, an easy-to-use PCA-n2v implementation is provided via a GitHub repository. Overall, PCA-n2v advances MSI data processing, enabling higher-throughput and higher-resolution experiments with improved fidelity.