Efficient Compression of Mass Spectrometry Images via Contrastive Learning-Based Encoding

Area of Science:

• Computational Biology
• Medical Imaging
• Data Science

Background:

• Mass spectrometry imaging (MSI) generates large datasets, posing significant storage and computational challenges.
• Existing methods for MSI data analysis are often limited by data size, hindering widespread application of advanced techniques.
• Efficient data compression is crucial for unlocking the full potential of MSI in diagnostics and research.

Purpose of the Study:

• To introduce a novel encoding algorithm for compressing mass spectrometry imaging data.
• To reduce storage requirements for MSI data while retaining essential diagnostic information.
• To enable advanced analytical techniques like t-SNE on previously computationally prohibitive datasets.

Main Methods:

• Developed a contrastive learning-based encoding algorithm to compress MSI data into fixed-length vectors.
• Tested the algorithm on diverse datasets, including mouse bladder and human Barrett's esophagus biopsies.
• Evaluated segmentation accuracy using traditional k-means and a proposed iterative k-means algorithm on raw and encoded images.

Main Results:

• The encoding algorithm successfully reduced MSI data size significantly.
• Encoded images maintained crucial diagnostic information, comparable or superior to raw data for segmentation tasks.
• Reduced data size enabled the application of t-SNE for enhanced tissue analysis, overcoming previous computational limitations.

Conclusions:

• The proposed contrastive learning algorithm offers an effective solution for MSI data compression.
• This method preserves data integrity, facilitating accurate segmentation and deeper tissue understanding.
• The open-source Python implementation facilitates broader adoption and further development in MSI analysis.