Interpreting MALDI imaging data for rare types of ampullary cancer using machine learning
Patrick M Jensen1,2, Jan Lellmann3, Christian Sperling4,5
1Department of Applied Mathematics and Computer Science, Technical University of Denmark, Lyngby, Denmark. patmjen@dtu.dk.
NPJ Systems Biology and Applications
|May 12, 2026
Summary
This study introduces machine learning (ML) with matrix-assisted laser desorption/ionization (MALDI) time-of-flight (TOF) imaging to improve rare ampullary cancer diagnostics. The method aids in identifying prognostic factors and developing new diagnostic solutions for rare tumors.
Area of Science:
- Oncology
- Proteomics
- Computational Biology
Background:
- Rare tumor diseases present diagnostic challenges due to a lack of routine procedures.
- Comprehensive identification of prognostic target proteins or transcripts is crucial for rare cancers.
- Ampullary adenocarcinomas represent a rare cancer requiring advanced diagnostic approaches.
Purpose of the Study:
- To investigate proteomic differences in ampullary adenocarcinomas using machine learning (ML) and MALDI time-of-flight (TOF) imaging.
- To establish MALDI Imaging as a complementary diagnostic tool for immunohistochemical analysis.
- To develop a MALDI Imaging neural network for broad tumor diagnostics.
Main Methods:
- Analysis of a cohort of ampullary adenocarcinomas (intestinal, pancreatic, unknown subtypes).
- Pathological assessment and immunohistochemical staining of human formalin-fixed paraffin-embedded (FFPE) tissues.
- MALDI Imaging detection followed by ML-based analysis and model explainability.
Main Results:
- Identified proteomic differences within the studied ampullary adenocarcinoma cohort.
- Developed a MALDI Imaging neural network applicable to tumor diagnostics.
- Determined influential m/z-values using ML model explainability tools.
Conclusions:
- MALDI Imaging, enhanced by ML, offers a powerful complement to immunohistochemistry for rare cancer diagnostics.
- The developed neural network facilitates broad application in tumor diagnostics.
- Enabling ML network transformation across proteomic data sources supports future rare cancer patient data collection.
