A new paradigm for clinical biomarker discovery and screening with Mass Spectrometry through biomedical image analysis principles

Biomarker discovery in amenably sampled body fluids has the potential to empower clinical screening programs for the early detection of disease. Liquid Chromatography interfaced to Mass Spectrometry (LC-MS) has emerged as a central technique for sensitive and automated analysis of proteins and metabolites from these clinical samples. However, the potential of LC-MS as a precise and reliable platform for discovery and screening is dependent on robust, sensitive and specific signal extraction and interpretation. The output of LC-MS is formed as a set of quantifiable images containing thousands of biochemical signals regulated in disease and treatment. We propose to tackle this problem for the first time with a biomedical image analysis paradigm. A novel workflow of image reconstruction, groupwise image registration and Bayesian functional mixed-effects modeling is presented. Poisson counting noise and lognormal biological variation are modeled in the raw image domain, resulting in markedly improved detection limit for differential analysis.