Although multispectral optoacoustic tomography (MSOT) significantly evolved over the last several years, there is a lack of quantitative methods for analysing this type of image data. Current analytical methods characterise the MSOT signal in manually defined regions of interest outlining selected tissue areas. These methods demand expert knowledge of the sample anatomy, are time consuming, highly subjective and prone to user bias. Here we present our fully automated open-source MSOT cluster analysis toolkit Mcat that was designed to overcome these shortcomings. It employs a deep learning-based approach for initial image segmentation followed by unsupervised machine learning to identify regions of similar signal kinetics. It provides an objective and automated approach to quantify the pharmacokinetics and extract the biodistribution of biomarkers from MSOT data. We exemplify our generally applicable analysis method by quantifying liver function in a preclinical sepsis model whilst highlighting the advantages of our new approach compared to the severe limitations of existing analysis procedures.
Keywords: AUC, Area under the curve; Biomarkers; DAG, Directed acyclic graph; DL, Deep learning; Deep learning; GUI, Graphical user interface; ICG, Indocyanine green; ImageJ plugin; MSE, Mean squared error; MSOT, Multispectral optoacoustic tomography; Mcat, MSOT cluster analysis toolkit; Multispectral optoacoustic tomography; PCI, Peritoneal contamination and infection; Pharmacokinetics; Quantitative image analysis; ROI, Region of interest; Sepsis; WAC, Weighted-average curve.
© 2022 The Authors.