U-Net-based SUV calculation in FDG-PET imaging of mice brain for enhanced analysis

Abstract Positron emission tomography (PET) is a widely used imaging modality in nuclear medicine for a variety of applications. Amongst the methods used for the quantifying and interpretation of PET images, the standardized uptake value (SUV) is a widely-adopted semi-quantitative tool that supplements visual understanding with quantitative information. SUV is used in both clinical and preclinical practices to report the status of various normal organs and tumors under investigation using PET imaging. While the determination of SUVs is typically done manually, which can be tedious, artificial intelligence (AI) can be utilized to enhance the efficiency of the process. In this study, a U-Net-based approach was employed for semi-automated determination of SUV in FDG-PET scans of mice brains. First, a U-Net model was trained using 50 FDG-PET images of six mice to perform the automatic segmentation task. The trained model then delineated the brain of a mouse which was then processed by a short in-house code to extract data and calculate the SUV. The process was also replicated in a manual way for comparison purposes. The comparison of the results from the U-Net-based segmentation method and the conventional manual method at nine different time points revealed that there were errors of less than 4.5% in eight out of the nine-time points. Although our U-Net model’s performance needs improvement, adapting a well-trained AIbased approach for SUV determination, particularly in preclinical studies, can help reduce the workload of organ delineation and minimize associated errors, facilitating SUV determination.