Automated Deep Learning Segmentation of Cardiac Inflammatory FDG PET

Background: Fluorodeoxyglucose positron emission tomography (FDG PET) with glycolytic metabolism suppression plays a pivotal role in diagnosing cardiac sarcoidosis. Reorientation of images to match perfusion datasets is critical and myocardial segmentation enables consistent image scaling and quantification. However, both are challenging and labor intensive. We developed a 3D U-Net deep learning (DL) algorithm for automated myocardial segmentation in cardiac sarcoidosis FDG PET. Methods: The DL model was trained on 316 patients' FDG PET scans, and left ventricular contours derived from perfusion datasets. Qualitative analysis of clinical readability was performed to compare DL segmentation with the current automated method on a 50-patient test subset. Additionally, left ventricle displacement and angulation, as well as SUVmax sampling were compared to inter-user reproducibility results. Results: DL segmentation enhanced readability scores in over 90% of cases compared to the standard segmentation currently used in the software. DL segmentation performed similarly to a trained technologist, surpassing standard segmentation for left ventricle displacement and angulation, as well as correlation of SUVmax. Conclusion: The DL-based automated segmentation tool presents a marked improvement in the processing of cardiac sarcoidosis FDG PET, promising enhanced clinical workflow. This tool holds significant potential for accelerating clinical practice and improving consistency and quality. Further research with varied datasets is warranted to broaden its applicability.