Identifying active Brown Adipose Tissue (BAT) typically uses costly and radiation-exposing [¹⁸F]-FDG PET/CT imaging. This study proposes using convolutional neural networks (CNNs) to predict [¹⁸F]-FDG uptake from unenhanced CT scans. Experiments on four cohorts showed that CNNs, particularly the Attention U-Net, achieved 23% to 40% better BAT segmentation accuracy than conventional CT thresholding. BAT volumes from segmentations distinguished subjects with and without active BAT (AUC 0.8 vs 0.6 for CT thresholding). This suggests CNNs can improve large-scale, cost-effective BAT imaging studies using only CT.

Ertunc Erdil, Anton S. Becker, Moritz Schwyzer, Borja Martinez-Tellez, Jonatan R. Ruiz, Thomas Sartoretti, H. Alberto Vargas, A. Irene Burger, Alin Chirindel, Damian Wild, Nicola Zamboni, Bart Deplancke, Vincent Gardeux, Claudia Irene Maushart, Matthias Johannes Betz, Christian Wolfrum, Ender Konukoglu