Abstract
We introduce Multi-Organ Objective Segmentation (MOOSE) software that generates subject-specific, multi-organ segmentations using data-centric AI principles to facilitate high-throughput systemic investigations of the human body via whole-body PET imaging. Methods: Image data from two PET/CT systems (uEXPLORER and Siemens TruePoint TrueView) was used in training MOOSE. For non-cerebral structures, a total 50 WB-CT images were used; 30 of which were acquired from healthy controls (HC, 14M/16F) and 20 datasets were acquired from oncology patients (14M/6F). Non-cerebral tissues consisted of 13 abdominal organs, 20 bone segments, subcutaneous fat, visceral fat, psoas, and skeletal muscle. An expert panel performed manual segmentation of all non-cerebral structures except for subcutaneous, visceral fat, and skeletal muscle, which was semi-automatically segmented using thresholding. A majority-voting algorithm was used to generate a ‘reference standard’ segmentation. From the 50 CT datasets, 40 were used for training and 10 for testing purposes. For cerebral structures, 34 18F-FDG PET/MRI brain image volumes were used from 10 HC (5M/5F imaged twice) and 14 non-lesional epilepsy patients (7M/7F). Only 18F-FDG PET images were considered for training: 24/34 and 10/34 volumes were used for training and testing, respectively. The dice score coefficient (DSC) was used as the primary and the average symmetric surface distance (ASSD) as a secondary metric to evaluate the automated segmentation performance. Results: An excellent overlap between the reference labels and MOOSE-derived organ segmentations was observed: 92% of non-cerebral tissues showed DSC values of >0.90, while a few organs exhibited lower DSC values (e.g., adrenal glands (0.72), pancreas (0.85), and bladder (0.86)). In comparison, the median DSC values of brain subregions derived from PET images were lower. Only 29% of the brain segments had a median DSC of >0.90, while segmentation of 60% of regions yielded a median DSC of 0.80-0.89. Results of the ASSD analysis demonstrated that the average distance between the reference standard and the automatically segmented tissue surfaces (organs, bones, brain regions) lies within the size of image voxels (2mm). Conclusion: The proposed segmentation pipeline MOOSE allows automatic segmentation of 120 unique tissues from whole-body 18F-FDG PET/CT images with high accuracy.
- Image Processing
- PET/CT
- Research Methods
- artificial neural networks
- automated segmentation
- multi-organ segmentation
- systems biology
- total-body PET imaging
- Copyright © 2022 by the Society of Nuclear Medicine and Molecular Imaging, Inc.