@article {Chen1195, author = {Shanshan Chen and Jianqiao Luo and Li Zhao and Janina Pearce and Francesco Celi}, title = {Fast Detection of Brown Adipose Tissue in Cancer Patients by Localizing Supraclavicular Fossae}, volume = {60}, number = {supplement 1}, pages = {1195--1195}, year = {2019}, publisher = {Society of Nuclear Medicine}, abstract = {1195Background: Brown adipose tissue (BAT) is a specialized, highly-metabolic fat that is associated with cancer cachexia. Despite its newly found importance in human metabolism, the prevalence of BAT in cancer patients is currently not well studied. Purpose: This study aims to develop an image-processing pipeline for automatically detecting BAT in 18F-FDG PET/CT images, in order to enable retrospective screening of BAT in a large population of cancer patients. Material: 10 patients{\textquoteright} images were selected (four BAT positive patients, six males, age =52.9{\textpm}24.6 years, weight =78.5{\textpm}21.4 kg) to develop the pipeline processing algorithm. For data collection, patients were administered an intravenous injection of 6.5{\textpm}1.7 MBq/kg 18F-FDG after at least 4 hours of fasting, 60 minutes before either a neck-to-thigh or whole-body scan using a GE Discovery 690 (GE Healthcare, Chicago, IL). The total scanning time per bed was 2.5 minutes. The spatial resolution of the PET was 6mm x 6mm x 6mm, the pixel spacing of the CT was 0.98mm by 0.98mm, and the CT slice thickness was 3.75mm. All data processing was performed in MATLAB 2017a (MathWorks, Inc). Methods: Inspired by the characteristic location of BAT, we localized supraclavicular fossae (SF) by detecting the clavicular bone and the neck region. With the SF region identified, we further detected BAT by identifying adipose tissue on CT images using the -190 ~ -30 Hounsfield (HU) interval, and by identifying the highly-metabolic regions from PET images. Various thresholds for detecting high SUV regions were tested: 1) fixed thresholds of 2g/ml; 2) 20\% 30\%, and 40\% of the global maximum of SUV (SUVmax); 3) 60\% of the global maximum on the coronal view of Maximum Intensity Projection (MIPmax) of PET images. Results: Validating against nuclear medicine specialists{\textquoteright} readings, we evaluated the proposed pipelines with five types of SUV thresholds on 10 cancer patients{\textquoteright} images. Overall, the highest accuracy (90\%, misidentifying only one patient) was achieved using 30\% of SUVmax and 60\% of MIPmax after localizing the SF region. Without SF region detection, the accuracy was diminished to 40\% ~ 60\% due to the false positive regions along the airway and upper thoracic diaphragm. Discussion: In this study, we explored the possibility of automating the detection BAT using threshold-based methods. Leveraging the anatomical landmarks (i.e. the clavicular bone) greatly reduced the false positives in BAT detection, which allows the tolerance of more thresholding methods. The proposed pipeline processing method is fully automated, fast and robust, suitable for retrospectively screening BAT in a large number of cancer patients{\textquoteright} PET/CT images.}, issn = {0161-5505}, URL = {https://jnm.snmjournals.org/content/60/supplement_1/1195}, eprint = {https://jnm.snmjournals.org/content}, journal = {Journal of Nuclear Medicine} }