An Automated Algorithm to Identify and Quantify Brown Adipose Tissue in Human 18F-FDG-PET/CT Scans

Objective: To develop an algorithm to identify and quantify BAT from PET/CT scans without radiologist interpretation. Design and Methods: Cases (n = 17) were randomly selected from PET/CT scans with documented "brown fat" by the reviewing radiologist. Controls (n = 18) had no documented `` brown fat'' and were matched with cases for age (49.7 [31.0-63.0] vs. 52.4 [24.0-70.0] yrs), outdoor temperature at scan date (51.8 [38.9-77.0] vs. 54.9 [35.2-74.6] degrees F), sex (F/M: 15/2 cases; 16/2 controls) and BMI (28.2 [20.0-45.7] vs. 26.8 [21.4-37.1] kg/m(2)]). PET/CT scans and algorithm-generated images were read by the same radiologist blinded to scan identity. Regions examined included neck, mediastinum, supraclavicular fossae, axilla and paraspinal soft tissues. BAT was scored 0 for no BAT; 1 for faint uptake possibly compatible with BAT or unknown; and 2 for BAT positive. Results: Agreement between the algorithm and PET/CT scan readings was 85.7% across all regions. The algorithm had a low false negative (1.6%) and higher false positive rate (12.7%). The false positive rate was greater in mediastinum, axilla and neck regions. Conclusion: The algorithm's low false negative rate combined with further refinement will yield a useful tool for efficient BAT identification in a rapidly growing field particularly as it applies to obesity.