Assessment of Adipose Tissue PET Activation Using Non-Contrast CT Imaging

Overview

This study demonstrates that a deep learning model can accurately predict brown adipose tissue (BAT) metabolic activity from standard non-contrast CT scans, providing PET-like insights without the need for radiotracers. The model showed strong agreement with PET-derived standardized uptake values (SUV) across independent cohorts, suggesting a radiation-sparing alternative for BAT assessment.

Background

Brown adipose tissue plays a critical role in energy metabolism and cardiometabolic health. Traditional detection of BAT activity relies on 18F-FDG PET imaging, which is costly, involves radiation exposure, and is impractical for large-scale screening. Non-contrast CT scans are routinely performed in clinical practice but lack direct metabolic information. Leveraging deep learning to estimate metabolic activity from CT could enable widespread, non-invasive assessment of BAT and related metabolic conditions.

Data Highlights

The study utilized paired PET/CT imaging data from two independent cohorts: one from Beth Israel Deaconess Medical Center (BIDMC) and a publicly available dataset from The Cancer Imaging Archive (TCIA). A conditional Generative Adversarial Network (cGAN) was trained to predict standardized uptake values (SUV) within adipose tissue regions identified on CT. The model incorporated a fat-focused loss function to enhance metabolic signal estimation. Predicted activations demonstrated strong reproducibility across anatomical regions and datasets, closely matching PET-derived values.

Key Findings

• A deep learning cGAN model can predict BAT metabolic activity from non-contrast CT scans with high accuracy.
• Predicted standardized uptake values (SUV) showed strong agreement with PET-derived measurements.
• The model was validated across two independent cohorts, confirming reproducibility.
• Incorporation of a fat-focused loss function improved metabolic signal estimation.
• This approach eliminates the need for radiotracers and additional radiation exposure inherent to PET imaging.
• Potential to enable large-scale, population-based studies of BAT and metabolic health using routine CT scans.

Clinical Implications

This method offers a practical, radiation-sparing alternative to PET for assessing brown adipose tissue activity, facilitating broader clinical and research applications. It enables metabolic evaluation of adipose tissue using existing non-contrast CT scans without additional imaging burden or cost. This could improve screening and monitoring of metabolic health and disease progression in diverse patient populations.

Conclusion

Deep learning-based prediction of adipose tissue metabolic activity from non-contrast CT scans provides a feasible and reproducible alternative to PET imaging. This innovation may expand access to BAT assessment and support metabolic health research without added radiation or cost.

References

1. Cypess AM et al. 2009 -- Identification and importance of brown adipose tissue in adult humans
2. Magudia K et al. 2021 -- Population-Scale CT-based Body Composition Analysis Using Deep Learning
3. Diaz AA et al. 2014 -- Chest CT Measures of Muscle and Adipose Tissue in COPD
4. Zheng K et al. 2024 -- CT-based muscle and adipose measurements predict prognosis in digestive malignancy
5. Wibmer AG et al. 2021 -- Brown adipose tissue is associated with healthier body fat distribution