Abstract
Patients with neuroendocrine tumors of gastro-entero-pancreatic origin (GEP-NET) experience changes in fat and muscle composition. Dual-energy X-ray absorptiometry (DXA) and bioelectrical impedance analysis (BIA) are currently used to analyze body composition. Changes thereof could indicate cancer progression or response to treatment. This study examines the correlation between CT-based (computed tomography) body composition analysis (BCA) and DXA or BIA measurement. 74 GEP-NET-patients received whole-body [68Ga]-DOTATOC-PET/CT, BIA, and DXA-scans. BCA was performed based on the non-contrast-enhanced, 5 mm, whole-body-CT images. BCA from CT shows a strong correlation between body fat ratio with DXA (r = 0.95, ρC = 0.83) and BIA (r = 0.92, ρC = 0.76) and between skeletal muscle ratio with BIA: r = 0.81, ρC = 0.49. The deep learning-network achieves highly accurate results (mean Sørensen-Dice-score 0.93). Using BCA on routine Positron emission tomography/CT-scans to monitor patients' body composition in the diagnostic workflow can reduce additional exams whilst substantially amplifying measurement in slower progressing cancers such as GEP-NET.
Introduction
Patients with gastro-entero-pancreatic neuroendocrine tumors (GEP-NETs) often face long-term clinical management where monitoring nutritional status is vital for survival. Changes in muscle mass (sarcopenia) and fat distribution are critical biomarkers for cancer progression and treatment response. Currently, clinicians rely on two main tools: Dual-energy X-ray absorptiometry (DXA) and Bioelectrical impedance analysis (BIA). While effective, these require additional appointments, extra costs, and—in the case of DXA—additional radiation exposure. Furthermore, these methods can be skewed by a patient's hydration status, which is common in advanced cancer.
Since these patients already undergo regular CT scans for tumor staging, researchers investigated whether deep learning could be used to extract the same body composition insights from existing CT data, creating a "one-stop-shop" diagnostic workflow.
Methods
The study utilized a prospective database of 74 GEP-NET patients. The researchers employed a fully automated Body Composition Analysis (BCA) system powered by a multi-resolution U-Net 3D convolutional neural network.
- Segmentation: The AI segments CT images into seven distinct tissue types: muscle, bone, and five types of adipose (fat) tissue (subcutaneous, visceral, intermuscular, epicardial, and paracardial).
- Quantification: Tissues are identified using Hounsfield Unit (HU) thresholds. Fat is identified between -190 and -30 HU, while muscle falls between -29 and 150 HU.
The network was trained on 300 CT studies and validated against 60 scans manually annotated by experienced human readers. The automated processing was remarkably efficient, averaging about 100 seconds per scan using an NVIDIA TITAN RTX GPU.
Key Results and Insights
The AI achieved a mean Sørensen-Dice score of 0.93, indicating near-perfect agreement with manual human segmentations. Statistical analysis showed strong correlations between the CT-derived Body Fat Ratio (BFR) and measurements from DXA (r = 0.95) and BIA (r = 0.92). The Skeletal Muscle Ratio (SMR) also correlated strongly with BIA (r = 0.81).
Beyond simple correlation, the CT-BCA method offers several advantages over traditional tools:
- Overcoming the BMI Paradox: Unlike Body Mass Index (BMI), this tool can distinguish between muscle and fat, allowing for the identification of "sarcopenic obesity."
- Hydration Independence: Unlike BIA, which is sensitive to water retention, CT-BCA is based on physical volume and density, making it more reliable for patients with fluctuating fluid balance.
- Anatomic Specificity: While DXA provides a 2D projection, CT-BCA provides 3D volumetric data, allowing for granular analysis of visceral versus subcutaneous fat.
Limitations and Future Work
Standard clinical CT protocols often truncate extremities, which led the CT-BCA to slightly overestimate fat ratios compared to whole-body DXA. Additionally, as a single-center study, multi-center validation is required to ensure the AI generalizes across different scanner manufacturers. Future research may focus on developing similar algorithms for MRI to provide superior soft-tissue differentiation without radiation exposure.
In conclusion, automated CT-based BCA offers a highly accurate and efficient alternative to DXA and BIA, potentially improving the quality of life for GEP-NET patients by reducing the number of required hospital visits.
Summary updated