Enhanced Tumor Detection in 18F-FDG-PET via Liver SUV Normalization Using Blood Tests

Overview

This study developed a machine learning-based method to normalize liver standardized uptake values (SUV) in 18F-FDG-PET imaging using blood test results and demographic data. By creating personalized Z-score maps of liver SUV, the approach improved detection accuracy of hepatic tumors compared to conventional SUV analysis.

Background

18F-FDG-PET/CT is widely used to evaluate neoplasms by measuring glucose metabolism via standardized uptake values (SUV). However, liver SUV is influenced by multiple factors including blood glucose, BMI, and liver function test results, complicating tumor detection. Normalizing liver SUV using clinical variables could enhance tumor visualization and diagnostic accuracy. This study aimed to develop a fully automated method to generate normalized liver SUV Z-score maps using machine learning and LASSO regression based on blood tests and demographic data.

Data Highlights

Key Findings

• Machine learning and LASSO regression effectively estimated average liver SUV and standard deviation from blood tests and demographic variables.
• Personalized Z-score maps of liver SUV were generated automatically for each patient, normalizing for individual variability.
• Normalized Z-score maps improved detection of hepatic tumors compared to conventional SUV maps, as demonstrated by ROC curve analyses.
• Blood test parameters such as AST, ALT, ALP, bilirubin, albumin, and clinical factors like age, BMI, and blood glucose significantly influenced liver FDG uptake.
• The method allows standardized and more accurate interpretation of liver PET images in clinical practice.

Clinical Implications

Normalizing liver SUV using routinely obtained blood test results and demographic data can enhance the sensitivity and specificity of 18F-FDG-PET for detecting hepatic tumors. This approach facilitates standardized image interpretation and may reduce missed diagnoses of liver lesions with moderate FDG uptake. Incorporating personalized Z-score maps into clinical workflows could improve diagnostic confidence and patient management.

Conclusion

The study demonstrates that liver SUV normalization using blood test-derived variables and machine learning improves tumor detection in 18F-FDG-PET imaging. This personalized approach offers a promising tool for enhancing diagnostic accuracy and standardization in hepatic lesion evaluation.

References

1. Medical Imaging and PET/CT Context [1-4]
2. Liver FDG Uptake and Clinical Variables [5-12]
3. Study Dataset and Methodology [13]