Radiomics Features Differentiate Between Normal and Tumoral High-Fdg Uptake

Area of Science:

• Nuclear Medicine
• Radiomics
• Oncology

Background:

• Distinguishing fluorodeoxyglucose-avid (FDG-avid) neoplasms from high-uptake normal tissues in PET imaging is challenging.
• This limitation hinders accurate assessment of disease natural history and treatment response.

Purpose of the Study:

• To develop and validate a radiomics-based classifier for differentiating FDG-avid normal tissues from tumors in PET scans.
• To improve the accuracy of tumor segmentation and classification in oncology imaging.

Main Methods:

• A random forest classifier was built using radiomics features (shape, first-order) derived from Standardized Uptake Value (SUV) in FDG-PET scans.
• The classifier was trained and tested on data from Hodgkin lymphoma and Ewing sarcoma patients.
• Automatic segmentation of tumor and normal tissue volumes was performed and compared with manual segmentations.

Main Results:

• The classifier achieved 90% accuracy in identifying normal tissues, with high performance across different organs (e.g., brain 100%, heart 97%).
• Automatically segmented tumor volumes demonstrated strong concordance with manually segmented volumes (R² = 0.97).
• Texture-based radiomics features offered minimal improvement to the classifier's performance.

Conclusions:

• Accurate segmentation and classification of normal tissues using radiomics are essential for precise identification of FDG-avid tumor tissues.
• This approach enhances the reliability of PET imaging in cancer diagnosis and management.
• The developed classifier shows promise for improving oncologic imaging analysis.