Radiomics with Clinical Data and [¹⁸F]FDG-PET for Differentiating Between Infected and Non-Infected Intracavitary...

Area of Science:

• Nuclear Medicine
• Radiology
• Artificial Intelligence

Background:

• Diagnosing infected intracavitary vascular grafts and endografts (iVGEI) is challenging.
• Current diagnostic methods may have limitations in accuracy and timeliness.

Purpose of the Study:

• To evaluate the feasibility of machine learning (ML) models for differentiating iVGEI.
• To compare models using clinical features, radiomics, and a combination of both.

Main Methods:

• Developed three ML models: "MAGIC-light" (clinical features), "PET-radiomics" (radiomics from [18F]FDG PET/CT), and a combined model.
• Utilized automated segmentation (SEQUOIA, TotalSegmentator) for radiomics extraction from 96 features.
• Validated models on training (92 patients) and external test sets (32 patients).

Main Results:

• The combined ML model achieved the highest AUC (0.91 ± 0.02) in the test set.
• Combined model demonstrated superior accuracy (0.91) compared to MAGIC criteria alone (0.82).
• The combined model showed high specificity (1.00) and comparable sensitivity (0.70).

Conclusions:

• ML models integrating clinical and PET-radiomics features show significant potential for iVGEI diagnosis.
• A combined approach offers high diagnostic performance and specificity.
• This method could aid clinical decision-making, reduce overtreatment, and improve patient outcomes.