Photon-counting Detector CT Spectral Reconstructions for Radiomics-based Liver Lesion Classification: A Multicenter Study.
Invest RadiolSep 17, 2026 (epub)
Friedrich L Pietsch, Abhinay K Vellala, Florian Haag et al.
✦ AI-curated · Sources linked
In 30 seconds
This multicenter clinical trial evaluated the effectiveness of radiomics-based machine learning for classifying benign and malignant liver lesions using photon-counting detector CT (PCD-CT). The study included 378 patients and found that the Random Forest model trained on virtual non-contrast images achieved an AUC of 0.899, indicating strong performance in lesion differentiation.
Key findings
- Random Forest model on virtual non-contrast images achieved an AUC of 0.899 and accuracy of 83.5%.
- Models tested on 110, 140, and 180 keV showed the highest stability with median AUCs ≥ 0.83.
- Lower energy levels (40 keV, 50 keV) and iodine density maps resulted in lower and more variable AUCs (median AUC ≤ 0.71).
Why it matters
Accurate noninvasive classification of liver lesions is crucial for effective patient management. This study suggests that advanced imaging techniques and machine learning can enhance diagnostic accuracy, potentially reducing the need for invasive procedures.