CNN models using FDG PET/CT differentiate benign from malignant ovarian tumors with high accuracy

Deep learning using F-18 FDG PET/CT distinguished benign from malignant ovarian tumors: ResNet-18 reached accuracy 0.882 and AUC 0.938; a simpler CNN reached AUC 0.957 and specificity 0.942 in 101 patients. Findings support CNN-based PET classification.

• Study included 101 patients with pelvic masses who underwent F-18 FDG PET/CT before surgery, with histopathology as the reference standard.
• ResNet-18 yielded accuracy 0.882, AUC 0.938, sensitivity 0.829, specificity 0.913; the simpler CNN yielded accuracy 0.876, AUC 0.957, sensitivity 0.761, specificity 0.942.
• Grad-CAM heatmaps visualized image regions the models found most relevant for classification.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.