Deep learning detects lower-limb DVT on CT venography with high accuracy

A CTPA-pretrained YOLO11-nano model detected lower-limb DVT on CT venography with 92.1% accuracy and 93.9% sensitivity, exceeding a Faster R-CNN model (66% accuracy); both models surpassed 97% accuracy for pulmonary embolism.

• The study enrolled 119 DVT-positive and 40 DVT-negative patients, plus 111 PE-positive and 20 PE-negative patients; three radiologists with 3–5 years’ experience interpreted CTV images.
• The YOLO11-nano model achieved 92.1% accuracy and 93.9% sensitivity for DVT, versus 66% accuracy and 68% sensitivity for Faster R-CNN; for PE, both algorithms exceeded 97% accuracy and sensitivity.
• By segment, YOLO11-nano exceeded 77% accuracy in pelvic and femoral-popliteal segments and reached 66.7% for calf thrombi.

Automated summary

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