Superficial femoral artery calcification segmentation and detection in CT angiography using convolutional neural...

Calcification detection and segmentation in CT angiography (CTA) is the basis of preoperative calcification assessment and treatment determination in endovascular interventional surgery for lower-extremity atherosclerotic occlusion disease. However, the complex calcification-lumen contrast and difficult-to-locate occluded superficial femoral artery (SFA) make it challenging. This paper proposes a fast and accurate method without artery extraction to segment and detect SFA calcification in CTA using a convolutional neural network.

The thigh region containing the target SFA is first automatically extracted based on the human anatomical position. Then, 3D Unet with a large receptive field is used to segment calcifications in image patches with a large field of view. The lumen label is introduced and a calcification-lumen contrast data augmentation method is developed to improve the segmentation performance on images with varying calcification-lumen contrast. Finally, false-positive errors far from the SFA are eliminated based on the SFA centerline estimated from the segmentation results.

Five-fold cross validation experiments were conducted on a local dataset of CTA images containing 128 SFAs. The average Dice scores of calcification segmentation on the entire, occluded and non-occluded arteries achieved 89.12%, 92.98% and 88.96%, respectively. The average recall and precision of calcification detection on each slice were 93.50% and 91.51%, respectively. The total processing time was about 2 min.

This paper proposes a CNN-based method to segment and detect SFA calcification in CTA without artery extraction for varying calcification-lumen intensity contrast and arterial occlusion situations. The work can be used to improve clinical calcification analysis.