AI Aided Annotation Tool

Medical imaging semantic segmentation using extreme points

Medical imaging segmentation needs a pixel map to annotate the target organ or malignant disease regions (i.e., tumor and polyps, etc.). Annotating target objects, like pixel by pixel by hand, is prohibitively expensive and labor-intensive.

We developed an efficient AI-aided medical imaging annotation tool using the deep extreme cut (the paper reference). Users specify four extreme points on the target region; the AI annotation tool automatically generates the binary mask. The delineated binary mask region is close to the edge boundary for the target organ. The video shows the training and testing across different medical image domain datasets on CT images.

VIDEOS

Training and Validating on the Visceral Dataset

Video file

Cross Domain: Decathlon Dataset – Testing Case: http://medicaldecathlon.com/

Video file

REFREFENCE

K.K. Maninis and S. Caelles and J. Pont-Tuset and L. Van Gool, Deep Extreme Cut: From Extreme Points to Object Segmentation, Computer Vision and Pattern Recognition (CVPR), 2018