Input: Training set , Testing set TE, the outputted ROIs of Algorithm (or 9) |
Output: The locations of detected lobulation regions |
Stage 1: Training |
for each image in |
Step : Preprocess the image; |
Step : Obtain the image patches by sliding window method; |
Step : Obtain the edges of image patch using canny algorithm; |
Step : Compute bending energy for each edge fragment in ; |
for each edge fragment in |
step : Compute the curvature for all points of ; |
step : Compute the bending energy of edge point using curvature ; |
step : Compute the average bending energy of edge points locating in a curved line sliding window with the given size ; |
end |
Step : Sort the average bending energy of edge points by descend; |
Step : Select Top 6 largest average bending energy ; |
Step : Compute the average value BE of as the bending energy feature (BEFeature, image feature) in current curved line |
sliding window; |
Step : Determine Threshold Th1 and Th2 empirically based on the training samples, so that BEFeature Th1 and |
BEFeature Th2 for lobulation. |
end |
Stage 2: Lobulation detection |
for each image in testing set TE |
Step : Obtain the result ROIs (i.e., bounding box region) for current image from ; |
for each ROI in R |
Step : Transform the gray value in to CT value by formula (1); |
Step : Compute the bending energy BE for using the similar steps in train stage; |
Step : lobulation labeling |
If BE > Th1 and BE < Th2 |
Label as lobulation; |
end |
end |
end |