Program Information
Smart Segmentation Algorithm Performance in Eclipse Planning System
H Erra Sriramulu1*, a Mousa2 , (1) ,,,(2) Kuwait Cancer Control Center, Kuwait, shuwaikh, PB No 42262, Kuwait
Presentations
SU-I-GPD-J-110 (Sunday, July 30, 2017) 3:00 PM - 6:00 PM Room: Exhibit Hall
Purpose: To evaluate the merit, stability and efficiency of smart segmentation in eclipse planning system since segmentation is governed by series of modeling. User selected models can be introduced to improve the delineation and extracting information from other modality.
Methods: Patient is scanned and Eclipse planning(ver 8.6) system has smart segmentation which provides automatic segmentation for the organs at risk. If the gradient persists then segmentation can be done either with derivative or laplacian. If the interface does not have any kind of cleavage then models must be used to deal with. Five patients contours are analyzed in order to check the stability and efficiency of contouring. The vector representation of the merit also useful for visual prediction and gives information where algorithm has to be developed in order to reduce the errors in contouring I have calculated covered percentage and overestimated percentage by algorithm. I have defined hari-gopal vector, magnitude and index, as vector represented in covered in x axis and over estimated in y axis and its magnitude is vector magnitude. Higher the magnitude below 100 but near, gives the better magnitude. Similarly index is defined as calculated minus overestimated. Its negative value gives more overestimated than the covered.
Results: All tubular structures surrounded by soft tissue having problem and index informs that negative value for optical chiasm, nerves and rectum are overestimated. Lungs, heart, liver and bladder show good coverage. Visual vector representation is very useful.
Conclusion: it is possible to merit the models. In order to improve optical structures coverage by algorithm, since it has to detect from very feeble background,intersection of other modality like MRI can introduce lining(interface information) depends on CT voxel neighborhood on CT Hounsfield matrix itself such that planning should not be disturbed and then smart segmentation can operate on that.
Contact Email: