(Sunday, 7/29/2018) 3:00 PM - 6:00 PM
Room: Exhibit Hall
Purpose: Contouring is an important first step in treatment planning, but contouring of targets and organs at risk is a time-consuming process. The use of automatically generated contours has the potential of dramatically improving treatment planning efficiency. However, the quality of such contours needs to be evaluated before their use in clinics. The aim of this study is to evaluate the consistency of auto-generated contours and its quality by comparing them with manually drawn contours.
Methods: Three lung patients’ CT images were selected and two set of contours for three critical organs (lung, heart and spinal cord) were automatically generated at different times. Conformation index (CI), which is defined as the ratio between the intersection and union of the two contours (V1∩V2/V1UV2) generated by the same software but on different days, were evaluated. CI between one set of automatically generated contours and manually-drawn contours are also computed.
Results: The two sets of automatically generated contours are similar but not exactly the same. CI for hearts are 0.95, 0.95 and 0.96. CI for left lungs are 0.95, 0.95 and 0.94, and for right lungs are 0.95, 0.95 and 0.94. For spinal cord, the CIs are 0.77, 0.77 and 0.78. For the case of auto contours and manually-drawn contours, CIs for hearts are 0.76, 0.48 and 0.8, respectively. CI for left lungs are 0.92, 0.91 and 0.89 whereas CI for right lungs are 0.92, 0.92 and 0.88. For spinal cord, CIs are 0.63, 0.62 and 0.67.
Conclusion: Auto contours are not necessarily consistent even when they are generated by the same software, and they differ from manual contours as well. Its impact on radiation therapy needs to be evaluated before their use in clinics.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
