ePoster Forums
Purpose: Auto-contouring organs at risk (OARs) and targets can save time in radiation oncology. However, there remains a question about its accuracy. Manteia AccContour is a image workstation that can be used in image auto-segmentation. This research is to evaluate its accuracy in auto-contouring of OARs.
Methods: 5 sets of DICOM images, including 2 head and neck patients, 2 breast patients, and 1 cervical patient were imported into Manteia AccuContour, and OARs were generated using auto-segmentation. These structures were then exported into Eclipse, and their volumes were compared to those manually drawn by physicians. In addition, the intersection and union structure of auto-contoured and physician-contoured OARs had been created using boolean operators, and these volumes were recorded as well. The difference (abs ( physician-contoured ⁻ auto-contoured ) /physician-contoured ), and dice (intersection² / intersection+union ) were calculated for each structure.
Results: In H&N cases, brachial plexus, lens, lobe temporal, optical chasm have the high difference and low dice. While the dice for the mandible bone, brainstem, eyes, pituitary have the dice higher than 80%. Inner ear, oral cavity, optical nerve, parotids, spinal cord have the dice higher than 70%, and the lowest one is 12.12% for optical chiasm . In breast cases, all the OARs except esophagus, spinal cord, and trachea owe the dice higher than 70%. While the dice for Contralateral breast, heart, and lungs are higher than 90%. The reason for the low dice of esophagus, spinal cord, trachea is different layers Manteia and physician drawn. In cervix case, bladder shows high dice to 95%, while femur head and rectum only 52% and 66%.
Conclusion: Manteia AccuContour has been used clinically at our center and shown to be helpful in clinical contouring but better accuracy is desired for certain organs.