(Tuesday, 7/31/2018) 10:00 AM - 10:30 AM
Room: Exhibit Hall | Forum 9
Purpose: To achieve the correct radiation treatment planning, we developed an automatic contouring program based on artificial intelligence using both computed tomography (CT) and nuclear medicine (NM) image.
Methods: The system was designed to work in the following format using machine learning method. To contour tumor region on CT image, CT image was divided into three part and NM image was overlapped on CT. After that, we make radiotracer concentrated area (i.e., tumor suspicion area) to be contoured through the NM image. System also received an additional user input to select and contour the entire region of a tumor. After all images were contoured, system calculated tumor data through deep learning technique. For the system verification, CT and positron emission tomography (PET) image set of human head were used.
Results: Actually developed program was operated as follows. CT images were normalized and divided into air, tissue and bone by decision tree. Tumor suspicion areas were determined by applying appropriate thresholds to a normalized NM image. Actual tumor region was contoured by mouse click based region labelling. The selected tumor region was expanded to adjacent image layers using user input and supervised learning algorithm. Through the image fusion process, air-tissue-bone-tumor separated head images are generated. To determine the treatment field, four values were calculated using deep learning technique: optimized tumor to skin length, tumor depth, tumor width and total tumor volume. In the verification process, program demonstrated the same contouring result from the identical condition except user.
Conclusion: The effectiveness of automatic contouring program using based on artificial intelligence both CT and NM image has been developed. This program can help to reduce planning time and planner’s work. However, the accuracy evaluation should be required to prove the superior of this program in the future.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
