In the research group of Assistant Professor Daisuke Kawahara, Specially Appointed Associate Professor Shuichi Ozawa, Professor Yasushi Nagata and others of Hiroshima University and Professor Sadaharu Nishio, a member of the Medical Physics Working Group of Japan Clinical Oncology Group (JCOG), a new automatic method using AI technology Developed a contouring system, Step-wise net. Step-wise net uses deep learning technology to perform two-step learning to extract contours of multiple organs in the head and neck. In the first stage, the area around the organ to be contoured is extracted, and in the second stage, the contour of the organ is created with high accuracy in the extracted area.

　When the contour creation accuracy of Step-wise net was evaluated, the accuracy of Step-wise net was higher in all organs than the existing automatic contour extraction system (Atlas method, which is a commercially available tool) that does not use AI. rice field.Furthermore, in comparison with the conventional method (U-net) using AI, the accuracy of Step-wise net was higher in all organs.

　Step-wise net will not only significantly improve contouring accuracy, but will also reduce contouring time to less than one-tenth of the conventional time, which is expected to contribute to the improvement and efficiency of medical field operations. Is done.

　In addition, since the automatic contouring of Step-wise net does not cause a difference between facilities like manual contouring, even in clinical trials where it is necessary to perform contouring of the same quality at each facility based on unified rules, this book. It is expected that the tool will be used.

Paper information:[Computers in Biology and Medicine] Stepwise deep neural network (stepwise-net) for head and neck auto-segmentation on CT images