Abstract：Objective To obtain an adaptive threshold that satisfied the accurate segmentation of the tumor biological target area,
and improved the accuracy of image segmentation by extracting the gradient information of the PET image. Methods This paper
analyzed the data of the PET image which took the point with the largest standardized uptake value (SUV) as the center point of
the tumor area, and also took the SUV in eight directions. Combined with the second partial derivative in the image processing
algorithm, it could reflect the characteristics of the image boundary information. Two gradient operations were performed on it
to obtain the optimal boundary thresholds in eight directions, and average the eight thresholds as the adaptive threshold for
image segmentation. The similarity and sensitivity of image segmentation using the adaptive threshold proposed in this
paper and the traditional SUV=2.5 and 40% SUVmax, respectively, were compared. Results This paper proposed an adaptive
threshold segmentation algorithm based on the accuracy evaluation measurement method of the region. The average value of the
Dice similarity coefficient (DSC) was 85.26%, and the average value of true positive rate (TPR) index was 82.31%. Compared
with the threshold value of SUV=2.5, the increase was 15.14% and 17.36% respectively, and the increase was 11.97% and 12.47%
respectively compared with the threshold value of 40% SUVmax. There were statistical differences in the similarity and sensitivity
of images obtained by three threshold segmentation algorithms (F=81.02, F=217.21, P<0.01). Conclusion The adaptive threshold
segmentation algorithm proposed in this study can improve the accuracy of tumor segmentation results in PET/CT images.
戴盈欣，吴锐先，王治国. 基于PET/CT图像的自适应阈值分割算法[J]. 中国医疗设备, 2022, 37(3): 79-83.
DAI Yingxin, WU Ruixian, WANG Zhiguo. Research on Adaptive Threshold Segmentation Algorithm Based on PET/CT. China Medical Devices, 2022, 37(3): 79-83.
