Objective Realize the fusion of clinical close-range treatment radioactive source position accuracy detection and artificial intelligence.

Method  The method is based on affine transformation function to correct video angle, LSD algorithm to detect line segments, Canny edge detection operator and find Contours function to identify moving radioactive source targets, and other machine vision algorithms to automatically identify the location of radioactive sources, so as to achieve the purpose of detecting the location accuracy of radioactive sources in close-range treatment machines, and compare it with traditional manual detection methods.

Results The repeated positioning errors measured by naked eye observation and machine vision analysis were (0.30±0.48) mm and (0.80±0.42) mm, respectively, and the P value was 0.015, which was statistically significant. The repeated positioning time errors were (0.32±0.12) s and (0.03±0) s, respectively, and the P value was 0, which was statistically significant. The cumulative positioning errors were (1.45±0.42) mm and (1.72±0.47) mm, respectively. The P value was 0.025, which was statistically significant. The cumulative positioning time errors were 1.00 s and 2.10 s, respectively. The machine vision analysis method was more accurate. In terms of the time of analysis results, the three parameters of radioactive source positioning, repetition and cumulative error, the time of visual observation and visual analysis are 1min, 10min,machine vision analysis takes less time.

Conclusion The automatic detection technology of measuring the accuracy of radioactive source position by machine vision analysis method can meet the requirements of quality control guidelines for short-range treatment machines, and can meet the requirements of basic and complex quality control projects. It has the advantages of time saving, high stability and automation. It is an application technology combining artificial intelligence with medical clinic.