Abstract: Objective The aim of this research is to diagnose and analyze the metastasis of mediastinal lymph nodes of non-small cell lung cancer(NSCLC)through the method of visual observation,measurement and the computer pattern recognition;then compared with the pathological diagnosis results,explored the diagnostic efficiency of every method,Which can provide more accuracy information about N staging of NSCLC for the clinicians.Methods 28 cases of preoperative whole body PET/CT imaging with pathological-proven NSCLC were included in the study.Three practised doctors read the lymph nodes images of CT,PET and PET/CT through the method of visual observation and measurement,then compared with the pathological diagnosis.To analyze the lymph node images which were extracted through the new computer pattern recognition of engineering.Results Among CT,PET and PET/CT,the sensibility of mediastinal lymph node involvement was 53.6%,80.4% and 82.1% respectively,the specificity was 92.2%,83.3% and 90.6% respectively,the positive predictive value was 68.2%,60% and 73.0% respectively,the negative predictive value was 86.5%,93.2% and 94.2% respectively and the diagnosis coincidence rate was 83.1%,82.6% and 88.6% respectively;For the method of visual observation,the sensibility of mediastinal lymph node involvement was 53.6%,71.4% and 69.6% respectively,the specificity was 85.0%,82.2% and 88.9% respectively,the positive predictive value was 52.6%,55.6% and 66.7% respectively,the negative predictive value was 85.5%,90.2% and 90.4% respectively and the diagnosis coincidence rate was 77.5%,79.7% and 84.3% respectively.For the computer pattern recognition,the sensibility of PET and CT imaging were 88% and 84% separately,and the specificity was 76% and 66% separately.Conclusion The method of visual observation and measurement of PET/CT imaging for diagnosing the metastasis of mediastinal lymph nodes in patients with non-small cell lung cancer was advantage over the CT and PET imaging.And the method of measurement was advantage over the method of visual observation.This research preliminary explore the engineering method of SVM and multi-resolution histogram to analyze the image of lymph node,and settle a basis for the further combination of multiple subjects.