机器学习对不同性质斑块致冠状动脉狭窄程度初步评估的价值

Value of machine learning in preliminary assessment of the degree of coronary artery stenosis caused by different types of plaques

目的探讨机器学习对不同性质斑块致冠状动脉狭窄程度初步评估的价值。方法选择2020年1月—2022年10月于我院行冠状动脉CT血管造影(CCTA)并在随后14 d内行冠状动脉造影(CAG)的患者80例,随机选取80例患者CCTA中103处冠状动脉狭窄位置,根据斑块性质分为钙化斑块组(38处)、非钙化斑块组(34处)和混合斑块组(31处)。分别采用主观评估(SA)法、后处理工作站测量(AW)法、人工智能(AI)法、人工与AI结合(Semi-AI)法评估各组斑块所致冠状动脉狭窄程度。以CAG结果为诊断冠状动脉狭窄程度的金标准,比较上述四种方法与金标准的符合率、低估率及高估率。结果四种方法中,AI法评估三种斑块的符合率、低估率、高估率与SA法比较均无显著差异(P>0.0083)。在非钙化斑块和混合斑块的评估中,AI法的符合率显著高于AW法和Semi-AI法(χ^(2)=7.65~16.20,P<0.0083);在钙化斑块的评估中,AI法的符合率与其他三种方法相比无统计学差异(P>0.05);在钙化斑块及混合斑块的评估中,Semi-AI法比其他三种方法的高估率显著降低(χ^(2)=8.77~23.62,P<0.0083)。结论AI法在一定程度上可取代影像科医师对不同性质斑块所致冠状动脉狭窄程度的主观评估,对于冠脉狭窄的评估流程具有优化作用,Semi-AI法则可改善对各种性质斑块所致冠脉狭窄的高估的情况。但目前上述方法仅能对冠状动脉狭窄程度进行初步评估,均无法作为金标准取代CAG法。

Objective To explore the application value of machine learning in preliminary evaluation of the degree of co-ronary artery stenosis caused by different types of plaques.Methods Eighty patients who underwent coronary CT angiography(CCTA)and coronary angiography(CAG)in the following 14 d from January 2020 to October 2022 were selected.During CCTA,103 coronary artery stenosis sites were randomly selected and divided into calcified plaque group(38 sites),non-calcified plaque group(34 sites),and mixed plaque group(31 sites)according to plaque properties.Subjective evaluation(SA),post-processing workstation measurement(AW),artificial intelligence(AI),and SA combined with AI(Semi-AI)were used to assess the degree of coronary artery stenosis caused by plaques in each group.CAG results were used as the gold standard for the degree of coronary artery stenosis.The coincidence,underestimation,and overestimation rates were calculated based on the gold standard and compared between the four methods.Results Among the four methods,there were no significant differences in the coincidence rate,underestimation rate,and overestimation rate between AI and SA(P>0.0083).In the evaluation of non-calcified plaque and mixed plaque,the coincidence rate of AI was significantly higher than those of AW and Semi-AI(χ^(2)=7.65-16.20,P<0.0083).In the evaluation of calcified plaque,the coincidence rate of AI was not significantly different from those of the other three methods(P>0.05).In the evaluation of calcified plaque and mixed plaque,the overestimation rate of Semi-AI was significantly lower than those of the other three methods(χ^(2)=8.77-23.62,P<0.0083).Conclusion AI can partly replace the subjective evaluation made by radiologists regarding coronary artery stenosis caused by different types of plaques,thus optimizing the evaluation process of coronary artery stenosis.The Semi-AI method can reduce the overestimation of coronary artery stenosis caused by various types of plaques.However,AI cannot be used as a gold standard,and can only be used to preliminarily evaluate the degree of coronary artery stenosis.