The application value of YOLO neural network for imaging-based diagnosis and differential diagnosis of acute aortic syndrome
- VernacularTitle:YOLO神经网络在急性主动脉综合征影像学诊断及鉴别诊断中的应用价值
- Author:
Mengyang KANG
1
;
Yang ZHAO
;
Feng CHI
;
You LI
;
Hongyan TIAN
Author Information
- Publication Type:Journal Article
- Keywords: artificial intelligence(AI); YOLO neural network; acute aortic syndrome(AAS); image recognition; diagnosis model
- From: Journal of Xi'an Jiaotong University(Medical Sciences) 2025;46(2):317-322
- CountryChina
- Language:Chinese
- Abstract: Objective To develop an artificial intelligence(AI)diagnostic system for computed tomography angiography(CTA)images related to acute aortic syndrome(AAS)and to evaluate its efficacy in diagnosing and differentiating between AAS subtypes.Methods We collected the CTA images of patients diagnosed with aortic dissection(AD),intramural hematoma(IMH),or penetrating atherosclerotic ulcer(PAU)who were treated in the Department of Peripheral Vascular Diseases,The First Affiliated Hospital of Xi'an Jiaotong University,from June 2016 to June 2022.Based on the strict inclusion and exclusion criteria,2 057 CTA images were selected and extracted.Normal human aorta CTA images served as the control group.The YOLO v7 neural network was used to diagnose and differentiate subgroups of AAS patients in the CTA images,and its diagnostic performance was evaluated.Results The sensitivity,specificity,positive predictive value,negative predictive value,and total accuracy of the AAS diagnostic system based on the YOLO v7 were 98.72％,83.10％,97.82％,89.40％,and 96.92％,respectively.The total accuracy for differential diagnosis of the subgroups was 85.58％.The total accuracy of diagnosis results was significantly higher than that of differential diagnosis results(P＜0.05).Conclusion The AI-based AAS diagnostic system utilizing YOLO v7 meets the established standards for disease diagnosis.However,further research is warranted involving larger image databases and more advanced deep learning networks to improve the differentiation among AAS subtypes.
