2D SECara-Net and 3D U2-Net for detecting unruptured saccular intracranial aneurysms with MR angiography
10.13929/j.issn.1003-3289.2025.02.013
- VernacularTitle:2D SECara-Net及3D U2-Net模型用于检测MR血管造影中未破裂囊状颅内动脉瘤
- Author:
Zongren NIU
1
;
Qiang MA
;
Jingjing DU
;
Yande REN
;
Mengjie LI
;
Yaqian QIAO
;
Yueshan TANG
;
Jianbo GAO
Author Information
1. 青岛大学附属医院放射科,山东青岛 266555;淄博市妇幼保健院影像科,山东淄博 255020
- Publication Type:Journal Article
- Keywords:
intracranial aneurysm;
magnetic resonance angiography;
artificial intelligence
- From:
Chinese Journal of Medical Imaging Technology
2025;41(2):245-249
- CountryChina
- Language:Chinese
-
Abstract:
Objective To observe the value of 2D SECara-Net and 3D U2-Net models constructed based on 2D maximal intensity projection(MIP)and 3D time-of-flight MR angiography(3D TOF-MRA)images,respectively,also of their combination for MRA detecting unruptured saccular intracranial aneurysms(USIA).Methods Totally 973 patients with single USIA and 300 subjects who underwent healthy physical examination were retrospectively collected and divided into training set(n=923,containing 723 cases of USIA and 200 healthy subjects)and test set(n=350,containing 250 cases of USIA and 100 healthy subjects)at the ratio of 7:3.Pre-processed 3D TOF-MRA and the obtained 2D-MIP images in training set were imported into 3D U2-Net and 2D SECara-Net models for training and adjusting parameters,respectively.The efficiency of 2 models and their combination for detecting USIA were evaluated.Results The sensitivity,specificity and accuracy of 2D SECara-Net model for detecting USIA in test set was 78.80％(197/250),95.00％(95/100)and 83.43％(292/350),of 3D U2-Net model was 82.80％(207/250),86.00％(86/100)and 83.71％(293/350),respectively.The specificity of 2D SECara-Net model was higher than that of 3D U2-Net model(P=0.030),while no significant difference of sensitivity nor accuracy was found between 2 models(both P＞0.05).The specificity of the combination of the 2 models was 99.00％(99/100),higher than that of 3D U2-Net model(P＜0.05),and the sensitivity and accuracy of the combination was 91.20％(228/250)and 93.43％(327/350),respectivelty,both higher than those of 2 single models(all P＜0.05).Conclusion 2D SECara-Net and 3D U2-Net models had similar,sensitivity and accuracy for MRA detecting USIA.Combination of them could improve the detecting efficacy.
