Fully automatic fossa ovalis segmentation from computed tomography images using deep neural network with atlas-based localization

Q1 Medicine

Informatics in Medicine Unlocked Pub Date : 2025-01-01 DOI:10.1016/j.imu.2025.101613

Gakuto Aoyama , Toru Tanaka , Yukiteru Masuda , Naoki Matsuki , Ryo Ishikawa , Masahiko Asami , Kiyohide Satoh , Takuya Sakaguchi

{"title":"Fully automatic fossa ovalis segmentation from computed tomography images using deep neural network with atlas-based localization","authors":"Gakuto Aoyama , Toru Tanaka , Yukiteru Masuda , Naoki Matsuki , Ryo Ishikawa , Masahiko Asami , Kiyohide Satoh , Takuya Sakaguchi","doi":"10.1016/j.imu.2025.101613","DOIUrl":null,"url":null,"abstract":"<div><h3>Background and objective</h3><div>Information on the location of the fossa ovalis (FO) is necessary for planning interventional procedures that require an inter-atrial septal (IAS) puncture. At present, this information is obtained manually from pre-procedural medical images, which is time consuming with limited reproducibility. In this paper, we propose a method to automatically segment the FO region from computed tomography (CT) images.</div></div><div><h3>Methods</h3><div>Our proposed method roughly crops CT images based on atlas information of the FO and heart chambers, and inputs the cropped CT images to a U-Net-based deep neural network (DNN) to segment the FO region. This method was evaluated by five-fold cross validation using 215 CT images with manually annotated FO regions, and its segmentation accuracy was compared to two previously reported methods based on thinness of the IAS wall and on simple DNN.</div></div><div><h3>Results</h3><div>The segmentation process was successful in all cases for the IAS-based method, but failed in 4 cases for the proposed method and in 30 cases for the DNN method due to irregular heart structure. For the segmentation accuracies of our proposed method, the IAS wall thinness-based method and DNN based-method, mean chamfer distances were 2.16 ± 1.43, 3.30 ± 1.37 and 2.66 ± 1.32 respectively, with the difference being statistically significant.</div></div><div><h3>Conclusions</h3><div>These results suggest that our proposed method can automatically segment the FO region more accurately with fewer failures.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"52 ","pages":"Article 101613"},"PeriodicalIF":0.0000,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Informatics in Medicine Unlocked","FirstCategoryId":"1085","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2352914825000012","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"Medicine","Score":null,"Total":0}

引用次数: 0

Abstract

Background and objective

Information on the location of the fossa ovalis (FO) is necessary for planning interventional procedures that require an inter-atrial septal (IAS) puncture. At present, this information is obtained manually from pre-procedural medical images, which is time consuming with limited reproducibility. In this paper, we propose a method to automatically segment the FO region from computed tomography (CT) images.

Methods

Our proposed method roughly crops CT images based on atlas information of the FO and heart chambers, and inputs the cropped CT images to a U-Net-based deep neural network (DNN) to segment the FO region. This method was evaluated by five-fold cross validation using 215 CT images with manually annotated FO regions, and its segmentation accuracy was compared to two previously reported methods based on thinness of the IAS wall and on simple DNN.

Results

The segmentation process was successful in all cases for the IAS-based method, but failed in 4 cases for the proposed method and in 30 cases for the DNN method due to irregular heart structure. For the segmentation accuracies of our proposed method, the IAS wall thinness-based method and DNN based-method, mean chamfer distances were 2.16 ± 1.43, 3.30 ± 1.37 and 2.66 ± 1.32 respectively, with the difference being statistically significant.

Conclusions

These results suggest that our proposed method can automatically segment the FO region more accurately with fewer failures.

查看原文本刊更多论文

使用基于地图集定位的深度神经网络从计算机断层扫描图像中全自动分割卵窝

背景和目的关于卵圆窝（FO）位置的信息对于规划需要房间隔穿刺的介入手术是必要的。目前，这些信息是手工从手术前的医学图像中获得的，这是耗时且可重复性有限的。本文提出了一种从计算机断层扫描（CT）图像中自动分割FO区域的方法。方法提出了一种基于前肢和心室图谱信息对CT图像进行粗略裁剪的方法，并将裁剪后的CT图像输入到基于u - net的深度神经网络（DNN）中对前肢区域进行分割。该方法通过5次交叉验证对215张带有人工注释FO区域的CT图像进行了评估，并与之前报道的基于IAS壁厚度和简单DNN的两种分割方法进行了比较。结果基于ias的方法分割成功，但由于心脏结构不规则，提出的方法和DNN方法分别有4例和30例失败。基于IAS壁薄的方法和基于DNN的方法的平均倒角距离分别为2.16±1.43、3.30±1.37和2.66±1.32，两者的分割精度差异有统计学意义。结论本文提出的方法能够更准确地自动分割前叶区域，且故障较少。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Informatics in Medicine Unlocked Medicine-Health Informatics

CiteScore

9.50

自引率

0.00%

发文量

282

审稿时长

39 days

期刊介绍： Informatics in Medicine Unlocked (IMU) is an international gold open access journal covering a broad spectrum of topics within medical informatics, including (but not limited to) papers focusing on imaging, pathology, teledermatology, public health, ophthalmological, nursing and translational medicine informatics. The full papers that are published in the journal are accessible to all who visit the website.