Detection of focal cortical dysplasia: Development and multicentric evaluation of artificial intelligence models.

IF 6.6 1区医学 Q1 CLINICAL NEUROLOGY Epilepsia Pub Date : 2024-12-31 DOI:10.1111/epi.18240

Lennart N Kersting, Lennart Walger, Tobias Bauer, Vadym Gnatkovsky, Fabiane Schuch, Bastian David, Elisabeth Neuhaus, Fee Keil, Anna Tietze, Felix Rosenow, Angela M Kaindl, Elke Hattingen, Hans-Jürgen Huppertz, Alexander Radbruch, Rainer Surges, Theodor Rüber

{"title":"Detection of focal cortical dysplasia: Development and multicentric evaluation of artificial intelligence models.","authors":"Lennart N Kersting, Lennart Walger, Tobias Bauer, Vadym Gnatkovsky, Fabiane Schuch, Bastian David, Elisabeth Neuhaus, Fee Keil, Anna Tietze, Felix Rosenow, Angela M Kaindl, Elke Hattingen, Hans-Jürgen Huppertz, Alexander Radbruch, Rainer Surges, Theodor Rüber","doi":"10.1111/epi.18240","DOIUrl":null,"url":null,"abstract":"Objective: Focal cortical dysplasia (FCD) is a common cause of drug-resistant focal epilepsy but can be challenging to detect visually on magnetic resonance imaging. Three artificial intelligence models for automated FCD detection are publicly available (MAP18, deepFCD, MELD) but have only been compared on single-center data. Our first objective is to compare them on independent multicenter test data. Additionally, we train and compare three new models and make them publicly available.Methods: We retrospectively collected FCD cases from four epilepsy centers. We chose three novel models that take two-dimensional (2D) slices (2D-nnUNet), 2.5D slices (FastSurferCNN), and large 3D patches (3D-nnUNet) as inputs and trained them on a subset of Bonn data. As core evaluation metrics, we used voxel-level Dice similarity coefficient (DSC), cluster-level F1 score, subject-level detection rate, and specificity.Results: We collected 329 subjects, 244 diagnosed with FCD (27.7 ± 14.4 years old, 54% male) and 85 healthy controls (7.1 ± 2.4 years old, 51% female). We used 118 subjects for model training and kept the remaining subjects as an independent test set. 3D-nnUNet achieved the highest F1 score of .58, the highest DSC of .36 (95% confidence interval [CI] = .30-.41), a detection rate of 55%, and a specificity of 86%. deepFCD showed the highest detection rate (82%) but had the lowest specificity (0%) and cluster-level precision (.03, 95% CI = .03-.04, F1 score = .07). MELD showed the least performance variation across centers, with detection rates between 46% and 54%.Significance: This study shows the variance in performance for FCD detection models in a multicenter dataset. The two models with 3D input data showed the highest sensitivity. The 2D models performed worse than all other models, suggesting that FCD detection requires 3D data. The greatly improved precision of 3D-nnUNet may make it a sensible choice to aid FCD detection.","PeriodicalId":11768,"journal":{"name":"Epilepsia","volume":" ","pages":""},"PeriodicalIF":6.6000,"publicationDate":"2024-12-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Epilepsia","FirstCategoryId":"3","ListUrlMain":"https://doi.org/10.1111/epi.18240","RegionNum":1,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"CLINICAL NEUROLOGY","Score":null,"Total":0}

引用次数: 0

Abstract

Objective: Focal cortical dysplasia (FCD) is a common cause of drug-resistant focal epilepsy but can be challenging to detect visually on magnetic resonance imaging. Three artificial intelligence models for automated FCD detection are publicly available (MAP18, deepFCD, MELD) but have only been compared on single-center data. Our first objective is to compare them on independent multicenter test data. Additionally, we train and compare three new models and make them publicly available.

Methods: We retrospectively collected FCD cases from four epilepsy centers. We chose three novel models that take two-dimensional (2D) slices (2D-nnUNet), 2.5D slices (FastSurferCNN), and large 3D patches (3D-nnUNet) as inputs and trained them on a subset of Bonn data. As core evaluation metrics, we used voxel-level Dice similarity coefficient (DSC), cluster-level F₁ score, subject-level detection rate, and specificity.

Results: We collected 329 subjects, 244 diagnosed with FCD (27.7 ± 14.4 years old, 54% male) and 85 healthy controls (7.1 ± 2.4 years old, 51% female). We used 118 subjects for model training and kept the remaining subjects as an independent test set. 3D-nnUNet achieved the highest F₁ score of .58, the highest DSC of .36 (95% confidence interval [CI] = .30-.41), a detection rate of 55%, and a specificity of 86%. deepFCD showed the highest detection rate (82%) but had the lowest specificity (0%) and cluster-level precision (.03, 95% CI = .03-.04, F₁ score = .07). MELD showed the least performance variation across centers, with detection rates between 46% and 54%.

Significance: This study shows the variance in performance for FCD detection models in a multicenter dataset. The two models with 3D input data showed the highest sensitivity. The 2D models performed worse than all other models, suggesting that FCD detection requires 3D data. The greatly improved precision of 3D-nnUNet may make it a sensible choice to aid FCD detection.

查看原文

微信好友朋友圈 QQ好友复制链接

本刊更多论文

局灶性皮质发育不良的检测：人工智能模型的发展和多中心评估。

目的：局灶性皮质发育不良（FCD）是耐药局灶性癫痫的常见病因，但在磁共振成像上很难发现。三种用于FCD自动检测的人工智能模型（MAP18, deepFCD， MELD）是公开可用的，但只在单中心数据上进行了比较。我们的第一个目标是在独立的多中心测试数据上对它们进行比较。此外，我们训练和比较了三个新模型，并将它们公开提供。方法：回顾性收集4个癫痫中心的FCD病例。我们选择了三种新颖的模型，它们以二维（2D）切片（2D- nnunet）、2.5D切片（FastSurferCNN）和大型3D斑块（3D- nnunet）作为输入，并在波恩数据的子集上对它们进行训练。作为核心评价指标，我们使用了体素级Dice相似系数（DSC）、聚类级F1评分、受试者级检出率和特异性。结果：共收集329例受试者，其中确诊为FCD的244例（年龄27.7±14.4，男性54%），健康对照85例（年龄7.1±2.4，女性51%）。我们使用118个受试者进行模型训练，其余受试者作为独立的测试集。3D-nnUNet最高F1评分为0.58，最高DSC为0.36(95%可信区间[CI] = 0.30 - 0.41)，检出率为55%，特异性为86%。deepFCD的检出率最高（82%），但特异性最低（0%），簇级精度最低（2%）。03, 95% ci = .03-。04， F1得分= .07)。MELD在各中心的表现差异最小，检出率在46%到54%之间。意义：本研究显示了多中心数据集中FCD检测模型的性能差异。两种具有三维输入数据的模型显示出最高的灵敏度。2D模型的表现比其他所有模型都差，这表明FCD检测需要3D数据。3D-nnUNet的精度大大提高，使其成为辅助FCD检测的明智选择。

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分钟内获得全文去求助

来源期刊

Epilepsia 医学-临床神经学

CiteScore

10.90

自引率

10.70%

发文量

319

审稿时长

2-4 weeks

期刊介绍： Epilepsia is the leading, authoritative source for innovative clinical and basic science research for all aspects of epilepsy and seizures. In addition, Epilepsia publishes critical reviews, opinion pieces, and guidelines that foster understanding and aim to improve the diagnosis and treatment of people with seizures and epilepsy.