Macchi Beatrice, Felisi Marco Maria Jacopo, Muti Gaia, Cicolari Davide, Parisotto Marco, Gennari Luciana, Sartori Ivana, Arosio Paolo, Piano Mariangela, Colombo Paola Enrica, Squarza Silvia
Dipartimento di Fisica "A. Pontremoli", Università degli Studi di Milano, Italy; Medical Physics Department, ASST Grande Ospedale Metropolitano Niguarda, Milano, Italy.
Medical Physics Department, ASST Grande Ospedale Metropolitano Niguarda, Milano, Italy.
Phys Med. 2025 Aug;136:105052. doi: 10.1016/j.ejmp.2025.105052. Epub 2025 Jul 19.
Task-based functional MRI (tb-fMRI) effectiveness as a support tool in brain mapping may be limited by patients' poor cooperation. Resting-state fMRI (rs-fMRI) represents an alternative or complementary approach. In this work, we developed and validated an analysis pipeline for rs-fMRI acquisitions, primarily aimed at language mapping in drug-resistant epileptic patients. The workflow relies on open-source software and semi-automatized solutions, ensuring easy clinical adoption.
Rs-fMRI data were acquired from 26 subjects (15 volunteers, 11 patients) using a 3 T-MRI scanner. The developed pipeline starts with preprocessing of raw data, subsequently analyzed through Independent Component Analysis (ICA), performed with MELODIC-FSL tool. Manual classification, semi-automated classifiers (FIX, ICA-AROMA) and a template matching procedure were employed to classify the ICA components and extract each patient rs-language network. Finally, verb-generation tb-fMRI and Diffusion Tensor Imaging were acquired to map language regions and reconstruct the arcuate fasciculus, respectively. The rs-language networks were validated evaluating the three acquisition modalities agreement.
Trained FIX showed AUC = 0.95 and ICA-AROMA 97 % of classification accuracy, considering manual classification as ground truth. Manual classification identified one (46 %), two (31 %), or three (19 %) language-related components per subject. The manually selected language components were among the top three ranked by the template matching in 88 % of cases, 100 % considering the top five. The Dice index between rs-fMRI and tb-fMRI language maps resulted 0.36 ± 0.13. Rs-language areas resulted qualitatively well-connected by the reconstructed arcuate fasciculus.
The developed pipeline confirmed strong potential for clinical applicability in a large general hospital, especially when tb-fMRI is infeasible.
基于任务的功能磁共振成像（tb-fMRI）作为脑图谱辅助工具的有效性可能因患者配合度差而受限。静息态功能磁共振成像（rs-fMRI）是一种替代或补充方法。在本研究中，我们开发并验证了一种用于rs-fMRI采集的分析流程，主要针对耐药性癫痫患者的语言图谱绘制。该工作流程依赖开源软件和半自动解决方案，确保易于临床应用。
使用3T磁共振成像扫描仪从26名受试者（15名志愿者，11名患者）获取rs-fMRI数据。所开发的流程从原始数据预处理开始，随后通过使用MELODIC-FSL工具进行的独立成分分析（ICA）进行分析。采用手动分类、半自动分类器（FIX、ICA-AROMA）和模板匹配程序对ICA成分进行分类，并提取每位患者的rs语言网络。最后，进行动词生成tb-fMRI和扩散张量成像，分别绘制语言区域和重建弓状束。通过评估三种采集方式的一致性来验证rs语言网络。
以手动分类为金标准，训练后的FIX显示曲线下面积（AUC）=0.95，ICA-AROMA的分类准确率为97%。手动分类发现每位受试者有一个（46%）、两个（~31%）或三个（19%）与语言相关的成分。在88%的病例中，手动选择的语言成分在模板匹配排名中位列前三，若考虑前五则为100%。rs-fMRI和tb-fMRI语言图谱之间的Dice指数为0.36±0.13。重建的弓状束使rs语言区域在定性上连接良好。
所开发的流程证实了在大型综合医院具有很强的临床应用潜力，特别是在tb-fMRI不可行时。
