CAREER: Small Data in a Big World: Balancing Interpretability and Generalizability for Data Integration in Clinical Neuroscience

职业：大世界中的小数据：平衡临床神经科学数据集成的可解释性和概括性

• 批准号: 1845430
• 负责人: Archana Venkataraman
• 金额: $ 50万
• 依托单位: Johns Hopkins University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-02-15 至 2023-04-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1845430&HistoricalAwards=false
• 关键词: CAREER; Small; Data; Big; World

Neurological and neuropsychiatric disorders affect millions of people worldwide and carry a staggering societal cost. Despite ongoing efforts, clinicians have a bare-bones understanding of these disorders, and hence, a limited ability to treat them. From an analytics perspective, clinical neuroscience is a field of high-dimensional datasets, small sample sizes, massive patient variability, and most importantly, an arguable lack of ground truth information. These challenges have led to a trade-off between the interpretability of a given model and its generalizability to new data. At one extreme, classical statistics allows us to formulate and test interpretable hypotheses about the brain, but it cannot make patient-specific generalizations. At the other extreme, conventional machine learning algorithms are geared towards patient generalizability but rarely illuminate a brain-basis for the prediction. This CAREER program will develop a Coupled Network Optimization (CNO) framework that balances the two analytical extremes. The resulting algorithms will reveal interpretable system-level interactions in the brain that can predict the behavioral and cognitive deficits of a given disorder. In parallel, the investigators have formulated a diverse range of educational initiatives to train the next generation of interdisciplinary data scientists.Mathematically, the CNO framework estimates a low-dimensional network manifold for functional neuroimaging data. The elemental bases of this manifold will correspond to interpretable group-level features, whereas the patient-specific projections will capture predictive information. The technical exploration of this award will unfold in three modular stages, each of which tackles an open challenge in the field. Thrust I will improve the CNO interpretability by imposing a patient-specific graph topology to guide the salient functional interactions. Thrust II will advance the CNO generalizability by introducing nonlinear and nonparametric regression models. Finally, Thrust III will leverage an equivalent Bayesian representation to tackle the challenges of multisite analysis. The CNO framework will be applied to two markedly different application testbeds: a large multi-site repository of neuroimaging, behavioral and genetic data for autism, and a focused clinical trial of functional electrical stimulation for spinal cord injury rehabilitation. Beyond the scientific goals, this award includes a three-pronged educational plan to build strong technical foundations, foster interdisciplinary collaborations through engagement and communication, and finally, motivate young women into the STEM fields. The investigators have outlined a comprehensive schedule of activities, ranging from curriculum development at the high school, undergraduate and graduate levels, to organizing student networking events, to mentoring high school women through the Johns Hopkins Women in Science and Engineering outreach.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

神经和神经精神疾病影响着全世界数百万人，并带来惊人的社会成本。尽管不断努力，临床医生对这些疾病只有粗略的了解，因此治疗它们的能力有限。从分析学的角度来看，临床神经科学是一个高维数据集、小样本量、大量患者变异性的领域，最重要的是，缺乏基础事实信息。这些挑战导致在给定模型的可解释性与其对新数据的可推广性之间进行权衡。在一个极端，经典统计学允许我们制定和测试关于大脑的可解释的假设，但它不能做出特定于患者的概括。在另一个极端，传统的机器学习算法面向患者的普遍性，但很少阐明预测的大脑基础。这个职业计划将开发一个耦合网络优化（CNO）框架，平衡两个极端的分析。由此产生的算法将揭示大脑中可解释的系统级相互作用，可以预测给定疾病的行为和认知缺陷。与此同时，研究人员制定了各种各样的教育计划，以培养下一代跨学科的数据科学家。在数学上，CNO框架估计了功能神经成像数据的低维网络流形。该流形的基本基础将对应于可解释的组水平特征，而患者特异性投影将捕获预测信息。该奖项的技术探索将分三个模块化阶段展开，每个阶段都将解决该领域的一个公开挑战。推力I将通过施加患者特异性图拓扑来指导突出的功能相互作用来提高CNO的可解释性。推力II将通过引入非线性和非参数回归模型来提高CNO的推广性。最后，Thrust III将利用等效的贝叶斯表示来应对多站点分析的挑战。CNO框架将应用于两个明显不同的应用测试平台：一个大型的多站点自闭症神经成像，行为和遗传数据库，以及一个针对脊髓损伤康复的功能性电刺激临床试验。除了科学目标之外，该奖项还包括三管齐下的教育计划，以建立强大的技术基础，通过参与和交流促进跨学科合作，最后激励年轻女性进入STEM领域。调查人员概述了一个全面的活动时间表，从高中、本科和研究生课程的开发，到组织学生网络活动，该奖项反映了NSF的法定使命，并被认为值得通过使用基金会的智力价值和更广泛的影响进行评估来支持审查标准。

项目成果

A joint network optimization framework to predict clinical severity from resting state functional MRI data.

• DOI: 10.1016/j.neuroimage.2019.116314
• 发表时间: 2020-02-01
• 期刊: NeuroImage
• 影响因子: 5.7
• 作者: D'Souza NS;Nebel MB;Wymbs N;Mostofsky SH;Venkataraman A
• 通讯作者: Venkataraman A

A Multi-Task Deep Learning Framework to Localize the Eloquent Cortex in Brain Tumor Patients using Both Static and Dynamic Functional Connectivity

使用静态和动态功能连接来定位脑肿瘤患者的口才皮层的多任务深度学习框架

• 发表时间: 2020
• 期刊: MLCN: MICCAI Workshop on Machine Learning for Clinical Neuroimaging
• 作者: Nandakumar, N.;D'Souza, N.S.;Manzoor, K.;Pillai, J.;Gujar, S.;Agarwal, S.;Sair, H.;Venkataraman, A
• 通讯作者: Venkataraman, A

RefineNet: An Automated Framework to Generate Task and Subject-Specific Brain Parcellations for Resting-State fMRI Analysis

RefineNet：用于生成用于静息态 fMRI 分析的任务和特定于主题的大脑分区的自动化框架

• 发表时间: 2022
• 期刊: Medical Image Computing and Computer Assisted Intervention
• 作者: Nandakumar, Naresh;Manzoor, Komal;Agarwal, Shruti;Sair, Haris I.;Venkataraman, Archana
• 通讯作者: Venkataraman, Archana

M-GCN: A Multimodal Graph Convolutional Network to Integrate Functional and Structural Connectomics Data to Predict Multidimensional Phenotypic Characterizations

• 发表时间: 2021
• 作者: N. S. D'Souza;M. B. Nebel;D. Crocetti;Joshua Robinson;S. Mostofsky;A. Venkataraman

Predicting Acute Kidney Injury via Interpretable Ensemble Learning and Attention Weighted Convoutional-Recurrent Neural Networks

• DOI: 10.1109/ciss50987.2021.9400242
• 发表时间: 2021-03
• 期刊: 2021 55th Annual Conference on Information Sciences and Systems (CISS)
• 作者: Yu-Chung Peng;N. S. D'Souza;Brian Bush;Charles H. Brown;A. Venkataraman