A machine learning approach for predicting post-stroke aphasia recovery: a pilot study
A machine learning approach for predicting post-stroke aphasia recovery: a pilot study
复制标题
DOI：
10.1145/3389189.3389204
复制
发表时间：
2020-06
期刊：
影响因子：
--
通讯作者：
Yiwen Gu;Murtadha Bahrani;Anne Billot;Sha Lai;Emily J Braun;M. Varkanitsa;Julia Bighetto;B. Rapp;T. Parrish;D. Caplan;C. Thompson;S. Kiran;Margrit Betke
中科院分区：
文献类型：
--
作者：
Yiwen Gu;Murtadha Bahrani;Anne Billot;Sha Lai;Emily J Braun;M. Varkanitsa;Julia Bighetto;B. Rapp;T. Parrish;D. Caplan;C. Thompson;S. Kiran;Margrit Betke
卒中后失语症的潜在恢复是高度可变的，康复结果很难预测。这种跨学科的合作建立在收集的数据基础上，这些数据是中风后失语症患者大量行为和大脑变量的一部分，绘制了与跨语言领域治疗相关的康复过程，并检查了神经可塑性的基础。在这项试点研究中，我们基于收集的数据子集创建并测试了一个预测框架，并开发了机器学习算法，该算法将一组复杂的大脑和行为特征作为输入，以分类和预测参与者对治疗的反应。我们开发了随机森林模型，使我们能够对这些功能的重要性进行排名。然后，我们比较了不同特征集的贡献，并讨论了它们的生理意义。我们的初步结果表明了我们框架的潜力，因此，这项研究朝着预测个性化康复结果迈出了重要的第一步。
The potential recovery of post-stroke aphasia is highly variable and the rehabilitation outcomes are difficult to predict. This interdisciplinary collaboration builds on data collected as part of a large set of behavioral and brain variables in patients with post-stroke aphasia, charting the course of recovery associated with therapy across language domains and examining the basis of neuroplasticity. In this pilot study, we created and tested a predictive framework based on a subset of the data collected and developed machine-learning algorithms that take as input a complex set of brain and behavioral features to classify and predict the participants' responsiveness to therapy. We developed Random Forest models that enabled us to rank the importance of these features. We then compared the contributions of different feature sets and discussed their physiological implications. Our preliminary results suggest the potential of our framework, and, thus, this study takes an important first step towards predicting individualized rehabilitation outcomes.