Federal Rural University of Pernambuco, Brazil; Catholic University of Pernambuco, Brazil.
Federal Rural University of Pernambuco, Brazil; University of Pernambuco, Brazil.
Comput Methods Programs Biomed. 2017 Dec;152:93-104. doi: 10.1016/j.cmpb.2017.09.013. Epub 2017 Sep 20.
A huge number of solutions based on computational systems have been recently developed for the classification of cognitive abnormalities in older people, so that individuals at high risk of developing neurodegenerative diseases, such as Cognitive Impairment and Alzheimer?s disease, can be identified before the manifestation of the diseases. Several factors are related to these pathologies, making the diagnostic process a hard problem to solve. This paper proposes a computational model based on the artificial neural network to classify data patterns of older adults.
The proposal takes into account the several parameters as diagnostic factors as gender, age, the level of education, study time, and scores from cognitive tests (Mini-Mental State Examination, Semantic Verbal Fluency Test, Clinical Dementia Rating and Ascertaining Dementia). This non-linear regression model is designed to classify healthy and pathological aging with machine learning techniques such as neural networks, random forest, SVM, and stochastic gradient boosting. We deployed a simple linear regression model for the sake of comparison. The primary objective is to use a regression model to analyze the data set aiming to check which parameters are necessary to achieve high accuracy in the diagnosis of neurodegenerative disorders.
The analysis demonstrated that the usage of cognitive tests produces median values for the accuracy greater than 90%. The ROC analysis shows that the best sensitivity performance is above 98% and specificity of 96% when the configurations have only cognitive tests.
The presented approach is a valuable tool for identifying patients with dementia or MCI and for supporting the clinician in the diagnostic process, by providing an outstanding support decision tool in the diagnostics of neurodegenerative diseases.
最近已经开发了大量基于计算系统的解决方案，用于对老年人的认知异常进行分类，以便能够在神经退行性疾病（如认知障碍和阿尔茨海默病）表现出来之前识别出有患病风险的个体。有几个因素与这些病理相关，使得诊断过程成为一个难以解决的问题。本文提出了一种基于人工神经网络的计算模型，用于对老年人的数据模式进行分类。
该提案考虑了几个参数作为诊断因素，包括性别、年龄、教育水平、学习时间以及认知测试（简易精神状态检查、语义流畅性测试、临床痴呆评定和确定痴呆）的分数。这个非线性回归模型旨在使用机器学习技术（如神经网络、随机森林、SVM 和随机梯度提升）对健康和病理性衰老进行分类。为了进行比较，我们部署了一个简单的线性回归模型。主要目标是使用回归模型来分析数据集，以检查哪些参数对于实现神经退行性疾病的高诊断准确性是必要的。
分析表明，使用认知测试产生的准确性中位数大于 90%。ROC 分析表明，当配置仅包括认知测试时，最佳的敏感性性能超过 98%，特异性为 96%。
所提出的方法是一种识别痴呆症或 MCI 患者的有价值的工具，并通过提供神经退行性疾病诊断中的出色支持决策工具，为临床医生提供支持诊断过程。
