基于机器学习的阿尔兹海默症智能分类研究

Intelligent Classification of Alzheimer’s Disease Based on Machine Learning

摘要: 阿尔兹海默症(AD)是一种使人记忆力衰退、大脑功能缓慢且逐渐变异的智力丧失表现的疾病。而早期诊断和预测AD可以延缓患者的发病，可能延长患者的寿命，对整个社会具有重要的科学意义。本文通过机器学习等方法对AD数据进行分析和建模，增添大脑特征指标，进行数据预处理，利用成对分类法的SVM实现AD的智能诊断，通过SVM得到的五分类的准确度达到0.71，高于随机森林的0.69与XGBoost提升树的0.706。并通过绘制ROC曲线，对于判断CN有64%准确率，判断AD有80%准确率，判断LMCI有68%准确率，判断SMC有58%准确率，判断EMCI有76%准确率，具有较好的测试效果。提取和发现影响AD诊断和预测的关键症状，并对患者类别进行分类，能够有效、快速地得到早期诊断和预测AD的患病概率，以达到预防AD的有效手段。基于成对分类法的SVM实现AD的智能诊断，并且能够用推广于更加多分类的问题。

Abstract: Alzheimer’s disease (AD) is a disease that causes memory loss and slow and gradual variation in brain function as a manifestation of intellectual loss. Early diagnosis and prediction of AD can delay the onset of the disease, potentially prolong the life span of patients, and is of great scientific im-portance to society as a whole. In this paper, we analyzed and modeled AD data by machine learning, added brain feature indicators, preprocessed data, and used SVM with pairwise classification method to achieve intelligent diagnosis of AD. The accuracy of the five classifications obtained by SVM reaches 0.71, which is higher than 0.69 of random forest and 0.706 of XGBoost boosted tree. The accuracy of the ROC curve is 64% for CN, 80% for AD, 68% for LMCI, 58% for SMC, and 76% for EMCI, which is a good test result. By extracting and discovering the key symptoms affecting the di-agnosis and prediction of AD, and classifying the patient categories, we can effectively and quickly obtain early diagnosis and predict the probability of AD, so as to achieve an effective means of pre-venting AD. The SVM based on pairwise classification method enables intelligent diagnosis of AD and can be extended to more multi-categorical problems.