Evaluation of Methods Concerning Health Risk Based on Artificial Intelligence

本研究的目的是阐明神经网络系统在医学领域，特别是在健康风险评估方面的有用性和局限性。这项研究的第一年被分配到研究环境的准备。完成了以下课题；从医学数据库中提取实验数据；2 .基于FORTRAN语言编写的误差反向传播算法的神经网络程序开发。建立微机计算神经网络程序。第二年是神经网络系统与普通统计方法（判别分析）诊断能力的比较。本研究使用的数据涉及三种肝胆疾病。研究对象如下：修改反向传播以显示学习过程中的顺序误差；神经网络对外部测试数据的总体诊断率为97.7%(212/217)，判别函数的总体诊断率为65.4%（142/217）。神经网络的诊断能力明显高于普通统计方法。此外，神经网络系统不需要数据的正态分布等统计假设。根据这些发现，神经网络系统可能是评估健康风险的有用工具。

The purpose of this study is to clarify the usefulness and the limitation of the neural network systems applied to the medical fields, especially to health risk assessment.First years of this study is assigned to the preparation of the study environment. The following subjects were done ;1. Extraction of experimental data from the medical database, 2. Development of the neural network program based on the error back propagation algorithm written by FORTRAN language, 3. Setup of the microcomputer for calculation of neural network program.Second year was assigned to the comparison of the diagnostic capabilities between the neural network system and the ordinary statistical method (discriminant analysis). The data used in this study was concerning three hepatobiliary disorders. Following subjects were done ;1. Modification of the back propagation to display the sequential error in the learning process, 2. Overall correct diagnostic rate of the external test data was 97.7% (212/217) in the neural network, and 65.4% (142/217) in the discriminant function. The diagnostic capability of the neural network was significantly higher than that of the ordinary statistical method.In addition, neural network system required no statistical assumptions such as the normal distribution of the data. According to these finding, neural network system might be useful tools for estimating the health risks.