Aggarwal Yogender, Karan Bhuwan Mohan, Das Barda Nand, Sinha Rakesh Kumar
Department of Biomedical Instrumentation, Birla Institute of Technology, Mesra, Ranchi, 835215, Jharkhand, India.
J Clin Monit Comput. 2008 Dec;22(6):425-30. doi: 10.1007/s10877-008-9152-x. Epub 2008 Nov 25.
Heat stress is known to induce high mortality rate due to multi-system illness, which demands urgent attention to reduce the fatality rate in such patients. Further, for the diagnosis and supportive therapy, one needs to define the severity of heat stress that can be distinguished as mild, intermediate and severe. The objective of this work is to develop an automated unsupervised artificial system to analyze the clinical outcomes of different levels of heat related illnesses. The Kohonen neural network program written in C++, which has seven normalized values of different clinical symptoms between 0-1 fed to the input layer of the network with 50 Kohonen output neurons, has been presented. The optimized initializing parameters such as neighborhood size and learning rate was set to 50 and 0.7, respectively, to simulate the network for 10 million iterations. The network was found smartly distinguishing all 51 patterns to three different states of heat illnesses. With the advent of these findings, it can be concluded that the Kohonen neural network can be used for automated classification of the severity of heat stress and other related psycho-patho-physiological disorders. However, to replace the expert clinicians with such type of smart diagnostic tool, extensive work is required to optimize the system with variety of known and hidden clinical and pathological parameters.
众所周知，热应激会因多系统疾病导致高死亡率，这需要紧急关注以降低此类患者的死亡率。此外，对于诊断和支持性治疗，需要确定热应激的严重程度，可分为轻度、中度和重度。这项工作的目的是开发一个自动化的无监督人工系统，以分析不同程度热相关疾病的临床结果。本文介绍了用C++编写的Kohonen神经网络程序，该程序将0到1之间的七个不同临床症状的归一化值输入到具有50个Kohonen输出神经元的网络输入层。将优化的初始化参数，如邻域大小和学习率分别设置为50和0.7，对网络进行1000万次迭代模拟。结果发现该网络能够巧妙地将所有51种模式区分为三种不同的热疾病状态。基于这些发现，可以得出结论，Kohonen神经网络可用于热应激严重程度及其他相关心理病理生理障碍的自动分类。然而，要用这种智能诊断工具取代专家临床医生，还需要进行大量工作，用各种已知和隐藏的临床及病理参数对系统进行优化。
