Giansanti Daniele, Maccioni Giovanni, Cesinaro Stefano, Benvenuti Francesco, Macellari Velio
Dipartimento di Tecnologie e Salute, Istituto Superiore di Sanità, Rome, Italy.
Med Eng Phys. 2008 Apr;30(3):367-72. doi: 10.1016/j.medengphy.2007.04.006. Epub 2007 Jun 8.
We have investigated the use of an Artificial Neural Network (ANN) for the assessment of fall-risk (FR) in patients with different neural pathologies. The assessment integrates a clinical tool based on a wearable device (WD) with accelerometers (ACCs) and rate gyroscopes (GYROs) properly suited to identify trunk kinematic parameters that can be measured during a posturography test with different constraints. Our ANN--a Multi Layer Perceptron Neural Network with four layers and 272 neurones--shows to be able to classify patients in three well-known fall-risk levels. The training of the neural network was carried on three groups of 30 subjects with different Fall-Risk Tinetti scores. The validation of our neural network was carried out on three groups of 100 subjects with different Fall-Risk Tinetti scores and this validation demonstrated that the neural network had high specificity (> or =0.88); sensitivity (> or =0.87); area under Receiver-Operator Characteristic Curves (>0.854).
我们研究了使用人工神经网络（ANN）评估不同神经病变患者的跌倒风险（FR）。该评估将基于可穿戴设备（WD）的临床工具与加速度计（ACC）和速率陀螺仪（GYRO）相结合，这些设备适用于识别在不同约束条件下姿势描记测试期间可测量的躯干运动学参数。我们的人工神经网络——一个具有四层和272个神经元的多层感知器神经网络——显示能够将患者分类为三个众所周知的跌倒风险级别。神经网络的训练是在三组各30名具有不同跌倒风险Tinetti评分的受试者上进行的。我们的神经网络在三组各100名具有不同跌倒风险Tinetti评分的受试者上进行了验证，该验证表明神经网络具有高特异性（≥0.88）；高敏感性（≥0.87）；受试者操作特征曲线下面积（>0.854）。
