Principal component analysis based backpropagation algorithm for diagnosis of peripheral arterial occlusive diseases

Abstract

Impedance cardio-vasography (ICVG) serves as a non-invasive screening procedure prior to invasive and expensive angiographic studies. Parameters like Blood Flow Index (BFI) and Differential Pulse Arrival Time (DPAT) at different locations in both lower limbs are computed from impedance measurements on the Impedance Cardiograph. A Backpropagation neural network is developed which uses these parameters for the diagnosis of peripheral vascular diseases such as Leriche's syndrome. The target outputs at the various locations are provided to the network with the help of a medical expert. The paper proposes the use of principal component analysis (PCA) based backpropagation network where the variance in the data is captured in the first seven principal components out of a set of fourteen features. Such a backpropagation algorithm with three hidden layers provides the least mean squared error for the network parameters. The results demonstrated that the elimination of correlated information in the training data by way of the PCA method improved the networks estimation performance. The cases of arterial Narrowing were predicted accurately with PCA based technique than with the traditional backpropagation Technique. The diagnostic performance of the neural network to discriminate the diseased cases from normal cases, evaluated using Receiver Operating Characteristic (ROC) analysis show a sensitivity of 95.5% and specificity of 97.36% an improvement over the performance of the conventional Backpropagation algorithm. The proposed approach is a potential tool for diagnosis and prediction for non-experts and clinicians.