Auxiliary Diagnosis of Breast Tumor Based on PNN Classifier Optimized by PCA and PSO Algorithm

Abstract

Facing the condition that misdiagnoses of breast tumor recognition are easily caused while using fine needle aspiration cytology, this paper proposes a method of breast tumor auxiliary diagnosis based on Probabilistic Neural Network (PNN) classifier optimized by Principal Component Analysis (PCA) and Particle Swarm Optimization (PSO). The main characteristics of breast tumor data are extracted by PCA, which can simplify the PNN structure and improve the training speed of PNN without losing recognition accuracy. The PSO is applied to select the best smoothing factor of PNN, and the optimal parameter will be used for training PNN. The recognition of breast tumor is finally implemented by using the improved PNN model, which realizes the auxiliary diagnosis of breast tumor. The simulation results show that the PCA-PSO-PNN model is superior to the traditional back-propagation (BP) neural network, the learning vector quantization (LVQ) neural network, and the unoptimized PNN in terms of recognition accuracy, sensitivity and specificity. The algorithm and simulation have been implemented in MATLAB.