A Hybrid Model of PSO Algorithm and Artificial Neural Network for Automatic Follicle Classification

Polycystic Ovarian Syndrome (PCOS) is one of the leading causes of infertility in the world, but is a preventable disease when detected early. Detection of follicles in ultrasound images of the ovary is required for the diagnosis of PCOS. The manual method of detecting follicles is time consuming, laborious, error-prone and inconvenient for patients. However, methods used by the existing automated systems often lead to a reduction in accuracy, sensitivity and specificity due to the irregular and jagged edges of the follicles. This research work aims at achieving an improved specificity, sensitivity and accuracy of the system. In this report, a new technique for the automatic detection of follicles is implemented. Lee filter was used to despeckle the ultrasound images. Multiple features were then extracted from the images. Further, twelve of these features were selected as optimal values by the Particle Swarm Optimization algorithm. Then, these features were fed as input to the Multilayer Perceptron Artificial Neural Network. Upon training and testing the network, 98.3% accuracy, 100% sensitivity and 96.8% specificity were achieved