Chest Pathology Detection in X-Ray Scans Using Social Spider Optimization Algorithm with Generalization Deep Learning

Abstract

Nowadays, medical diagnosis field using artificial intelligence is state of the art technology, which is growing and covering a lot of technical approaches to build pathology detection models. In our work, we used a new optimizer using Social Spider Optimization (SSO) algorithm with convolutional layer in deep neural network for chest X-ray pathology detection. SSO is one of the newest optimization algorithms. It is related to simulate the behavior of social spiders living in groups. Three main phases are used in this work to overcome the limited resources, the first phase is to build Convolutional Neural Network (CNN) with the benefits of transfer learning to reuse two blocks from VGG16 model (Visual Geometry Group), while the second one is to train CNN using gradient-based optimizer to extract feature vectors from the input images, and the third phase is to feed these vectors to two-fully connected layers and train it using SSO. These three phases reduce the number of trainable parameters, and get results that is much smaller and easy to handle network. We used SSO with deep neural network to achieve an accuracy of 89% and recall 98%, that leads to success of SSO optimizer to detect pathology of chest X-ray.