Convolutional Neural Networks for Epileptic Disease Detection

The Convolutional Neural Networks (CNN) offers automatic monitoring and understanding of facial activities of patients in clinical systems. We investigate the feasibility of predicting epileptic disease by using measurements of detection systems based on CNN during the outbreak of epileptic disease. In this work, we present a detector which achieves the balance of the accuracy and speed. We perform the pre-training model on the COCO dataset, reducing computation cost and memory usage. Then, we effectively optimize SSD-Inception V2 model by adjusting the weights based on the VOC2012 dataset. The fine-tuned CNN model increases the predictive accuracy and accelerates the convergence speed of training. Our experiments show that the fine-turned method outperforms approaches based on other learned methods on the effectiveness for epileptic disease detection tasks. The results of these studies indicate that fine-tuned CNN model could be a powerful tool on the current search for epileptic disease and other brain-based disorders prediction.