Improvement in the Convolutional Neural Network for Computed Tomography Images

Background and purpose. This study evaluated a modified specialized convolutional neural network (CNN) to improve the accuracy of medical images. Materials Methods. We defined computed tomography (CT) images as belonging one following 10 classes: head, neck, chest, abdomen, pelvis with without contrast media, 10,000 per class. CNN based on AlexNet an input size 512 × 512. resized filter sizes convolution layer max pooling. Using these CNNs, various models were created evaluated. The improved was classify presence or absence pancreas in CT compared overall accuracy, which calculated from not used for training, that ResNet. Results. accuracies most ResNet classes 94.8% 89.3%, respectively. (13, 13), (7, 7), (5, 5), 5) order first layer, max-pooling 7). calculation times 56 120 min, Regarding classification pancreas, 75.75% 58.25%, 36 55 Conclusion. By optimizing images, we quickly obtained highly accurate image model. can be useful classifying lesions anatomies related diagnostic aid applications.