End to End Neural Network Application for Ultrasonic Computed Tomography Image Recognition

Deep learning techniques have known a big revolution in various areas including the industry, robotics, data security, signal processing and especially medical imaging. In this context, a new neural approach has been proposed in the ultrasonic medical imaging area. Thus, we have developed an end-to-end deep learning approach for the recognition of ultrasonic images. Our neural network system consists of an end-to-end variable structure of neural network (VSMN-VGG-Unet) for the segmentation of Ultrasonic Computed Tomographic (USCT) images. The experimental results are performed on our new USCT dataset, proving the efficiency of our proposed VSMN-VGG-Unet approach. The whole system is implemented on GPU Geforce GTX 1050Ti, achieving excellent accuracy results reaching 99.08% for training and 98% for the validation process. As for the test, the segmented USCT images show excellent segmented images with high resolution and a small error equal to 0.05. Moreover, we accelerate the time process speed, reaching 0.01s/image, hence surpassing the state of art.