Convolutional Neural Networks Implemented by TensorFlow for the Segmentation of Left Ventricular MRI Images

摘要

Convolutional neural networks applied to medical image segmentation with low-compute power computers are investigated. A convolutional neural network architecture is implemented by TensorFlow for a synthetic image and left ventricular MRI image segmentation. Finding parameters of a trained model is an optimization problem. A flowchart is proposed to solve the compute resource problem for the training procedure. The main point is to train a model by updating the checkpoint step by step. With the limitation of the low compute power, experiments can also have good enough results. The experiment results in image segmentation are good enough. The accuracy of both of them is above 95%.