Estimation of the volume of the left ventricle from MRI images using deep neural networks

摘要

Segmenting human left ventricle (LV) in magnetic resonance imaging (MRI)images and calculating its volume are important for diagnosing cardiacdiseases. In 2016, Kaggle organized a competition to estimate the volume of LVfrom MRI images. The dataset consisted of a large number of cases, but onlyprovided systole and diastole volumes as labels. We designed a system based onneural networks to solve this problem. It began with a detector combined with aneural network classifier for detecting regions of interest (ROIs) containingLV chambers. Then a deep neural network named hypercolumns fully convolutionalnetwork was used to segment LV in ROIs. The 2D segmentation results wereintegrated across different images to estimate the volume. With ground-truthvolume labels, this model was trained end-to-end. To improve the result, anadditional dataset with only segmentation label was used. The model was trainedalternately on these two datasets with different types of teaching signals. Wealso proposed a variance estimation method for the final prediction. Ouralgorithm ranked the 4th on the test set in this competition.