An Automated Deep Learning-Based Software for Total Lung Volume Calculation and Lung 3D Model Reconstruction

Abstract

In this paper, we propose an automated deep learning-based software, especially for reconstructing 3D lung images and estimating Total Lung Volume (TLV). The system is mainly designed for Vietnamese users with Vietnamese as the default language. The purpose of our study is to build an automated system based on deep learning and machine-learning models to measure TLV and construct a 3D prototype of the lung. The training data is collected from The Cancer Imaging Archive (TCIA) dataset, provided by the 2017 Lung CT Segmentation Challenge for training and testing the proposed model. Our proposed system utilizes a modified Bi-directional Convolutional (ConvLSTM) U-Net (BCDU-Net) neural network. We use the lung segmentation results as the data to build 3D lung image and calculate the lung volume. We test the reliability of the software with real dataset collected from Bach Mai hospital. The overall results have shown that Accuracy against actual data from commercial products is approximately 99% when using our model to calculate the TLV.