A CNN-Based Approach for Lung 3D-CT Registration

Abstract

Deep learning techniques have been applied to certain rigid or non-rigid medical image registration due to its potential advantages in meeting the clinical requirements of real-time and accuracy. Based on the deep learning model, this study aims to explore specific network models suitable for lung CT images. The proposed model took unlabeled 3D image pairs as input, and the convolutional neural network (CNN) was utilized and identified as a function with ability of sharing parameters to obtain displacement field. The image pair could be aligned by applying the acquired displacement field to the target image through spatial transformation. The similarity between the aligned image pair combined with the constraints on the displacement field was taken as the objective function to obtain the optimal parameters. Two models with different depths were designed and the consequent registration effects with different optimization methods and convolution kernel sizes were explored. The results proved that the designs with deeper level using Adam optimizer and smaller convolution kernels in obtaining displacement fields had higher accuracy and stronger robustness. The accuracy of the unsupervised model was comparable to state-of-theart methods, while operating orders of magnitude faster. This study proposed a feasible registration method for lung 3D-CT, and its usefulness in aligning CT images has been demonstrated.

An unsupervised multi-scale framework with attention-based network (MANet) for lung 4D-CT registration

“…A regularization term of DVF smoothness was contained in the loss function to ensure the realistic of the generated DVF. According to the methods presented in literatures (Bender et al 2012, Jaderberg et al 2015, Zhang 2018, Hu et al 2020, commonly used DVF regularization terms mainly include DVF smoothness constraint, anti-folding constraint and inverse consistency constraint. However, whether these constraints are adequate for proper regularization of DVF is ambiguous.…”

“…Figure 2(A) demonstrates the attention gates in the generator architecture that are used to combine the features of adjacent layers from different scales. The overall framework of CNN used here was proposed in our previous publication (Hu et al 2020). Although the convolution model performed well in the registration task, the architecture usually relied on multiple cascaded layers to progressively extract features and make point-to-point predictions when the target organs had large differences in shape and size.…”

Modeling Respiratory Signals by Deformable Image Registration on 4DCT Lung Images

“…In this research, the 4DCT dataset was acquired as a part of the standard planning process for the treatment of thoracic malignancies at The University of Texas M. D. Anderson Cancer Center in Houston and offered by DIR-LAB [ 3 ]. In 4DCT imaging, thoracic movements are monitored by a Varian Real-time Position Management (RPM) system during the CT scan.…”

“…This approach is rather complicated and requires a model initialization for the process. For registration, some researchers use deep learning approaches based on the displacement field to obtain the optimal parameters [ 3 ], which must be trained with big data until reaching the optimization. Some other approaches require a landmark tracking process [ 4 ], which must be determined by specialists.…”

Adversarial-Learning-Based Taguchi Convolutional Fuzzy Neural Classifier for Images of Lung Cancer

“…Zhang et al [1] combined multiple convolutional neural network (CNN) models to construct an ensemble learner for the classification of pulmonary nodules. Hu et al [2] proposed STN model with unsupervised learning method to label malignant lung cancer. Shen et al [3] introduced an interpretable deep hierarchical semantic CNN to predict the malignancy of pulmonary nodules observed in CT images.…”