Lung CT Image Registration through Landmark-constrained Learning with Convolutional Neural Network.

Accurate registration of lung computed tomography (CT) image is a significant task in thorax image analysis. Recently deep learning-based medical image registration methods develop fast and achieve promising performance on accuracy and speed. However, most of them learned the deformation field through intensity similarity but ignored the importance of aligning anatomical landmarks (e.g., the branch points of airway and vessels). Accurate alignment of anatomical landmarks is essential for obtaining anatomically correct registration. In this work, we propose landmark constrained learning with a convolutional neural network (CNN) for lung CT registration. Experimental results of 40 lung 3D CT images show that our method achieves 0.93 in terms of Dice index and 3.54 mm of landmark Euclidean distance on lung CT registration task, which outperforms state-of-the-art methods in registration accuracy.

Prediction of major torso organs in low-contrast micro-CT images of mice using a two-stage deeply supervised fully convolutional network.

Delineation of major torso organs is a key step of mouse micro-CT image analysis. This task is challenging due to low soft tissue contrast and high image noise, therefore anatomical prior knowledge is needed for accurate prediction of organ regions. In this work, we develop a deeply supervised fully convolutional network which uses the organ anatomy prior learned from independently acquired contrast-enhanced micro-CT images to assist the segmentation of non-enhanced images. The network is designed with a two-stage workflow which firstly predicts the rough regions of multiple organs and then refines the accuracy of each organ in local regions. The network is trained and evaluated with 40 mouse micro-CT images. The volumetric prediction accuracy (Dice score) varies from 0.57 for the spleen to 0.95 for the heart. Compared to a conventional atlas registration method, our method dramatically improves the Dice of the abdominal organs by 18%-26%. Moreover, the incorporation of anatomical prior leads to more accurate results for small-sized low-contrast organs (e.g. the spleen and kidneys). We also find that the localized stage of the network has better accuracy than the global stage, indicating that localized single organ prediction is more accurate than global multiple organ prediction. With this work, the accuracy and efficiency of mouse micro-CT image analysis are greatly improved and the need for using contrast agent and high x-ray dose is potentially reduced.

Estimation of thyroid volume from scintigraphy through 2D/3D registration of a statistical shape model.

Accurate measurement of thyroid volume is important for thyroid disease diagnosis and therapy. In nuclear medicine, the thyroid volume is usually estimated from scintigraphy images using empirical equations. However, due to the lack of volumetric information from the scintigraphy image, the accuracy of equation-based estimation is imperfect. To solve this problem, this paper proposes a method which registers a 3D thyroid statistical shape model (SSM) to a single-view scintigraphy image to achieve more accurate volume estimation. The SSM was constructed based on a training set of segmented 3D CT images, and the thyroid shape variations between the training subjects were modelled using the point distribution model. For thyroid volume estimation, the SSM was projected into the scintigraphy image of the target patient, and then the projected model shape was nonrigidly registered with the patient's scintigraphy image. The resultant 2D deformation file was back-projected to 3D space to guide the deformation of the 3D SSM. This process was repeated iteratively until convergence, and the volume of the finally deformed SSM was considered as the estimation of the patient's thyroid volume. For validation, this method was evaluated based on a test set of 20 scintigraphy images, achieving an estimation error of -2.10% ± 5.20% which was much less than the error of the conventional equation-based method (35.76% ± 15.20%) based on the same test set. The robustness of this method was further tested using a challenging case, i.e. a scintigraphy image with a large thyroid tumor. For this case, the volume estimation error was only 6.08%. Our method has significantly improved the accuracy of thyroid volume estimation from scintigraphy images, and it will enhance the value of scintigraphy imaging for thyroid disease diagnosis and radioiodine therapy.