Abstract
In this study, we propose a novel point cloud based 3D registration and segmentation framework using reinforcement learning. An artificial agent, implemented as a distinct actor based on value networks, is trained to predict the optimal piece-wise linear transformation of a point cloud for the joint tasks of registration and segmentation. The actor network estimates a set of plausible actions and the value network aims to select the optimal action for the current observation. Point-wise features that comprise spatial positions (and surface normal vectors in the case of structured meshes), and their corresponding image features, are used to encode the observation and represent the underlying 3D volume. The actor and value networks are applied iteratively to estimate a sequence of transformations that enable accurate delineation of object boundaries. The proposed approach was extensively evaluated in both segmentation and registration tasks using a variety of challenging clinical datasets. Our method has fewer trainable parameters and lower computational complexity compared to the 3D U-Net, and it is independent of the volume resolution. We show that the proposed method is applicable to mono- and multi-modal segmentation tasks, achieving significant improvements over the state-of-the-art for the latter. The flexibility of the proposed framework is further demonstrated for a multi-modal registration application. As we learn to predict actions rather than a target, the proposed method is more robust compared to the 3D U-Net when dealing with previously unseen datasets, acquired using different protocols or modalities. As a result, the proposed method provides a promising multi-purpose segmentation and registration framework, particular in the context of image-guided interventions.
Highlights
Segmentation and registration of anatomical structures in 3D imaging are important pre-processing steps for clinical applications such as computer aided-diagnosis, treatment planning, or X-ray dose management
We have shown in Eq (14) that the Q-value is bounded, it is not practical to set the number of steps t → ∞, during inference
We evaluated our method for two different tasks, namely, segmentation and registration
Summary
Segmentation and registration of anatomical structures in 3D imaging are important pre-processing steps for clinical applications such as computer aided-diagnosis, treatment planning, or X-ray dose management. Reducing the inference time in multi-modal segmentation and registration is, essential to enable applications in interventional imaging and endovascular treatment settings, settings such as liver embolization or transarterial chemo embolization (TRACE) To address this challenge, we introduce a general scheme for joint point cloud-based joint segmentation and registration based on reinforcement learning. For model-based settings, CNNs have been used to predict the deformation of a segmented template[19], enabling single- and multi-atlas based segmentation[20] These methods have shown great success and found widespread adoption in most medical image segmentation tasks, there are some disadvantages. It could be caused by use of an entirely different imaging modality, for example, magnetic resonance tomography (MRT) instead of CT These factors inhibit the application of conventional CNN-based segmentation approaches (such as the U-Net) in clinical settings, where different imaging devices from various vendors set at site-specific acquisition protocols are used. This can be expressed using the optimal action-value (Q-value) function,
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have