Automatic View Planning with Multi-scale Deep Reinforcement Learning Agents

Abstract

We propose a fully automatic method to find standardized view planes in 3D image acquisitions. Standard view images are important in clinical practice as they provide a means to perform biometric measurements from similar anatomical regions. These views are often constrained to the native orientation of a 3D image acquisition. Navigating through target anatomy to find the required view plane is tedious and operator-dependent. For this task, we employ a multi-scale reinforcement learning (RL) agent framework and extensively evaluate several Deep Q-Network (DQN) based strategies. RL enables a natural learning paradigm by interaction with the environment, which can be used to mimic experienced operators. We evaluate our results using the distance between the anatomical landmarks and detected planes, and the angles between their normal vector and target. The proposed algorithm is assessed on the mid-sagittal and anterior-posterior commissure planes of brain MRI, and the 4-chamber long-axis plane commonly used in cardiac MRI, achieving accuracy of 1.53mm, 1.98mm and 4.84mm, respectively.

Learning to map 2D ultrasound images into 3D space with minimal human annotation

“…Such prediction is computed iteratively during inference. Recent studies (Alansary et al, 2018;Dou et al, 2019) proposed different reinforcement frameworks for standard plane localization in 3D MRI and ultrasound volumes. These RL frameworks provide feedback from the environment (i.e.…”

Artificial intelligence in pediatric and adult congenital cardiac MRI: an unmet clinical need

“…RL is performed by interacting with an environment instead of using a set of labeled examples. Alansary et al (43,44) proposed an RL-based approach for fully automatic view plane detection from 3D CMR data. Their model involves a complex search strategy with hierarchical action steps.…”

A Maximum Entropy Deep Reinforcement Learning Neural Tracker

“…Deep neural networks have been used successfully as function approximators of the policy and the state-action value function in multiple applications, including real-world visual navigation tasks where RL agents can learn directly from raw pixel values [2]. Their applications in biomedical imaging include landmark detection [6], view planning [1] and vascular centreline tracing [23], which all make use of deep Q-network algorithm [12], constraining them to using discrete action spaces. In order to predict the centreline observation points to subpixel accuracy, a continuous action space is required.…”