Partial Policy-based Reinforcement Learning for Anatomical Landmark\n Localization in 3D Medical Images

Abstract

Deploying the idea of long-term cumulative return, reinforcement learning has\nshown remarkable performance in various fields. We propose a formulation of the\nlandmark localization in 3D medical images as a reinforcement learning problem.\nWhereas value-based methods have been widely used to solve similar problems, we\nadopt an actor-critic based direct policy search method framed in a temporal\ndifference learning approach. Successful behavior learning is challenging in\nlarge state and/or action spaces, requiring many trials. We introduce a partial\npolicy-based reinforcement learning to enable solving the large problem of\nlocalization by learning the optimal policy on smaller partial domains.\nIndependent actors efficiently learn the corresponding partial policies, each\nutilizing their own independent critic. The proposed policy reconstruction from\nthe partial policies ensures a robust and efficient localization utilizing the\nsub-agents solving simple binary decision problems in their corresponding\npartial action spaces. The proposed reinforcement learning requires a small\nnumber of trials to learn the optimal behavior compared with the original\nbehavior learning scheme.\n

Ähnliche Arbeiten