3D Medical Image Segmentation using Deep Reinforcement Learning: Analytical Study

Abstract

Detection and segmentation of medical objects are complex and critical steps in the field of medical image processing and analysis in order to detect the Region Of Interest areas (e.g., tumors and vertebral bodies), providing a reliable basis for clinical diagnosis to help clinicians make a more accurate diagnosis.

In recent years, deep learning techniques have achieved advanced results in this field, but they have suffered from some challenges such as the complex nature of 3D medical data, the small number of samples and the imbalance datasets, in addition to the problem of overfitting.

In this context, deep reinforcement learning has emerged as a trend in this field. This research aims to provide a comprehensive study of deep reinforcement learning techniques, their various algorithms, and their use in segmenting medical images, through a comparison of a range of published research up to 2026 in terms of: the type of datasets used, the algorithm used, the model structure, its application, the evaluation methods, the strengths and weaknesses, and the research gaps of deep reinforcement learning in the medical field.