Multi-scale Deep Reinforcement Machine Learning for N-Dimensional Segmentation in Medical Imaging

Simple SummaryContent extracted from patent full text and abstract with AI.

This patent discloses a method and system for automated segmentation of objects (such as organs or tissues) in multi-dimensional medical images (e.g., 3D CT or MRI scans). It uses a machine learning model based on multi-scale deep reinforcement learning to iteratively refine the segmentation boundaries. By formulating the segmentation as a reinforcement learning problem and incorporating shape models (statistical, front propagation, or voxel masks) at different scales, the system can robustly identify object boundaries across varied image types and patient data.

Use CasesContent extracted from patent full text and abstract with AI.

• Automatic segmentation of organs, tumors, or anatomical regions in 3D CT or MRI scans for diagnostic support.
• Pre-operative and intra-operative planning in surgery, providing accurate identification and visualization of anatomical structures.
• Assisting radiologists and clinicians in quantifying organ size, tumor volume, or disease progression by providing repeatable, objective segmentations.
• Supporting radiation therapy planning by segmenting treatment regions from medical images.
• Assisting in the construction of patient-specific anatomical models for simulation, education, or device customization.

BenefitsContent extracted from patent full text and abstract with AI.

• Enables more accurate and robust segmentation of complex anatomical structures compared to traditional methods.
• Reduces manual effort and inter-operator variability, saving clinician time and increasing reproducibility.
• Can adapt across different patients, image types, and object geometries due to its learning-based and multi-scale approach.
• Supports real-time or near real-time analysis for clinical workflows, especially with integration into imaging modalities and PACS.
• Iterative, policy-driven refinement improves resistance to artifacts, image noise, or poor initial segmentations compared to one-shot methods.

Technical Classifications (CPCs)

Inventors & Applicants

Patent Abstract

Multi-scale deep reinforcement learning generates a multi-scale deep reinforcement model (22) for multi-dimensional (e.g., 3D) segmentation of an object. In this context, segmentation is formulated as learning an image-driven policy (38) for shape evolution (40) that converges to the object boundary. The segmentation is treated as a reinforcement learning problem, and scale-space theory is used to enable robust and efficient multi-scale shape estimation. By learning an iterative strategy to find the segmentation, the learning challenges of end-to-end regression systems may be addressed.