Volume Data Rendering Using Historical Frames and Neural Path Tracing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Physically based rendering techniques for volume data, such as medical 3D imaging, are computationally intensive and require restarting path tracing operations when the camera moves, leading to inefficiencies and increased computational demands.

Innovation Solution

A system and method that utilizes a neural network model to generate high-resolution medical images by inputting initial images, casting parameters, and historical image data, allowing for efficient volume data rendering by leveraging previous frame data to avoid restarting path tracing operations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If path tracing operation is restarted from the beginning when camera moves, then rendering accuracy is maintained, but computational load increases significantly and processing time extends

Engineering Contradiction:
Improverendering accuracyVSAvoidprocessing speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The system performs preliminary path tracing operations for multiple camera positions in advance, storing the results as pre-computed data. When the camera moves to a new position, the system retrieves and combines pre-computed results from nearby positions rather than performing complete path tracing from scratch, significantly reducing computational load while maintaining rendering quality

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system creates copies of path tracing results from previous frames and camera positions. Instead of computing entirely new images when the camera moves, it copies and interpolates results from nearby pre-computed positions, adjusting them to match the new camera viewpoint. This approach maintains accuracy while dramatically improving processing speed

Inventive Principle:
Principle #26Copying

2Reliability

If path tracing operation is restarted from the beginning when camera moves, then rendering quality is preserved, but continuity of rendering process is broken and computational resources are wasted

Engineering Contradiction:
Improverendering qualityVSAvoidcomputational time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system maintains continuous rendering by pre-computing and storing path tracing results for multiple camera positions before actual rendering begins. When the camera moves during rendering, the system continuously retrieves and combines pre-computed results without interrupting the rendering process, ensuring both quality and continuity

Inventive Principle:
Principle #20Continuity of useful action

Solution Approach 2:

All necessary path tracing computations are performed in advance for multiple camera positions and stored as pre-computed data. This preliminary action ensures that when rendering occurs, the system can quickly retrieve and combine results without performing computationally intensive operations during the actual rendering process, eliminating quality loss while reducing computational time

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4377889B1Systems and methods for volume data rendering
Publication Date: 2026.01.07 SHANGHAI UNITED IMAGING HEALTHCARE
  • EP4377889B1 patent drawingFigure 1~2
  • EP4377889B1 patent drawingFigure 3~4
  • EP4377889B1 patent drawingFigure 5~6

AI summary

The present disclosure relates to systems and methods for volume data rendering. The methods may include obtaining an initial image relating to volume data of a subject by performing a path tracing operation on the volume data, obtaining a total sampling operation count corresponding to the initial image, determining, based on the volume data, one or more reference parameters relating to the initial image, and generating, based on historical image data relating to the volume date of the subject, the initial image, the total sampling operation count, and the one or more reference parameters, a target image using an image processing model.