3D Neural Voxel Correction for CT Metal Artifacts

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for correcting metal artifacts in CT images, particularly in low-dose cone beam CT scans, are inefficient and require substantial computational resources, failing to provide a generic solution for automatically removing or reducing artifacts in x-ray data without knowledge of the scanning and backprojection methods used.

Innovation Solution

A system utilizing two 3D deep neural networks to automatically correct metal-affected voxel representations by distinguishing between voxels affected by metal artifacts and tissue, replacing artifact voxels with accurate estimates based on clinical training data, without processing sinogram domain data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional backprojection algorithms are used to reconstruct CT images, then image reconstruction is achieved, but metal artifacts appear due to high absorption coefficients of metallic materials

Engineering Contradiction:
Improveimage reconstruction accuracyVSAvoidmetal artifacts
Core Design Contradiction:
ReliabilityVSObject-generated harmful factors

Solution Approach 1:

The patent segments the image reconstruction process into two distinct phases: first reconstructing the image using conventional backprojection algorithms, then applying a separate deep learning-based correction phase to remove metal artifacts. This segmentation allows each phase to optimize for its specific function without interfering with the other.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces a deep learning-based correction system as an intermediary between the conventional backprojection algorithm and the final diagnostic image. This intermediary processes the reconstructed image to identify and remove metal artifacts while preserving genuine anatomical structures, effectively mediating between the raw reconstruction and the clinically useful image.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Object-affected harmful factors

If dose reduction is implemented in CT examinations, then radiation exposure is reduced, but noise in reconstructed images increases

Engineering Contradiction:
Improveradiation exposureVSAvoidimage quality
Core Design Contradiction:
Object-affected harmful factorsVSReliability

Solution Approach 1:

The patent replaces traditional mechanical/deterministic image processing methods with a data-driven deep learning approach. The neural network learns to distinguish between noise and genuine anatomical structures from training data, enabling effective noise reduction in low-dose images without requiring increased radiation exposure.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Object-generated harmful factors

If existing metal artifact reduction methods are used, then some artifacts are reduced, but substantial computational resources are required and specific knowledge of scanning methods is needed

Engineering Contradiction:
Improvemetal artifact reductionVSAvoidcomputational resources and system knowledge requirements
Core Design Contradiction:
Object-generated harmful factorsVSDevice complexity

Solution Approach 1:

The patent creates a corrected copy of the reconstructed image by generating a metal artifact mask and subtracting it from the original image. This copying approach allows the correction to be applied universally without requiring knowledge of the specific scanning or backprojection methods used to generate the original image.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The deep learning-based correction system is designed to be universal and method-agnostic, working with images from different CT scanners and reconstruction algorithms without requiring specific knowledge of the scanning or processing methods. The system processes the final reconstructed image directly, making it compatible with various imaging systems.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

Effectively removes metal artifacts from 3D CT images, improving image quality by preserving crucial tissue and metal information, while reducing computational overhead and eliminating the need for specific knowledge of the scanning system.

Implementation Method 1

different areas of a tissue are exposed at different angles by an x-ray radiation beam of a predetermined intensity while an array of sensors measures the intensity of the x-ray beam that travelled through the tissue. Based on the attenuated radiation beam measured by the sensors, the absorption of the x-ray radiation at different areas in the tissue can be determined.

Methodology Applied
Scientific EffectX-ray attenuation: Absorption (EM radiation)

Implementation Method 2

a second 3D deep neural network receiving the initial voxel representation and the voxel map generated by the first 3D deep neural network at its input and generating a corrected voxel representation at its output, the corrected voxel representation including voxel estimations for voxels that are identified by a voxel map as being part of a metal affected region

Methodology Applied
Scientific EffectDeep learning pattern recognition:

Data Source

PatentEP3785234B1Automated correction of metal affected voxel representations of x-ray data using deep learning techniques
Publication Date: 2025.12.31 PROMATON HLDG BV
  • EP3785234B1 patent drawingFigure 1~2
  • EP3785234B1 patent drawingFigure 3
  • EP3785234B1 patent drawingFigure 4

AI summary

A computer-implemented method for correction of a voxel representation of metal affected x-ray data is described,themetal affected x-ray data representing artefacts in the x-ray data caused by metal or metallic objects in a volume of tissue that is imaged by an x-ray imager, wherein the method comprises a first 3D deep neural network receiving an initial voxel representation of metal affected x-ray data at its input and generating a voxel map at its output, the voxel map identifying voxels of the initial voxel representation that belong to a region of voxels that are affected by metal; and, a second 3D deep neural network receiving the initial voxel representation and the voxel map generated by the first 3D deep neural network at its input and generating a corrected voxel representation,the corrected voxel representation including voxel estimations for voxels that are identified by the voxel map as being part of a metal affected region, the first 3D deep neural being trained on the basis of training data and reference data that include voxel representations of clinical x- ray data of a predetermined body part of a patient.