Medical Image Denoising via Neural Network Patch Processing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current medical imaging technologies face challenges in accurately removing noise from medical images, leading to blurred features and the need for repeated scans or increased radiation exposure, which can be detrimental to patients and delay diagnosis.

Innovation Solution

A method and system utilizing a machine learning model, specifically convolutional neural networks, to process medical images by constructing pixel matrices, identifying noise, and applying pre-defined pixel matrices to generate denoised images, thereby improving image clarity and reducing the need for multiple scans.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Object-affected harmful factors

If traditional noise removal methods are applied to medical images, then noise reduction is achieved, but image clarity and diagnostic accuracy deteriorate due to blurred features

Engineering Contradiction:
Improvenoise in medical imageVSAvoidimage clarity and feature sharpness
Core Design Contradiction:
Object-affected harmful factorsVSManufacturing precision

Solution Approach 1:

The image is divided into multiple patches, and each patch is processed independently to identify and remove noise while preserving local features. This segmentation approach allows targeted noise removal without affecting overall image clarity.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

A deep learning model acts as an intermediary between the noisy medical image and the final denoised output. The model learns to distinguish between noise and actual image features, transforming the noisy input into a clear output while preserving diagnostic information.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Manufacturing precision

If patients undergo multiple image acquisition processes to obtain clear images, then diagnostic accuracy improves, but radiation exposure and time consumption increase

Engineering Contradiction:
Improveimage quality for diagnosisVSAvoidradiation exposure to patient
Core Design Contradiction:
Manufacturing precisionVSObject-affected harmful factors

Solution Approach 1:

The invention converts the harmful noise present in low-dose or single-scan images into a solvable problem through deep learning denoising. Instead of requiring additional scans to overcome noise, the model learns to recognize and remove noise patterns, transforming noisy single-scan images into diagnostic-quality images without additional radiation exposure.

Inventive Principle:
Principle #22Blessing in disguise (Convert harm into benefit)

Solution Approach 2:

The deep learning model is trained on pairs of noisy and clean images, learning to generate clean image copies from noisy inputs. This allows the system to produce multiple high-quality image copies from a single noisy scan, eliminating the need for repeated patient exposure.

Inventive Principle:
Principle #26Copying

3Object-affected harmful factors

If deep learning models are used for denoising, then noise removal accuracy improves, but computational complexity and processing time increase

Engineering Contradiction:
Improvenoise in medical imageVSAvoidcomputational complexity of denoising system
Core Design Contradiction:
Object-affected harmful factorsVSDevice complexity

Solution Approach 1:

The deep learning model is trained in advance on large datasets of noisy and clean medical images. This preliminary training phase allows the model to learn noise patterns and removal strategies beforehand, so that during actual denoising operations, the model can quickly process images without requiring complex real-time computations.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system dynamically adjusts processing based on the specific characteristics of each input image. The model can adapt its denoising strength and approach based on the detected noise levels and image features, optimizing computational resources for each case rather than applying fixed complex processing to all images.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3889881B1Method, device and system for generating a denoised medical image
Publication Date: 2024.05.08 SIEMENS HEALTHINEERS AG
  • EP3889881B1 patent drawingFigure 1
  • EP3889881B1 patent drawingFigure 2
  • EP3889881B1 patent drawingFigure 3

AI summary

A method (200), device and system (100) for generating a denoised medical image is provided. In one aspect, the method (200) includes obtaining a medical image from a medical imaging device. Further, the method (200) includes extracting at least one category of medical information associated with the medical image. Additionally, the method (200) includes processing the medical image using at least one trained machine learning model to obtain pixel information associated with the medical image. The method (200) also includes retrieving at least one pre-defined pixel matrix from a database (107), wherein the pre-defined pixel matrix is identified based on the at least one category of medical information and the pixel information associated with the medical image. Furthermore, the method (200) includes correcting the pixel information associated with the medical image using the at least one pre-defined pixel matrix so as to generate a denoised medical image.