Deep Learning CNN for PET Image Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Positron emission tomography (PET) images often suffer from high noise levels and poor spatial resolution due to limitations in injected doses and scan durations, leading to misdiagnoses and incorrect treatment decisions, and existing denoising methods are not robust to variations in noise levels.
Innovation Solution
A deep learning (DL) convolutional neural network (CNN) approach is trained to be robust to varying noise levels, using a 2.5D orthogonal training and denoising method, feature-oriented training to preserve small features, and multi-modality training with other medical images for partial volume correction, to produce consistently high-quality PET images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If traditional denoising methods are applied to PET images, then noise levels are reduced, but image quality becomes non-uniform and artifacts are introduced
Solution Approach 1:
The patent applies parameter changes by transforming the input PET image into a transformed domain (e.g., frequency domain or other representations) and applying denoising operations in this transformed domain. This allows the denoising method to operate with different parameters that preserve image quality uniformity while reducing noise, avoiding the artifacts introduced by traditional spatial-domain denoising methods.
2Object-affected harmful factors
If denoising methods are applied to PET images, then noise is reduced, but spatial resolution deteriorates
Solution Approach 1:
The patent segments the denoising process into multiple stages or components that can independently address different aspects of the image. By dividing the denoising task into separate operations (e.g., noise reduction in transformed domain, then reconstruction and refinement in spatial domain), the method achieves both noise reduction and preservation of spatial resolution, avoiding the trade-off inherent in traditional single-stage denoising methods.
3Reliability
If deep learning denoising is applied to PET images with varying noise levels, then consistent image quality is achieved, but computational complexity increases
Solution Approach 1:
The patent employs parameter changes by using transformations that convert the PET image into a different domain where denoising can be performed more efficiently. This transformed-domain approach, combined with deep learning, reduces the computational complexity required to achieve consistent image quality across varying noise levels, as the transformation enables more effective noise characterization and removal with fewer computational resources.
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
The DL-CNN approach provides consistent noise texture, improved robustness, and better preservation of small-lesion contrast, reducing inter- and intra-patient variation, and improving diagnostic outcomes by adapting to different noise levels without requiring adjustable parameters.
Implementation Method 1
a first set of multiple two-dimensional (2D) convolution operations are applied in parallel to a first set of respective 2D slices in a PET image in first orientations (e.g., sagittal, coronal, and transaxial planes), and a second set of multiple 2D convolution operations are applied in parallel to a second set of respective 2D slices in a PET image in second orientations
Implementation Method 2
PET imaging systems use detectors positioned around the patient to detect coincidence pairs of gamma rays
Implementation Method 3
To reconstruct the spatio-temporal distribution of the tracer via tomographic reconstruction principles
Implementation Method 4
The tracer emits positrons, resulting in an annihilation event occurs when the positron collides with an electron that produces two gamma rays (at 511 keV) traveling at substantially 180 degrees apart
Data Source
AI summary
A deep learning (DL) convolution neural network (CNN) reduces noise in positron emission tomography (PET) images, and is trained using a range of noise levels for the low-quality images having high noise in the training dataset to produce uniform high-quality images having low noise, independently of the noise level of the input image. The DL-CNN network can be implemented by slicing a three-dimensional (3D) PET image into 2D slices along transaxial, coronal, and sagittal planes, using three separate 2D CNN networks for each respective plane, and averaging the outputs from these three separate 2D CNN networks. Feature-oriented training can be implemented by segmenting each training image into lesion and background regions, and, in the loss function, applying greater weights to voxels in the lesion region. Other medical images (e.g. MRI and CT) can be used to enhance resolution of the PET images and provide partial volume corrections.