Dilated Convolutional Neural Network for PET Image Denoising
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current PET imaging technologies face challenges in achieving high signal-to-noise ratio (SNR) without increasing radiation dose or scan time, leading to poor image quality due to physical degradation factors like photon detection limitations.
Innovation Solution
A dilated convolutional neural network system is employed for PET image denoising, which involves image normalization, encoding with increasing dilation rate, decoding with decreasing dilation rate, and synthesizing denoised output images, thereby enhancing image quality without sacrificing subject burden.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the scanner registers a large number of radioactive decay events to obtain high SNR images, then image quality and signal-to-noise ratio are improved, but radiation dose increases and scan time increases
Solution Approach 1:
The patent applies deep learning denoising techniques to convert the harmful noise present in low-count PET images into beneficial signal information. The neural network is trained to recognize and remove noise patterns while preserving true anatomical and functional information, effectively transforming the noisy low-count data into high-quality images without requiring additional radiation exposure or scan time
Solution Approach 2:
The patent changes the parameter of image processing by introducing deep learning-based denoising algorithms that operate on the reconstructed PET images. By applying learned transformations through convolutional neural networks, the system achieves high SNR image quality from low-count data, eliminating the need to increase radiation dose or scan time to improve image quality
2Measurement precision
If the scanner registers a large number of radioactive decay events to obtain high SNR images, then image quality and signal-to-noise ratio are improved, but scan time increases
Solution Approach 1:
The patent applies deep learning denoising techniques to convert the harmful noise present in low-count PET images into beneficial signal information. The neural network is trained to recognize and remove noise patterns while preserving true anatomical and functional information, effectively transforming the noisy low-count data into high-quality images without requiring additional radiation exposure or scan time
Solution Approach 2:
The patent performs preliminary denoising operations during the image reconstruction process using pre-trained deep learning models. By applying denoising transformations before final image display and analysis, the system achieves high-quality images from reduced-count data, eliminating the need for extended scan times
3Measurement precision
If deep learning techniques are applied for image denoising, then image quality is improved, but computational complexity increases
Solution Approach 1:
The patent segments the deep learning denoising process into distinct functional modules: an encoder that extracts features from input images, a bottleneck that processes the encoded representation, and a decoder that reconstructs the denoised image. This segmentation allows each module to be optimized independently and facilitates efficient implementation using standard deep learning frameworks and hardware accelerators
Solution Approach 2:
The patent uses pre-trained deep learning models that can be applied to new PET images without requiring retraining. By copying the learned denoising transformations from training data to clinical images, the system achieves high image quality while avoiding the computational burden of training models on each individual patient scan
Data Source
AI summary
A method for performing positron emission tomography (PET) image denoising using a dilated convolutional neural network system includes: obtaining, as an input to the dilated convolutional neural network system, a noisy image; performing image normalization to generate normalized image data corresponding to the noisy image; encoding the normalized image data using one or more convolutions in the dilated convolutional neural network, whereby a dilation rate is increased for each encoding convolution performed to generate encoded image data; decoding the encoded image data using one or more convolutions in the dilated convolutional neural network, whereby dilation rate is decreased for each decoding convolution performed to generate decoded image data; synthesizing the decoded image data to construct a denoised output image corresponding to the noisy image; and displaying the denoised output image on an image display device, the denoised output image having enhanced image quality compared to the noisy image.