Related Experiment Video
Updated: Nov 14, 2025
PET and MRI Guided Irradiation of a Glioblastoma Rat Model Using a Micro-irradiator
Published on: December 28, 2017
Micro-Networks for Robust MR-Guided Low Count PET Imaging
Casper O da Costa-Luis1, Andrew J Reader1
1Department of Biomedical EngineeringSchool of Biomedical Engineering and Imaging Sciences, St. Thomas' HospitalKing's College LondonLondonSE1 7EHU.K.
Abstract:
Noise suppression is particularly important in low count positron emission tomography (PET) imaging. Post-smoothing (PS) and regularization methods which aim to reduce noise also tend to reduce resolution and introduce bias. Alternatively, anatomical information from another modality such as magnetic resonance (MR) imaging can be used to improve image quality. Convolutional neural networks (CNNs) are particularly well suited to such joint image processing, but usually require large amounts of training data and have mostly been applied outside the field of medical imaging or focus on classification and segmentation, leaving PET image quality improvement relatively understudied. This article proposes the use of a relatively low-complexity CNN (micro-net) as a post-reconstruction MR-guided image processing step to reduce noise and reconstruction artefacts while also improving resolution in low count PET scans. The CNN is designed to be fully 3-D, robust to very limited amounts of training data, and to accept multiple inputs (including competitive denoising methods). Application of the proposed CNN on simulated low (30 M) count data (trained to produce standard (300 M) count reconstructions) results in a 36% lower normalized root mean squared error (NRMSE, calculated over ten realizations against the ground truth) compared to maximum-likelihood expectation maximization (MLEM) used in clinical practice. In contrast, a decrease of only 25% in NRMSE is obtained when an optimized (using knowledge of the ground truth) PS is performed. A 26% NRMSE decrease is obtained with both RM and optimized PS. Similar improvement is also observed for low count real patient datasets. Overfitting to training data is demonstrated to occur as the network size is increased. In an extreme case, a U-net (which produces better predictions for training data) is shown to completely fail on test data due to overfitting to this case of very limited training data. Meanwhile, the resultant images from the proposed CNN (which has low training data requirements) have lower noise, reduced ringing, and partial volume effects, as well as sharper edges and improved resolution compared to conventional MLEM.
Insights
This study introduces a novel convolutional neural network (CNN) for enhanced Positron Emission Tomography (PET) imaging. The micro-net significantly reduces noise and improves resolution in low-count PET scans, outperforming traditional methods.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Image Processing
Background:
- Low count Positron Emission Tomography (PET) imaging requires effective noise suppression.
- Traditional methods like post-smoothing (PS) and regularization reduce noise but also decrease resolution and introduce bias.
- Anatomical information from modalities like Magnetic Resonance (MR) imaging can enhance PET image quality.
Purpose of the Study:
- To propose a low-complexity, 3-D convolutional neural network (CNN) for post-reconstruction MR-guided image processing in low count PET scans.
- To reduce noise and reconstruction artifacts while simultaneously improving resolution.
- To develop a CNN robust to limited training data and capable of multi-input processing.
Main Methods:
- A fully 3-D convolutional neural network (CNN) with low complexity (micro-net) was designed for MR-guided PET image post-processing.
- The CNN was trained using limited data to process low count (30 M) PET scans, aiming for quality comparable to standard (300 M) count reconstructions.
- The CNN was evaluated against Maximum-Likelihood Expectation Maximization (MLEM) and optimized post-smoothing (PS) methods using simulated and real patient data.
Main Results:
- The proposed CNN achieved a 36% lower Normalized Root Mean Squared Error (NRMSE) compared to MLEM on simulated low count data.
- Optimized PS and MR (RM) with optimized PS yielded only 25% and 26% NRMSE reduction, respectively.
- The CNN demonstrated robustness against overfitting, unlike larger networks (U-net), and produced images with reduced noise, artifacts, and improved resolution on real patient data.
Conclusions:
- A low-complexity, 3-D CNN (micro-net) effectively improves image quality in low count PET scans by reducing noise and enhancing resolution.
- The proposed MR-guided CNN approach significantly outperforms conventional MLEM and PS methods, especially with limited training data.
- This method offers a promising solution for enhancing diagnostic accuracy in low count PET imaging without requiring extensive datasets.
