Related Experiment Video
Updated: Feb 18, 2026
14:27
Identification of Disease-related Spatial Covariance Patterns using Neuroimaging Data
Published on: June 26, 2013
16.3K
Functional Brain Imaging Synthesis Based on Image Decomposition and Kernel Modeling: Application to Neurodegenerative
Francisco J Martinez-Murcia1, Juan M Górriz1, Javier Ramírez1
1Signal Processing and Biomedical Application, Department of Signal Theory, Networking and Communication, University of Granada, Granada, Spain.
Frontiers in Neuroinformatics
|November 30, 2017
Summary
This study introduces a novel brain image synthesis method for nuclear imaging. The generated synthetic images are suitable for standardizing computer-aided diagnosis (CAD) evaluation and augmenting machine learning datasets.
Area of Science:
- Medical Imaging
- Machine Learning
- Computer-Aided Diagnosis
Background:
- Neuroimaging and machine learning advancements drive computer-aided diagnosis (CAD).
- Limited access to proprietary datasets hinders direct comparison and development of CAD algorithms.
- Small sample sizes impede the creation of accurate machine learning models.
Purpose of the Study:
- To develop a brain image synthesis procedure for generating new image sets with characteristics similar to original ones.
- To address limitations in dataset size and accessibility for CAD development and evaluation.
- To create a method applicable to nuclear imaging modalities like PET and SPECT.
Main Methods:
- Principal Component Analysis (PCA) was applied to original datasets.
- A Probability Density Function (PDF) estimator modeled data distribution in the 'eigenbrain' space.
- New data points were generated in the eigenbrain space and projected back to image space.
Main Results:
- Synthetic images maintained inter-group differences present in original datasets.
- No significant differences were found between synthetic and real-world samples.
- Synthetic datasets demonstrated comparable performance and generalization capabilities to original datasets.
Conclusions:
- The proposed image synthesis method is suitable for standardizing CAD pipeline evaluation.
- Synthetic images can serve as data augmentation for machine learning, particularly deep learning.
- This technique can be valuable for training medical professionals.
