ImprovingF-FDG PET Quantification Through a Spatial Normalization Method.

in Journal of nuclear medicine : official publication, Society of Nuclear Medicine by Daewoon Kim, Seung Kwan Kang, Seong A Shin, Hongyoon Choi, Jae Sung Lee

TLDR

  • This study developed a new way to analyze brain images using a deep learning model. The model was trained using a smaller number of datasets than traditional methods, making it more efficient and potentially useful for a wider range of applications. The model was tested on two sets of brain images and found to be better at analyzing the images than traditional methods. The study highlights the potential of transfer learning in medical imaging analysis.

Abstract

Quantification ofF-FDG PET images is useful for accurate diagnosis and evaluation of various brain diseases, including brain tumors, epilepsy, dementia, and Parkinson disease. However, accurate quantification ofF-FDG PET images requires matched 3-dimensional TMRI scans of the same individuals to provide detailed information on brain anatomy. In this paper, we propose a transfer learning approach to adapt a pretrained deep neural network model from amyloid PET to spatially normalizeF-FDG PET images without the need for 3-dimensional MRI.The proposed method is based on a deep learning model for automatic spatial normalization ofF-FDG brain PET images, which was developed by fine-tuning a pretrained model for amyloid PET using only 103F-FDG PET and MR images. After training, the algorithm was tested on 65 internal and 78 external test sets. All TMR images with a 1-mm isotropic voxel size were processed with FreeSurfer software to provide cortical segmentation maps used to extract a ground-truth regional SUV ratio using cerebellar gray matter as a reference region. These values were compared with those from spatial normalization-based quantification methods using the proposed method and statistical parametric mapping software.The proposed method showed superior spatial normalization compared with statistical parametric mapping, as evidenced by increased normalized mutual information and better size and shape matching in PET images. Quantitative evaluation revealed a consistently higher SUV ratio correlation and intraclass correlation coefficients for the proposed method across various brain regions in both internal and external datasets. The remarkably good correlation and intraclass correlation coefficient values of the proposed method for the external dataset are noteworthy, considering the dataset's different ethnic distribution and the use of different PET scanners and image reconstruction algorithms.This study successfully applied transfer learning to a deep neural network forF-FDG PET spatial normalization, demonstrating its resource efficiency and improved performance. This highlights the efficacy of transfer learning, which requires a smaller number of datasets than does the original network training, thus increasing the potential for broader use of deep learning-based brain PET spatial normalization techniques for various clinical and research radiotracers.

Overview

  • The study aims to develop a transfer learning approach for automatic spatial normalization ofF-FDG brain PET images without the need for 3-dimensional MRI. The proposed method is based on a deep learning model for automatic spatial normalization ofF-FDG brain PET images, which was developed by fine-tuning a pretrained model for amyloid PET using only 103F-FDG PET and MR images. The algorithm was tested on 65 internal and 78 external test sets. The primary objective of the study is to demonstrate the resource efficiency and improved performance of the transfer learning approach compared to traditional spatial normalization methods.

Comparative Analysis & Findings

  • The proposed method showed superior spatial normalization compared with statistical parametric mapping, as evidenced by increased normalized mutual information and better size and shape matching in PET images. Quantitative evaluation revealed a consistently higher SUV ratio correlation and intraclass correlation coefficients for the proposed method across various brain regions in both internal and external datasets. The remarkably good correlation and intraclass correlation coefficient values of the proposed method for the external dataset are noteworthy, considering the dataset's different ethnic distribution and the use of different PET scanners and image reconstruction algorithms.

Implications and Future Directions

  • The study highlights the efficacy of transfer learning, which requires a smaller number of datasets than does the original network training, thus increasing the potential for broader use of deep learning-based brain PET spatial normalization techniques for various clinical and research radiotracers. Future research directions could include the application of the proposed method to other brain PET radiotracers and the development of a deep learning-based spatial normalization method for other types of medical imaging data.
Read Full Article