Segmentation of white matter hyperintensities on 18F-FDG PET/CT images with a generative adversarial network
Kyeong Taek Oh1, Dongwoo Kim2, Byoung Seok Ye3
1Department of Medical Engineering, Yonsei University College of Medicine, Seoul, Republic of Korea.
Purpose:
White matter hyperintensities (WMH) are typically segmented using MRI because WMH are hardly visible on 18F-FDG PET/CT. This retrospective study was conducted to segment WMH and estimate their volumes from 18F-FDG PET with a generative adversarial network (WhyperGAN).
Methods:
We selected patients whose interval between MRI and FDG PET/CT scans was within 3 months, from January 2017 to December 2018, and classified them into mild, moderate, and severe groups by following the semiquantitative rating method of Fazekas. For each group, 50 patients were selected, and of them, we randomly selected 35 patients for training and 15 for testing. WMH were automatically segmented from FLAIR MRI with manual adjustment. Patches of WMH were extracted from 18F-FDG PET and segmented MRI. WhyperGAN was compared with H-DenseUnet, a deep learning method widely used for segmentation tasks, for segmentation performance based on the dice similarity coefficient (DSC), recall, and average volume differences (AVD). For volume estimation, the predicted WMH volumes from PET were compared with ground truth volumes.
Results:
The DSC values were associated with WMH volumes on MRI. For volumes >60 mL, the DSC values were 0.751 for WhyperGAN and 0.564 for H-DenseUnet. For volumes ≤60 mL, the DSC values rapidly decreased as the volume decreased (0.362 for WhyperGAN vs. 0.237 for H-DenseUnet). For recall, WhyperGAN achieved the highest value in the severe group (0.579 for WhyperGAN vs. 0.509 for H-DenseUnet). For AVD, WhyperGAN achieved the lowest score in the severe group (0.494 for WhyperGAN vs. 0.941 for H-DenseUnet). For the WMH volume estimation, WhyperGAN performed better than H-DenseUnet and yielded excellent correlation coefficients (r = 0.998, 0.983, and 0.908 in the severe, moderate, and mild group).
Conclusions:
Although limited by visual analysis, the WhyperGAN based can be used to automatically segment and estimate volumes of WMH from 18F-FDG PET/CT. This would increase the usefulness of 18F-FDG PET/CT for the evaluation of WMH in patients with cognitive impairment.
Insights
This study introduces WhyperGAN, a generative adversarial network for segmenting white matter hyperintensities (WMH) on 18F-FDG PET/CT scans. The method accurately estimates WMH volumes, improving PET/CT utility for cognitive impairment evaluation.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Neuroscience
Background:
- White matter hyperintensities (WMH) are crucial imaging biomarkers, typically assessed with MRI.
- WMH are often difficult to visualize and quantify using 18F-FDG PET/CT alone.
- Accurate WMH volume estimation is vital for understanding cognitive impairment progression.
Purpose of the Study:
- To develop and evaluate a generative adversarial network (WhyperGAN) for segmenting and estimating WMH volumes from 18F-FDG PET/CT.
- To compare the performance of WhyperGAN against a deep learning benchmark (H-DenseUnet).
- To assess the potential of WhyperGAN in enhancing the clinical utility of 18F-FDG PET/CT for WMH evaluation.
Main Methods:
- Retrospective study utilizing MRI and 18F-FDG PET/CT scans from patients within a 3-month interval.
- WMH segmentation performed using FLAIR MRI with manual adjustments, followed by patch extraction for WhyperGAN training and testing.
- WhyperGAN performance evaluated against H-DenseUnet using Dice Similarity Coefficient (DSC), recall, and Average Volume Difference (AVD).
Main Results:
- WhyperGAN demonstrated superior segmentation performance (DSC) compared to H-DenseUnet, particularly for larger WMH volumes (>60 mL).
- WhyperGAN achieved higher recall and lower average volume differences, especially in the severe WMH group.
- Volume estimation using WhyperGAN showed excellent correlation coefficients with ground truth volumes across all severity groups.
Conclusions:
- WhyperGAN enables automatic segmentation and volume estimation of WMH directly from 18F-FDG PET/CT data.
- This approach significantly enhances the diagnostic value of 18F-FDG PET/CT for assessing WMH.
- The method holds promise for improving the evaluation of WMH in patients experiencing cognitive impairment.
