Deep Learning-Driven Estimation of Centiloid Scales from Amyloid PET Images with 11C-PiB and 18F-Labeled Tracers in Alzheimer's Disease

Tensho Yamao; Kenta Miwa; Yuta Kaneko; Noriyuki Takahashi; Noriaki Miyaji; Koki Hasegawa; Kei Wagatsuma; Yuto Kamitaka; Hiroshi Ito; Hiroshi Matsuda

doi:10.3390/brainsci14040406

Deep Learning-Driven Estimation of Centiloid Scales from Amyloid PET Images with ¹¹C-PiB and ¹⁸F-Labeled Tracers in Alzheimer's Disease

Brain Sci. 2024 Apr 21;14(4):406. doi: 10.3390/brainsci14040406.

Authors

Affiliations

¹ Department of Radiological Sciences, School of Health Science, Fukushima Medical University, Fukushima 960-8516, Japan.
² Department of Radiology, Fukushima Medical University Hospital, Fukushima 960-1295, Japan.
³ School of Allied Health Sciences, Kitasato University, Tokyo 252-0373, Japan.
⁴ Research Team for Neuroimaging, Tokyo Metropolitan Institute for Geriatrics and Gerontology, Tokyo 173-0015, Japan.
⁵ Department of Radiology and Nuclear Medicine, Fukushima Medical University, Fukushima 960-1295, Japan.
⁶ Department of Biofunctional Imaging, Fukushima Medical University, Fukushima 960-1295, Japan.

Abstract

Background: Standard methods for deriving Centiloid scales from amyloid PET images are time-consuming and require considerable expert knowledge. We aimed to develop a deep learning method of automating Centiloid scale calculations from amyloid PET images with ¹¹C-Pittsburgh Compound-B (PiB) tracer and assess its applicability to ¹⁸F-labeled tracers without retraining.

Methods: We trained models on 231 ¹¹C-PiB amyloid PET images using a 50-layer 3D ResNet architecture. The models predicted the Centiloid scale, and accuracy was assessed using mean absolute error (MAE), linear regression analysis, and Bland-Altman plots.

Results: The MAEs for Alzheimer's disease (AD) and young controls (YC) were 8.54 and 2.61, respectively, using ¹¹C-PiB, and 8.66 and 3.56, respectively, using ¹⁸F-NAV4694. The MAEs for AD and YC were higher with ¹⁸F-florbetaben (39.8 and 7.13, respectively) and ¹⁸F-florbetapir (40.5 and 12.4, respectively), and the error rate was moderate for ¹⁸F-flutemetamol (21.3 and 4.03, respectively). Linear regression yielded a slope of 1.00, intercept of 1.26, and R² of 0.956, with a mean bias of -1.31 in the Centiloid scale prediction.

Conclusions: We propose a deep learning means of directly predicting the Centiloid scale from amyloid PET images in a native space. Transferring the model trained on ¹¹C-PiB directly to ¹⁸F-NAV4694 without retraining was feasible.

Keywords: amyloid PET; centiloid scale; deep learning.

Grants and funding

This work was supported in part by a KAKENHI Grant-in-Aid for Young Scientists (No. 21K18097) from the Ministry of Education, Culture, Sports, Science and Technology (MEXT), the Japanese Government.