A hybrid 3D CNN model for amyloid positivity prediction based on white matter structural integrity
Poster No:
1299
Submission Type:
Abstract Submission
Authors:
Pavithran Pattiam Giriprakash1, Zhengshi Yang1, Dietmar Cordes1, Andrew Bender1
Institutions:
1Cleveland Clinic, Las Vegas, NV
First Author:
Co-Author(s):
Introduction:
Accumulation of β-amyloid (Aβ) plaques, as detected by Positron Emission Tomography (PET), has been shown to be sufficient for the diagnosis of Alzheimer's disease (AD) (Jack et al., 2024). White matter degeneration is believed to precede cortical atrophy in early AD (Moody et al., 2022). Increased amyloid levels showed strong associations with age-related white matter alterations (Caballero et al., 2020). Also, higher education levels was proposed to have a compensatory effect on Aβ accumulation (Joannette et al., 2020). In this study, we propose a hybrid 3DCNN network for prediction of Aβ positivity using NODDI parameters and patient demographics.
Methods:
The study included 224 non-demented participants (129 cognitively normal (CN),95 mild cognitive impairment (MCI); age: 73.1±7.1; education: 16.1±2.5) from the Center for Neurodegeneration and Translational Neuroscience. PET images were processed to obtain standardized uptake value ratio (SUVR) values (Landau et al., 2012). A threshold of 1.11 for amyloid positivity yielded 130 Aβ negative and 94 Aβ positive cases (Figure 1(a)). The diffusion MRI (dMRI) acquisition parameters were: slice thickness=1.5 mm, TR=5218ms, TE=100ms, diffusion directions=79, b-values=0,500,1000, 2500s/mm2. Standard preprocessing was done for dMRI data in MRtrix3. Three parameters were derived from the NODDI (Zhang et al., 2012) model: neurite density index (NDI), free water fraction (FWF), and orientation dispersion index (ODI). The normalized NODDI images in the MNI space in addition to patient demographics (age, education and sex) served as inputs for the model (Figure 1(b)). These model parameters were chosen heuristically: learning rate=1e-3, epochs=50, batch size=8 and optimizer=Adam. Focal loss was used to account for class imbalance. Spatiotemporal feature extraction consisted of a (2+1)D convolution (Tran et al., 2017) layer with filter size=8 and kernel size=3. Average and max pooling layers were used for capturing both global and local information. A dense layer with 8 units was used to extract features from the patient demographics. Classification constituted a dense layer with sigmoid activation. Model generalizability was tested using stratified k-fold cross validation (k=3). Regularization was achieved using dropout layers (dropout rate=0.2) and unit norm constraint. Model evaluation included three methods: (i) 3DCNN features(F=F3DCNN); (ii) 3DCNN and demographics(F'=F3DCNN⊕FDEMO); (iii) using convolutional block attention module (F''=gCBAM(F')) proposed by (Woo et al., 2018). Model building and evaluation was done using TensorFlow v2.13.
Results:
The model combining CNN features and demographics (F') shows better performance compared to using only CNN features (F). The increased performance is evident in the higher AUC-ROC, accuracy and sensitivity values (figure 2). Model with attention mechanism (F'') shows the highest AUC-ROC values (0.82 in training, 0.68 in validation). Irrespective of the feature set used, the proposed model is comparatively better at identifying true amyloid positives (higher sensitivity than specificity in validation) as seen in figure 2(b)-2(d).
Conclusions:
The present findings show that white matter integrity and patient demographics are associated with amyloid deposition, a hallmark of AD. Using an attention mechanism in CNNs helps enhance the spatiotemporal learning, contributing to a model with good generalizability. However, further work is needed to evaluate the model's robustness across different thresholds of amyloid positivity as well as in an independent testing set. In addition, other patient demographics, such as Apolipoprotein E status could potentially improve model performance due to its association with Aβ burden. This work was supported by NIH grants, R01AG071566 and P20GM109025, and by a Cleveland Clinic Neurological Institute Emergent Clinician-Scientist Career Award to ARB.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s)
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Diffusion MRI Modeling and Analysis 1
Methods Development
Task-Independent and Resting-State Analysis
Keywords:
Computational Neuroscience
Data analysis
WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC
Other - Convolutional Neural Networks (CNNs)
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Are you Internal Review Board (IRB) certified? Please note: Failure to have IRB, if applicable will lead to automatic rejection of abstract.
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
2. Jack, C. R. (2024). Revised criteria for diagnosis and staging of Alzheimer’s disease: Alzheimer’s Association Workgroup. Alzheimer’s & Dementia: The Journal of the Alzheimer’s Association, 20(8), 5143–5169. https://doi.org/10.1002/alz.13859
3. Joannette, M. (2020). Education as a Moderator of the Relationship Between Episodic Memory and Amyloid Load in Normal Aging. The Journals of Gerontology. Series A, Biological Sciences and Medical Sciences, 75(10), 1820–1826. https://doi.org/10.1093/gerona/glz235
4. Landau, S. M. (2012). Amyloid deposition, hypometabolism, and longitudinal cognitive decline. Annals of Neurology, 72(4), 578–586. https://doi.org/10.1002/ana.23650
5. Moody, J. F. (2022). Associations between diffusion MRI microstructure and cerebrospinal fluid markers of Alzheimer’s disease pathology and neurodegeneration along the Alzheimer’s disease continuum. Alzheimer’s & Dementia (Amsterdam, Netherlands), 14(1), e12381. https://doi.org/10.1002/dad2.12381
6. Raghavan, S. (2021). Diffusion models reveal white matter microstructural changes with ageing, pathology and cognition. Brain Communications, 3(2), fcab106. https://doi.org/10.1093/braincomms/fcab106
7. Tran, D. (2017). A Closer Look at Spatiotemporal Convolutions for Action Recognition (Version 3). arXiv. https://doi.org/10.48550/ARXIV.1711.11248
8. Woo, S. (2018). CBAM: Convolutional Block Attention Module (Version 2). arXiv. https://doi.org/10.48550/ARXIV.1807.06521
9. Zhang, H. (2012). NODDI: Practical in vivo neurite orientation dispersion and density imaging of the human brain. NeuroImage, 61(4), 1000–1016. https://doi.org/10.1016/j.neuroimage.2012.03.072