3D Convolutional Neural Networks for Classification of Alzheimer’s and Parkinson’s Disease with T1-Weighted Brain MRI

Abstract

Parkinson's disease (PD) and Alzheimer's disease (AD) are progressive neurodegenerative disorders that affect millions of people worldwide. In this work, we propose a deep learning approach to classify these diseases based on 3D T1-weighted brain MRI. We analyzed several datasets including the Parkinson's Progression Markers Initiative (PPMI), an independent dataset from the University of Pennsylvania School of Medicine (UPenn), the Alzheimer's Disease Neuroimaging Initiative (ADNI), and the Open Access Series of Imaging Studies (OASIS) dataset. PPMI and ADNI were partitioned to train (70%), validate (20%), and test (10%) a 3D convolutional neural network (CNN) for PD and AD classification. The UPenn and OASIS datasets were used as independent test sets to evaluate the model performance during inference. We also implemented a random forest classifier as a baseline model by extracting key radiomics features from the same T1-weighted MRI scans. The proposed 3D CNN model was trained from scratch for the classification tasks. For AD classification, the 3D CNN model achieved an ROC-AUC of 0.878 on the ADNI test set and an average ROC-AUC of 0.789 on the OASIS dataset. For PD classification, the proposed 3D CNN model achieved an ROC-AUC of 0.667 on the PPMI test set and an average ROC-AUC of 0.743 on the UPenn dataset. We also found that model performance was largely maintained when using only 25% of the training dataset. The 3D CNN outperformed the random forest classifier for both the PD and AD tasks. The 3D CNN also generalized better on unseen MRI data from different imaging centers. Our results show that the proposed 3D CNN model was less prone to overfitting for AD than for PD classification. This approach shows promise for screening of PD and AD patients using only T1-weighted brain MRI, which is relatively widely available. This model with additional validation could also be used to help differentiate between challenging cases of AD and PD when they present with similarly subtle motor and non-motor symptoms.

The differential diagnosis value of radiomics-based machine learning in Parkinson’s disease: a systematic review and meta-analysis

“…Following the exclusion of 121 duplicate studies, 43 studies were screened based on their titles or abstracts. Ultimately, a total of 28 articles ( Cheng et al, 2019 ; Shinde et al, 2019 ; Wu et al, 2019 ; Xiao et al, 2019 ; Cao et al, 2020 , 2021 ; Liu et al, 2020 ; Pang et al, 2020 , 2022 ; Shu et al, 2020 ; Dhinagar et al, 2021 ; Hu et al, 2021 ; Li et al, 2021 , 2022 ; Ren et al, 2021 ; Shi et al, 2021 , 2022a , 2022b ; Sun et al, 2021 , 2022 ; Tupe-Waghmare et al, 2021 ; Zhang et al, 2021 ; Ben Bashat et al, 2022 ; Guan et al, 2022 ; Kang et al, 2022 ; Kim et al, 2022 ; Shiiba et al, 2022 ; Zhao et al, 2022 ) were deemed eligible and included in this meta-analysis.…”

“…The characteristics of the studies included in this research are shown in Table 1 and Supplementary Table 4 . The original 28 studies were published between 2019 and 2022, with 27 of them from Asia ( Cheng et al, 2019 ; Shinde et al, 2019 ; Wu et al, 2019 ; Xiao et al, 2019 ; Cao et al, 2020 , 2021 ; Liu et al, 2020 ; Pang et al, 2020 , 2022 ; Shu et al, 2020 ; Hu et al, 2021 ; Li et al, 2021 , 2022 ; Ren et al, 2021 ; Shi et al, 2021 , 2022a , 2022b ; Sun et al, 2021 , 2022 ; Tupe-Waghmare et al, 2021 ; Zhang et al, 2021 ; Ben Bashat et al, 2022 ; Guan et al, 2022 ; Kang et al, 2022 ; Kim et al, 2022 ; Shiiba et al, 2022 ; Zhao et al, 2022 ) and one from North America ( Dhinagar et al, 2021 ). The study comprised a total of 6,057 participants, with 3,422 patients diagnosed with PD, 1,983 healthy controls, and 652 cases of APS (476 with MSA and 176 with PSP).…”

“…The study comprised a total of 6,057 participants, with 3,422 patients diagnosed with PD, 1,983 healthy controls, and 652 cases of APS (476 with MSA and 176 with PSP). Among these studies, 22 focused on the diagnosis of PD ( Cheng et al, 2019 ; Shinde et al, 2019 ; Wu et al, 2019 ; Xiao et al, 2019 ; Cao et al, 2020 , 2021 ; Liu et al, 2020 ; Shu et al, 2020 ; Dhinagar et al, 2021 ; Li et al, 2021 , 2022 ; Ren et al, 2021 ; Shi et al, 2021 , 2022a , 2022b ; Sun et al, 2021 , 2022 ; Zhang et al, 2021 ; Ben Bashat et al, 2022 ; Guan et al, 2022 ; Kang et al, 2022 ; Shiiba et al, 2022 ), while six studies focused on the differential diagnosis of PD and APS ( Pang et al, 2020 , 2022 ; Hu et al, 2021 ; Tupe-Waghmare et al, 2021 ; Kim et al, 2022 ; Zhao et al, 2022 ). In addition, two studies addressed the differential diagnosis of PD with or without depression ( Li et al, 2021 ; Zhang et al, 2021 ), and one study fixated on the differential diagnosis of TD and PIGD ( Sun et al, 2021 ).…”

Evaluation of transfer learning methods for detecting Alzheimer’s disease with brain MRI

“…2 Structural magnetic resonance imaging (MRI) based on standard anatomical T1-weighted (T1-w) is routinely used and accessible worldwide and does not subject the patient to ionizing radiation. In recent times with an increase in the available training data, deep learning and machine learning methods have shown great promise for various neuroimaging tasks such as brain age prediction, 3 diagnostic classification 4 and disease subtyping and staging, 5 based on detecting profiles of brain atrophy with T1-w MRI. In 2019, Wen et al 6 published a comprehensive review of over 30 papers that used convolutional neural networks (CNNs) for AD classification from anatomical MRI.…”

FetalGAN: Automated Segmentation of Fetal Functional Brain MRI Using Deep Generative Adversarial Learning and Multi-Scale 3D U-Net

“…However, there are several limitations in using CNN-based approaches for segmentation ( Ronneberger et al, 2015 ; Xue et al, 2018 ; Li et al, 2019 ; Rutherford et al, 2021 ). Although U-Nets can use skip connections to combine both low- and high-level features, there is no guarantee of spatial consistency in the final segmentation map, especially at the boundaries ( Isola et al, 2017 ; Yang et al, 2018 ; Zhao et al, 2018 ; Dhinagar et al, 2021 ). To address this limitation, methods that consider spatial correlations among neighboring pixels such as conditional random field and other graph cut techniques are used as post-processing refinement ( Pereira et al, 2016b ; Nancy, 2019 ; Son et al, 2020 ).…”