Automated classification of Alzheimer's disease and mild cognitive impairment using a single MRI and deep neural networks

Abstract

We built and validated a deep learning algorithm predicting the individual diagnosis of Alzheimer's disease (AD) and mild cognitive impairment who will convert to AD (c-MCI) based on a single cross-sectional brain structural MRI scan. Convolutional neural networks (CNNs) were applied on 3D T1-weighted images from ADNI and subjects recruited at our Institute (407 healthy controls [HC], 418 AD, 280 c-MCI, 533 stable MCI [s-MCI]). CNN performance was tested in distinguishing AD, c-MCI and s-MCI. High levels of accuracy were achieved in all the classifications, with the highest rates achieved in the AD vs HC classification tests using both the ADNI dataset only (99%) and the combined ADNI + non-ADNI dataset (98%). CNNs discriminated c-MCI from s-MCI patients with an accuracy up to 75% and no difference between ADNI and non-ADNI images. CNNs provide a powerful tool for the automatic individual patient diagnosis along the AD continuum. Our method performed well without any prior feature engineering and regardless the variability of imaging protocols and scanners, demonstrating that it is exploitable by not-trained operators and likely to be generalizable to unseen patient data. CNNs may accelerate the adoption of structural MRI in routine practice to help assessment and management of patients.

Transfer Learning for Predicting Conversion from Mild Cognitive Impairment to Dementia of Alzheimer’s Type based on 3D-Convolutional Neural Network

“…With the development of computational methods such as machine learning and deep learning, the utility of biomarker-based diagnosis for the classification and prediction of disease is becoming recognized. Various methods have been proposed to tackle the problem of predicting MCI patients who convert to DAT (MCI-Converters or MCI-C) vs. those who do not (MCI-Non-Converters or MCI-NC) [4,7,21,38]. For example, using Random Forest with weak hierarchical lasso feature selection, Li et al [21] achieved 74.8% classification accuracy with 161 MCI-NC and 132 MCI-C sMRI scans.…”

“…Similarly, Suk et al [38] had 74.8% classification accuracy in classifying 226 MCI-NC and 167 MCI-C patients by using 2D-CNN based on 93 regions of interest (ROIs) as features. 1 Lastly, Basaia and colleagues [4] showed 74.9% classification accuracy in classifying 533 MCI-NC and 280 MCI-C patients by using 3D-Convoluational Neural Network (CNN) based on gray matter tissue probability maps.…”

“…When designing a study, it is important to assign a portion of the whole data set in a random manner to be included in the independent test set [19]. For example, Basaia and colleagues [4] assigned 10% of the whole dataset as test. While the most effective splitting ratio of the training, validation, and test sets is still under discussion, the ratio of 60:20:20 or 70:15:15 is traditionally accepted for small data set.…”

“…chosen feature is the most informative but may miss important information contained in the raw data. For example, studies that selected gray matter as the feature for model training [4] did not consider CSF or white matter that also play a role in DAT [17,20,40]. Additionally, Cheng et al [7] manually selected "useful" samples using DTFS and DTSS-based features.…”

“…Previous research suggests that the classification task of Cognitively Normal Control (NC) vs. DAT is similar to the classification task of MCI-NC vs. MCI-C [8,9,43]. In previous studies, the classification task of NC vs. DAT has been used to pre-train the model [4,7]. Therefore, we utilize a classification task of NC vs. DAT as the source task for transfer learning to our target task model.…”