Differential Diagnosis of Frontotemporal Dementia, Alzheimer's Disease, and Normal Aging Using a Multi-Scale Multi-Type Feature Generative Adversarial Deep Neural Network on Structural Magnetic Resonance Images

Abstract

Methods: Alzheimer's disease and Frontotemporal dementia are the first and third most common forms of dementia. Due to their similar clinical symptoms, they are easily misdiagnosed as each other even with sophisticated clinical guidelines. For disease-specific intervention and treatment, it is essential to develop a computer-aided system to improve the accuracy of their differential diagnosis. Recent advances in deep learning have delivered some of the best performance for medical image recognition tasks. However, its application to the differential diagnosis of AD and FTD pathology has not been explored. Approach: In this study, we proposed a novel deep learning based framework to distinguish between brain images of normal aging individuals and subjects with AD and FTD. Specifically, we combined the multi-scale and multi-type MRI-base image features with Generative Adversarial Network data augmentation technique to improve the differential diagnosis accuracy. Results: Each of the multi-scale, multitype, and data augmentation methods improved the ability for differential diagnosis for both AD and FTD. A 10-fold cross validation experiment performed on a large sample of 1,954 images using the proposed framework achieved a high overall accuracy of 88.28%. Conclusions: The salient contributions of this study are three-fold: (1) our experiments demonstrate that the combination of multiple structural features extracted at different scales with our proposed deep neural network yields superior performance than individual features; (2) we show that the use of Generative Adversarial Network for data augmentation could further improve the discriminant ability of the network regarding challenging tasks such as differentiating dementia sub-types; (3) and finally, we show that ensemble classifier strategy could make the network more robust and stable.

Keywords: Alzheimer's disease; differential diagnosis; frontotemporal dementia (FTD); generative adversarial network; magnetic resonance imaging.

Figure 1

Multi-scale and multi-type feature Deep Neural Network. The input feature dimension (number of patches) extracted from different scales is 1,488, 705, and 343 for volume of gray matter ROI and 527, 255, and 131 for the cortical thickness. The numbers of units in each layer representation are displayed on the top left.

Figure 2

Architecture of generative adversarial network. The numbers of units of Generator layers are 512 and 3,449(1,488+705+343+527+255+131), respectively.

Figure 3

Comparison about the distribution of the concatenated multi-level multi-type W-score feature set among different disease groups: (A) NC(ADNI) vs. NC(FTDNI); (B) NC vs. AD in ADNI database; (C) NC vs. FTD in FTDNI database; and (D) FTD vs. AD. (A) No statistical difference was shown when comparing the W-scores of the Healthy Control subjects between the ADNI and FTDNI, confirming no database-specific biases remained in the input w-score feature of the normative group. (B) Similar level of significant differences were shown when comparing the NC and AD subjects in the ADNI database, or (C) When comparing the NC and the FTD subjects in the FTDBI database, indicating similarity between the AD and FTD group. (D) When comparing the FTD and AD group alone, significant differences were observed in both the volume-based and thickness-based features, indicating discrepancy between these two types of Dementia subtypes which can be utilized to achieve potential differential diagnosis. Unpaired t-test were performed for each pair of the comparison, with multiple comparison corrected by setting false discovery rate (FDR) = 0.05.

Figure 4

Statistical analysis of the classification performance among different experiments. One-tailed pairwise t-tests were conducted to access the performance improvements. Multiple comparisons were corrected with False discovery rate FDR = 0.05. O: Significant improvement over PCV+SVM(Multi-type); X: significant improvement over MDNN (thickness); +: significant improvement over MDNN (volume); #: significant improvement over MMDNN (Multitype). (A) Overall accuracy, (B) NC sensitivity, (C) AD sensitivity, (D) FTD sensitivity.

Figure 5

Boxplot for classification accuracy of single classifiers (classifier 1–10 on x axis) and an ensemble of classifiers (E on the x axis). The stars in each box are the mean of accuracy and the red lines represent the median accuracy. (A) MDNN+Thickness, (B) MDNN+Volume, (C) MMDNN+Multitype, (D) GAN+Multitype.