Alzheimer's Disease Classification With a Cascade Neural Network

Zeng You, Runhao Zeng, Xiaoyong Lan, Huixia Ren, Zhiyang You, Xue Shi, Shipeng Zhao, Yi Guo, Xin Jiang, Xiping HuFrontiers in Public Health 2020 · DOI

Abstract

Early detection of Alzheimer's Disease (AD) is vital but challenging, particularly in distinguishing Mild Cognitive Impairment (MCI) from healthy aging or full-blown AD. This paper introduces a two-stage cascade neural network that leverages both gait and EEG data. By first filtering healthy controls using gait analysis and then classifying patients using raw EEG signals, the model achieves a 91.07% accuracy in three-way classification, significantly outperforming single-modality methods.

Introduction

The global rise in Alzheimer's cases necessitates efficient screening tools. While gait data is easy to acquire via sensors like Kinect, it often lacks the sensitivity to distinguish MCI from AD. Conversely, EEG provides high precision but is resource-intensive to perform on every screening candidate. Furthermore, traditional EEG models often lose spatial-temporal connectivity by converting signals to the frequency domain. This study proposes a multimodal cascade approach to optimize both efficiency and accuracy.

Methods: The Cascade Architecture

The proposed system operates as a clinical triage pipeline, mimicking a standard medical workflow to maximize diagnostic speed and precision.

The proposed cascade neural network architecture combining gait and EEG analysis.

Step 1: Gait-Based Screening

The first stage uses an Attention-based Spatial Temporal Graph Convolutional Network (AST-GCN) to analyze skeleton sequences. The human body is modeled as a graph where joints are nodes and bones are edges.

Spatial-temporal graph representation and convolution strategy for gait analysis.

An "hourglass" attention module identifies the most relevant frames and joints in the walking sequence. This stage effectively separates Healthy Controls (HC) from the patient group (MCI/AD) with an accuracy of 93.09%.

Step 2: EEG-Based Patient Classification

Subjects identified as patients in Step 1 undergo an EEG. A Spatial Temporal Convolutional Network (ST-CNN) processes raw EEG data, preserving the temporal synchronization across different brain regions. This avoids the information loss associated with frequency-domain transforms and allows the model to capture implicit connectivity features.

Results and Discussion

The model was evaluated on a dataset of 87 subjects (35 HC, 35 MCI, 17 AD). The cascade approach achieved a three-way classification accuracy of 91.07%, a massive leap over the 68.18% achieved by single-modality baselines.

Joint Importance in Gait

A key finding was that the lower body and right upper limb are the most significant indicators for early AD. Researchers suggest this aligns with clinical observations that the left hemisphere (which controls the right side of the body in right-handed individuals) is often more sensitive to early neurodegeneration.

Efficiency

The cascade design is computationally efficient, with an inference speed of 3.99 ms and a reduced parameter count (4.72M) compared to non-cascaded models. By using gait as a primary filter, the system reduces the clinical burden of performing EEG tests on healthy individuals.

Conclusion

By combining the accessibility of gait sensors with the precision of raw EEG analysis, this cascade neural network provides a robust framework for Alzheimer's diagnosis. While the study is limited by a small sample size of AD patients, the results demonstrate the potential of multimodal deep learning in clinical triage.

Summary updated

Abstract

Classification of Alzheimer's Disease (AD) has been becoming a hot issue along with the rapidly increasing number of patients. This task remains tremendously challenging due to the limited data and the difficulties in detecting mild cognitive impairment (MCI). Existing methods use gait [or EEG (electroencephalogram)] data only to tackle this task. Although the gait data acquisition procedure is cheap and simple, the methods relying on gait data often fail to detect the slight difference between MCI and AD. The methods that use EEG data can detect the difference more precisely, but collecting EEG data from both HC (health controls) and patients is very time-consuming. More critically, these methods often convert EEG records into the frequency domain and thus inevitably lose the spatial and temporal information, which is essential to capture the connectivity and synchronization among different brain regions. This paper proposes a cascade neural network with two steps to achieve a faster and more accurate AD classification by exploiting gait and EEG data simultaneously. In the first step, we propose attention-based spatial temporal graph convolutional networks to extract the features from the skeleton sequences (i.e., gait) captured by Kinect (a commonly used sensor) to distinguish between HC and patients. In the second step, we propose spatial temporal convolutional networks to fully exploit the spatial and temporal information of EEG data and classify the patients into MCI or AD eventually. We collect gait and EEG data from 35 cognitively health controls, 35 MCI, and 17 AD patients to evaluate our proposed method. Experimental results show that our method significantly outperforms other AD diagnosis methods (91.07 vs. 68.18%) in the three-way AD classification task (HC, MCI, and AD). Moreover, we empirically found that the lower body and right upper limb are more important for the early diagnosis of AD than other body parts. We believe this interesting finding can be helpful for clinical researches.
BibTeX
@inproceedings{you2020alzheimer,
  author = {Zeng You and Runhao Zeng and Xiaoyong Lan and Huixia Ren and Zhiyang You and Xue Shi and Shipeng Zhao and Yi Guo and Xin Jiang and Xiping Hu},
  title = {{Alzheimer's Disease Classification With a Cascade Neural Network}},
  booktitle = {Frontiers in Public Health},
  year = {2020},
  doi = {10.3389/fpubh.2020.584387},
}

Loading related content...

Figure