Enriching Neural Models with Targeted Features for Dementia Detection

Abstract

Alzheimer's disease (AD) is an irreversible brain disease that can dramatically reduce quality of life, most commonly manifesting in older adults and eventually leading to the need for full-time care. Early detection is fundamental to slowing its progression; however, diagnosis can be expensive, time-consuming, and invasive. In this work we develop a neural model based on a CNN-LSTM architecture that learns to detect AD and related dementias using targeted and implicitly-learned features from conversational transcripts. Our approach establishes the new state of the art on the De-mentiaBank dataset, achieving an F1 score of 0.929 when classifying participants into AD and control groups.

Explainable CNN-attention Networks (C-Attention Network) for Automated Detection of Alzheimer’s Disease

“…We evaluate the proposed C-Attention Network architectures on the DementiaBank dataset and compare the performances of these architectures with each other as well as some recently published results [11, 12].…”

“…We used stochastic gradient descent + momentum (SGD + Momentum) [42] as the optimizer for training. Since the cookie theft sub-dataset is unbalanced we added a class weight correction by increasing the penalty for misclassifying the less frequent class during model training to reduce the affect of data bias, as in [12]. The class weight correction ratio used in this paper is 7 : 3.…”

“…C-LSTM: The C-LSTM architecture consists of a CNN followed by an LSTM layer. PoS and word embedding features are used [12].…”

“…[15] used regression models to achieve an accuracy of 81% on classifying between AD patients and healthy controls. Some researchers have combined latent features obtained via language embeddings like word2vec [16] ‘GloVe’ [17] sentence2vector [18] along with hand-crafted features in a [19, 11, 12] hybrid approach using a CNN and RNN architecture to achieve accuracy around 89%.…”

“…A Convolutional Neural Network (CNN) and Long Short Term Memory Network (LSTM) model (CNN-LSTM) was proposed in [11, 12], by using part-of-speech (PoS) tags to get accuracy up to 89.7% on classification for AD disease on DementiaBank cookie sub-corpus. Deep neural networks and deep language models were combined [13] to classifying the disease.…”