On Automatic Diagnosis of Alzheimer’s Disease Based on Spontaneous Speech Analysis and Emotional Temperature

Lopez-de-Ipiña et al. (2015) found positive and promising results in differentiating different stages of Alzheimer's disease using emotional temperature and spontaneous speech analysis, and concluded there is a decrease in fluency in speech of people with Alzheimer's disease.

“The test showed positive as well as promising results in differentiating different stages of AD. It concluded that there is a decrease in fluency in speech of people suffering from AD [130].”

Speech or voice analysis along with analysis of emotional temperature can be used as a diagnostic tool for detecting Alzheimer's disease in its preliminary stage.

“Therefore, speech or voice analysis along with analysis of emotional temperature of patients suffering from AD is very important and can be used as a diagnostic tool for detecting Alzheimer's in its preliminary stage [130].”

The AZTIAHORE dataset contains the youngest healthy group (20-90 years old) and a typical AD group (68-98 years old), introducing potential biases due to this imbalance.

“An exception is the dataset AZTIAHORE [79, 80], which contains the youngest healthy group (20–90 years old) and a typical AD group (68–98 years old), introducing potential biases due to this imbalance.”

Papers lacking health assessment information do not specify their criteria for group assignment.

“Therefore, papers lacking that information do not specify their criteria for group assignment [60, 61, 75, 79, 80, 82–85].”

Lopez-de Ipiña et al. describe a dataset where 20% of HC speech data but 80% of AD speech data is removed during pre-processing.

“For instance, Lopez-de Ipiña et al. [79, 80] describe a dataset where 20% of the HC speech data, but 80% of the AD speech data, is removed during pre-processing.”

Lopez-de Ipiña et al. reported accuracy between 60% and 93.79% using only ad hoc acoustic features.

“Lopez-de Ipiña et al. [80] reported an accuracy that varied between 60% and 93.79% using only acoustic features that were generated ad hoc.”

Very few reviewed papers report use of artificial neural networks, recurrent neural networks, multi-layer perceptron, or convolutional neural networks.

“As regards choice of machine learning methods, very few papers report the use of artificial neural networks [91, 95], recurrent neural networks [82], multi-layer perceptron [44, 67, 79, 80, 86], or convolutional neural networks [60, 85].”