Towards an Automatic Speech-Based Diagnostic Test for Alzheimer’s Disease

Roozbeh Sadeghian,J David Schaffer,Stephen A Zahorian

doi:10.3389/fcomp.2021.624594

Abstract

Automatic Speech Recognition (ASR) is widely used in many applications and tools. Smartphones, video games, and cars are a few examples where people use ASR routinely and often daily. A less commonly used, but potentially very important arena for using ASR, is the health domain. For some people, the impact on life could be enormous. The goal of this work is to develop an easy-to-use, non-invasive, inexpensive speech-based diagnostic test for dementia that can easily be applied in a clinician’s office or even at home. While considerable work has been published along these lines, increasing dramatically recently, it is primarily of theoretical value and not yet practical to apply. A large gap exists between current scientific understanding, and the creation of a diagnostic test for dementia. The aim of this paper is to bridge this gap between theory and practice by engineering a practical test. Experimental evidence suggests that strong discrimination between subjects with a diagnosis of probable Alzheimer’s vs. matched normal controls can be achieved with a combination of acoustic features from speech, linguistic features extracted from a transcription of the speech, and results of a mini mental state exam. A fully automatic speech recognition system tuned for the speech-to-text aspect of this application, including automatic punctuation, is also described.

Highlights

Dementia is broadly defined as deterioration in memory, thinking and behavior that decreases a person’s ability to function independently in daily life (McKhann et al, 2011)
The greedy algorithm combined with the neural network two-way classifier was very promising for both feature selection and final recognizer
This approach was at least two orders of magnitude faster than the genetic algorithm (GA) method

Summary

INTRODUCTION

Dementia is broadly defined as deterioration in memory, thinking and behavior that decreases a person’s ability to function independently in daily life (McKhann et al, 2011). Using pattern discovery algorithms to identify minimal size feature sets, we provide evidence that combining selected speech features with the MMSE can yield an improved diagnostic test for detecting probable Alzheimer’s disease These results were obtained using features extracted automatically by algorithms applied to the speech signal (wave file) and either manually produced transcripts or fully automated transcripts produced by a custom designed ASR and punctuation system. Every node in a decision tree, represents a decision (target) based on a single feature and a threshold which splits the dataset into two so that similar response values are collected in the same set On this fully automated system, 90% of the data (66 subjects) were used for training while the remaining speakers were used just one time for testing. A summary of the key results for the random forest and neural network two way classifiers (AD/NL) are given in Table 5, in terms of accuracy, sensitivity, and specificity

CONCLUSIONS

Findings

ETHICS STATEMENT