Published By :Anjali Raja

Language is a valuable source of clinical information in Alzheimer’s disease, as it declines concurrently with neurodegeneration. Consequently, speech and language data have been extensively studied in connection with its diagnosis.

Scientists from the University of Edinburgh Medical School conducted a study to summarize the existing findings on the use of artificial intelligence, speech, and language processing to predict cognitive decline in the context of Alzheimer’s disease. The detailed current research procedures, highlight their limitations, and suggest strategies to address them.

Recent studies on the use of AI in Alzheimer’s disease (AD) entail using language and speech data collected in different ways and applying computational speech processing for diagnosis, prognosis, or progression modelling. This technology encompasses methods for recognizing, analyzing, and understanding spoken discourse. It implies that at least part of the AD detection process could be automated (passive). Machine learning methods have been central to this research program. Research on automatic processing of speech and language with AI and machine learning methods have yielded encouraging results and attracted increasing interest. Different approaches have been studied, including computational linguistics.

However, investigations of the use of language and speech technology in AD research are heterogeneous, which makes consensus, conclusions, and translation into larger studies or clinical practice problematic. Despite progress in research, the small, inconsistent, single-laboratory and non-standardized nature of most studies has yielded results that are not robust enough to be aggregated and thereafter implemented toward those goals. This has resulted in gaps between research contexts, clinical potential, and actual clinical applications of this new technology.

AI, ML and speech technologies

While AI encompasses a variety of symbol manipulation systems and manual encoding of expert knowledge, the majority of methods and techniques employed by the studies reviewed here concern machine learning methods. The general architecture of a machine learning system as used in AD prediction based on speech and language can be described in terms of the learning task, data representation, learning algorithm, nature of the “training data”, and performance measures. The learning task concerns the specification of the function to be learned by the system.

According to the study, most of the reviewed articles aim to use acoustic and/or linguistic features in order to distinguish the speech produced by healthy participants from the one produced by participants with a certain degree of cognitive impairment. The majority of studies attempt binary models to detecting AD and, less often, MCI, in comparison to HC. A few studies also attempt to distinguish between MCI an AD. Even when the dataset contains three or four groups most studies only report pairwise group comparisons.

Despite many titles mentioning cognitive monitoring, most research addresses only the presence or absence of cognitive impairments. The researchers suggest that future research could take further advantage of this longitudinal aspect to build models able to generate a score reflecting risk of developing an impairment.

The study conducted the first systematic review on the potential application of interactive AI methods to AD detection and progression monitoring using natural language processing and speech technology to extract “digital biomarkers” for machine learning modelling. They sound that it seems reasonable to conclude that this is a very promising field, with potential to gradually introduce changes into clinical practice. Almost all studies report relatively high performance, despite the difficulties inherent to the type of data used and the heterogeneity of the methods.

The conclude that novel AI/ML digital biomarkers could be used in combination with established biomarkers to target populations at risk of later dementia onset, as has already been proposed. It needs to be emphasized that recorded data are considered personal data (i.e., with potential to identify a subject), with the ethical and regulatory hurdles this entails as regards data collection and analysis.

Sources of Article

Full Report

Author

Anjali Raja

Anjali Raja K, Content & Research Associate, Anjali is a firm believer in the developmental power of AI and its capacity to redefine the technological advancement of the world.