Machine learning approaches to diagnosis and laterality effects in semantic dementia discourse

AI and Aphasia in the Digital Age: A Critical Review

Aphasiology has a long and rich tradition of contributing to understanding how culture, language, and social environment contribute to brain development and function. Recent breakthroughs in AI can transform the role of aphasiology in the digital age by leveraging speech data in all languages to model how damage to specific brain regions impacts linguistic universals such as grammar. These tools, including generative AI (ChatGPT) and natural language processing (NLP) models, could also inform practitioners working with clinical populations in the assessment and treatment of aphasia using AI-based interventions such as personalized therapy and adaptive platforms. Although these possibilities have generated enthusiasm in aphasiology, a rigorous interrogation of their limitations is necessary before AI is integrated into practice. We explain the history and first principles of reciprocity between AI and aphasiology, highlighting how lesioning neural networks opened the black box of cognitive neurolinguistic processing. We then argue that when more data from aphasia across languages become digitized and available online, deep learning will reveal hitherto unreported patterns of language processing of theoretical interest for aphasiologists. We also anticipate some problems using AI, including language biases, cultural, ethical, and scientific limitations, a misrepresentation of marginalized languages, and a lack of rigorous validation of tools. However, as these challenges are met with better governance, AI could have an equitable impact.

Natural language processing techniques for studying language in pathological ageing: A scoping review

BACKGROUND

In the past few years there has been a growing interest in the employment of verbal productions as digital biomarkers, namely objective, quantifiable behavioural data that can be collected and measured by means of digital devices, allowing for a low-cost pathology detection, classification and monitoring. Numerous research papers have been published on the automatic detection of subtle verbal alteration, starting from written texts, raw speech recordings and transcripts, and such linguistic analysis has been singled out as a cost-effective method for diagnosing dementia and other medical conditions common among elderly patients (e.g., cognitive dysfunctions associated with metabolic disorders, dysarthria).

AIMS

To provide a critical appraisal and synthesis of evidence concerning the application of natural language processing (NLP) techniques for clinical purposes in the geriatric population. In particular, we discuss the state of the art on studying language in healthy and pathological ageing, focusing on the latest research efforts to build non-intrusive language-based tools for the early identification of cognitive frailty due to dementia. We also discuss some challenges and open problems raised by this approach.

METHODS & PROCEDURES

We performed a scoping review to examine emerging evidence about this novel domain. Potentially relevant studies published up to November 2021 were identified from the databases of MEDLINE, Cochrane and Web of Science. We also browsed the proceedings of leading international conferences (e.g., ACL, COLING, Interspeech, LREC) from 2017 to 2021, and checked the reference lists of relevant studies and reviews.

MAIN CONTRIBUTION

The paper provides an introductory, but complete, overview of the application of NLP techniques for studying language disruption due to dementia. We also suggest that this technique can be fruitfully applied to other medical conditions (e.g., cognitive dysfunctions associated with dysarthria, cerebrovascular disease and mood disorders).

CONCLUSIONS & IMPLICATIONS

Despite several critical points need to be addressed by the scientific community, a growing body of empirical evidence shows that NLP techniques can represent a promising tool for studying language changes in pathological aging, with a high potential to lead a significant shift in clinical practice.

WHAT THIS PAPER ADDS

What is already known on this subject Speech and languages abilities change due to non-pathological neurocognitive ageing and neurodegenerative processes. These subtle verbal modifications can be measured through NLP techniques and used as biomarkers for screening/diagnostic purposes in the geriatric population (i.e., digital linguistic biomarkers-DLBs). What this paper adds to existing knowledge The review shows that DLBs can represent a promising clinical tool, with a high potential to spark a major shift to dementia assessment in the elderly. Some challenges and open problems are also discussed. What are the potential or actual clinical implications of this work? This methodological review represents a starting point for clinicians approaching the DLB research field for studying language in healthy and pathological ageing. It summarizes the state of the art and future research directions of this novel approach.

Bigram semantic distance as an index of continuous semantic flow in natural language: Theory, tools, and applications

Much of our understanding of word meaning has been informed through studies of single words. High-dimensional semantic space models have recently proven instrumental in elucidating connections between words. Here we show how bigram semantic distance can yield novel insights into conceptual cohesion and topic flow when computed over continuous language samples. For example, "Cats drink milk" is comprised of an ordered vector of bigrams (cat-drink, drink-milk). Each of these bigrams has a unique semantic distance. These distances in turn may provide a metric of dispersion or the flow of concepts as language unfolds. We offer an R-package ("semdistflow") that transforms any user-specified language transcript into a vector of ordered bigrams, appending two metrics of semantic distance to each pair. We validated these distance metrics on a continuous stream of simulated verbal fluency data assigning predicted switch markers between alternating semantic clusters (animals, musical instruments, fruit). We then generated bigram distance norms on a large sample of text and demonstrated applications of the technique to a classic work of short fiction, To Build a Fire (London, 1908). In one application, we showed that bigrams spanning sentence boundaries are punctuated by jumps in the semantic distance. We discuss the promise of this technique for characterizing semantic processing in real-world narratives and for bridging findings at the single word level with macroscale discourse analyses. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

Frontotemporal dementia: A unique window on the functional role of the temporal lobes

Frontotemporal dementia (FTD) is an umbrella term covering a plethora of progressive changes in executive functions, motor abilities, behavior, and/or language. Different clinical syndromes have been described in relation to localized atrophy, informing on the functional networks that underlie these specific cognitive, emotional, and behavioral processes. These functional declines are linked with the underlying neurodegeneration of frontal and/or temporal lobes due to diverse molecular pathologies. Initially, the accumulation of misfolded proteins targets specifically susceptible cell assemblies, leading to relatively focal neurodegeneration that later spreads throughout large-scale cortical networks. Here, we discuss the most recent clinical, neuropathological, imaging, and genetics findings in FTD-spectrum syndromes affecting the temporal lobe. We focus on the semantic variant of primary progressive aphasia and its mirror image, the right temporal variant of FTD. Incipient focal atrophy of the left anterior temporal lobe (ATL) manifests with predominant naming, word comprehension, reading, and object semantic deficits, while cases of predominantly right ATL atrophy present with impairments of socioemotional, nonverbal semantic, and person-specific knowledge. Overall, the observations in FTD allow for crucial clinical-anatomic inferences, shedding light on the role of the temporal lobes in both cognition and complex behaviors. The concerted activity of both ATLs is critical to ensure that percepts are translated into concepts, yet important hemispheric differences should be acknowledged. On one hand, the left ATL attributes meaning to linguistic, external stimuli, thus supporting goal-oriented, action-related behaviors (e.g., integrating sounds and letters into words). On the other hand, the right ATL assigns meaning to emotional, visceral stimuli, thus guiding socially relevant behaviors (e.g., integrating body sensations into feelings of familiarity).

Language Decline Characterizes Amnestic Mild Cognitive Impairment Independent of Cognitive Decline

Purpose This research investigated the nature of cognitive decline in prodromal Alzheimer's disease (AD), particularly in mild cognitive impairment, amnestic type (aMCI). We assessed language in aMCI as compared with healthy aging (HA) and healthy young (HY) with new psycholinguistic assessment of complex sentences, and we tested the degree to which deficits on this language measure relate to performance in other general cognitive domains such as memory. Method Sixty-one individuals with aMCI were compared with 24 HA and 10 HY adults on a psycholinguistic measure of complex sentence production (relative clauses). In addition, HA, HY, and a subset of the aMCI participants (n = 22) were also tested on a multidomain cognitive screen, the Addenbrooke's Cognitive Examination-Revised (ACE-R), and on a verbal working memory Brown-Peterson (BP) test. General and generalized linear mixed models were used to test psycholinguistic results and to test whether ACE-R and BP performance predicted performance on the psycholinguistic test similarly in the aMCI and HA groups. Results On the psycholinguistic measure, sentence imitation was significantly deficited in aMCI in comparison with that in HA and HY. Experimental factorial designs revealed that individuals with aMCI had particular difficulty repeating sentences that especially challenged syntax-semantics integration. As expected, the aMCI group also performed significantly below the HY and HA groups on the ACE-R. Neither the ACE-R Memory subtest nor the BP total scores predicted performance on the psycholinguistic task for either the aMCI or the HA group. However, the ACE-R total score significantly predicted psycholinguistic task performance, with increased ACE-R performance predicting increased psycholinguistic task performance only for the HA group, not for the aMCI group. Conclusions Results suggest a selective deterioration in language in aMCI, specifically a weakening of syntax-semantics integration in complex sentence processing, and a general independence of this language deficit and memory decline. Results cohere with previous assessments of the nature of difficulty in complex sentence formation in aMCI. We argue that clinical screening for prodromal AD can be strengthened by supplementary testing of language, as well as memory, and extended evaluation of strength of their relation.

Connected speech markers of amyloid burden in primary progressive aphasia

INTRODUCTION

Positron emission tomography (PET) amyloid imaging has become an important part of the diagnostic workup for patients with primary progressive aphasia (PPA) and uncertain underlying pathology. Here, we employ a semi-automated analysis of connected speech (CS) with a twofold objective. First, to determine if quantitative CS features can help select primary progressive aphasia (PPA) patients with a higher probability of a positive PET amyloid imaging result. Second, to examine the relevant group differences from a clinical perspective.

METHODS

117 CS samples from a well-characterised cohort of PPA patients who underwent PET amyloid imaging were collected. Expert consensus established PET amyloid status for each patient, and 40% of the sample was amyloid positive.

RESULTS

Leave-one-out cross-validation yields 77% classification accuracy (sensitivity: 74%, specificity: 79%).

DISCUSSION

Our results confirm the potential of CS analysis as a screening tool. Discriminant CS features from lexical, syntactic, pragmatic, and semantic domains are discussed.

Enhancing the Classification of Aphasia: A Statistical Analysis Using Connected Speech

BACKGROUND

Large shared databases and automated language analyses allow for the application of new data analysis techniques that can shed new light on the connected speech of people with aphasia (PWA).

AIMS

To identify coherent clusters of PWA based on language output using unsupervised statistical algorithms and to identify features that are most strongly associated with those clusters.

METHODS & PROCEDURES

Clustering and classification methods were applied to language production data from 168 PWA. Language samples were from a standard discourse protocol tapping four genres: free speech personal narratives, picture descriptions, Cinderella storytelling, procedural discourse.

OUTCOMES & RESULTS

Seven distinct clusters of PWA were identified by the K-means algorithm. Using the random forests algorithm, a classification tree was proposed and validated, showing 91% agreement with the cluster assignments. This representative tree used only two variables to divide the data into distinct groups: total words from free speech tasks and total closed class words from the Cinderella storytelling task.

CONCLUSION

Connected speech data can be used to distinguish PWA into coherent groups, providing insight into traditional aphasia classifications, factors that may guide discourse research and clinical work.

Telltale silence: temporal speech parameters discriminate between prodromal dementia and mild Alzheimer's disease

This study presents a novel approach for the early detection of mild cognitive impairment (MCI) and mild Alzheimer's disease (mAD) in the elderly. Participants were 25 elderly controls (C), 25 clinically diagnosed MCI and 25 mAD patients, included after a clinical diagnosis validated by CT or MRI and cognitive tests. Our linguistic protocol involved three connected speech tasks that stimulate different memory systems, which were recorded, then analyzed linguistically by using the PRAAT software. The temporal speech-related parameters successfully differentiate MCI from mAD and C, such as speech rate, number and length of pauses, the rate of pause and signal. Parameters pauses/duration and silent pauses/duration linearly decreased among the groups, in other words, the percentage of pauses in the total duration of speech continuously grows as dementia progresses. Thus, the proposed approach may be an effective tool for screening MCI and mAD.

Language training for oral and written naming impairment in primary progressive aphasia: a review

BACKGROUND

Primary progressive aphasia (PPA) is a neurodegenerative disorder characterized by a gradual, insidious and progressive loss of language abilities, with naming difficulties being an early and persistent impairment common to all three variants. In the absence of effective pharmacological treatments and given the progressive nature of the disorder, in the past few decades, many studies have investigated the effectiveness of language training to minimize the functional impact of word-finding difficulties in daily life.

MAIN BODY

We review language treatments most commonly used in clinical practice among patients with different variants of PPA, with a focus on the enhancement of spoken and written naming abilities. Generalization of gains to the ability to name untrained stimuli or to other language abilities and the maintenance of these results over time are also discussed. Forty-eight studies were included in this literature review, identifying four main types of language treatment: a) lexical retrieval treatment, b) phonological and/or orthographic treatment, c) semantic treatment, and d) a multimodality approach treatment. Overall, language training is able to induce immediate improvements of naming abilities in all variants of PPA. Moreover, despite the large variability among results, generalization and long-term effects can be recorded after the training. The reviewed studies also suggest that one factor that determines the choice of a particular approach is the compromised components of the lexical/semantic processing system.

CONCLUSION

The majority of studies have demonstrated improvements of naming abilities following language treatments. Given the progressive nature of PPA, it is essential to apply language treatment in the early stages of the disease.

The Effects of Symptom Onset Location on Automatic Amyotrophic Lateral Sclerosis Detection Using the Correlation Structure of Articulatory Movements

Purpose Kinematic measurements of speech have demonstrated some success in automatic detection of early symptoms of amyotrophic lateral sclerosis (ALS). In this study, we examined how the region of symptom onset (bulbar vs. spinal) affects the ability of data-driven models to detect ALS. Method We used a correlation structure of articulatory movements combined with a machine learning model (i.e., artificial neural network) to detect differences between people with ALS and healthy controls. The performance of this system was evaluated separately for participants with bulbar onset and spinal onset to examine how region of onset affects classification performance. We then performed a regression analysis to examine how different severity measures and region of onset affects model performance. Results The proposed model was significantly more accurate in classifying the bulbar-onset participants, achieving an area under the curve of 0.809 relative to the 0.674 achieved for spinal-onset participants. The regression analysis, however, found that differences in classifier performance across participants were better explained by their speech performance (intelligible speaking rate), and no significant differences were observed based on region of onset when intelligible speaking rate was accounted for. Conclusions Although we found a significant difference in the model's ability to detect ALS depending on the region of onset, this disparity can be primarily explained by observable differences in speech motor symptoms. Thus, when the severity of speech symptoms (e.g., intelligible speaking rate) was accounted for, symptom onset location did not affect the proposed computational model's ability to detect ALS.

Analysis and Classification of Word Co-Occurrence Networks From Alzheimer’s Patients and Controls

In this paper we construct word co-occurrence networks from transcript data of controls and patients with potential Alzheimer’s disease using the ADReSS challenge dataset of spontaneous speech. We examine measures of the structure of these networks for significant differences, finding that networks from Alzheimer’s patients have a lower heterogeneity and centralization, but a higher edge density. We then use these measures, a network embedding method and some measures from the word frequency distribution to classify the transcripts into control or Alzheimer’s, and to estimate the cognitive test score of a participant based on the transcript. We find it is possible to distinguish between the AD and control networks on structure alone, achieving 66.7% accuracy on the test set, and to predict cognitive scores with a root mean squared error of 5.675. Using the network measures is more successful than using the network embedding method. However, if the networks are shuffled we find relatively few of the measures are different, indicating that word frequency drives many of the network properties. This observation is borne out by the classification experiments, where word frequency measures perform similarly to the network measures.

Modeling Large Sparse Data for Feature Selection: Hospital Admission Predictions of the Dementia Patients Using Primary Care Electronic Health Records

A growing elderly population suffering from incurable, chronic conditions such as dementia present a continual strain on medical services due to mental impairment paired with high comorbidity resulting in increased hospitalization risk. The identification of at risk individuals allows for preventative measures to alleviate said strain. Electronic health records provide opportunity for big data analysis to address such applications. Such data however, provides a challenging problem space for traditional statistics and machine learning due to high dimensionality and sparse data elements. This article proposes a novel machine learning methodology: entropy regularization with ensemble deep neural networks (ECNN), which simultaneously provides high predictive performance of hospitalization of patients with dementia whilst enabling an interpretable heuristic analysis of the model architecture, able to identify individual features of importance within a large feature domain space. Experimental results on health records containing 54,647 features were able to identify 10 event indicators within a patient timeline: a collection of diagnostic events, medication prescriptions and procedural events, the highest ranked being essential hypertension. The resulting subset was still able to provide a highly competitive hospitalization prediction (Accuracy: 0.759) as compared to the full feature domain (Accuracy: 0.755) or traditional feature selection techniques (Accuracy: 0.737), a significant reduction in feature size. The discovery and heuristic evidence of correlation provide evidence for further clinical study of said medical events as potential novel indicators. There also remains great potential for adaption of ECNN within other medical big data domains as a data mining tool for novel risk factor identification.

Detecting Alzheimer's Disease from Continuous Speech Using Language Models

BACKGROUND

Recently, many studies have been carried out to detect Alzheimer's disease (AD) from continuous speech by linguistic analysis and modeling. However, few of them utilize language models (LMs) to extract linguistic features and to investigate the lexical-level differences between AD and healthy speech.

OBJECTIVE

Our goals include obtaining state-of-art performance of automatic AD detection, emphasizing N-gram LMs as powerful tools for distinguishing AD patients' narratives from those of healthy controls, and discovering the differences of lexical usages between AD patients and healthy people.

METHOD

We utilize a subset of the DementiaBank corpus, including 242 control samples from 99 control participants and 256 AD samples from 169 "PossibleAD" or "ProbableAD" participants. Baseline models are built through area under curve-based feature selection and using five machine learning algorithms for comparison. Perplexity features are extracted using LMs to build enhanced detection models. Finally, the differences of lexical usages between AD patients and healthy people are investigated by a proportion test based on unigram probabilities.

RESULTS

Our baseline model obtains a detection accuracy of 80.7%. This accuracy increases to 85.4% after integrating the perplexity features derived from LMs. Further investigations show that AD patients tend to use more general, less informative, and less accurate words to describe characters and actions than healthy controls.

CONCLUSION

The perplexity features extracted by LMs can benefit the automatic AD detection from continuous speech. There exist lexical-level differences between AD and healthy speech that can be captured by statistical N-gram LMs.

A Comprehensive Machine-Learning Model Applied to Magnetic Resonance Imaging (MRI) to Predict Alzheimer's Disease (AD) in Older Subjects

Increasing evidence suggests the utility of magnetic resonance imaging (MRI) as an important technique for the diagnosis of Alzheimer’s disease (AD) and for predicting the onset of this neurodegenerative disorder. In this study, we present a sophisticated machine learning (ML) model of great accuracy to diagnose the early stages of AD. A total of 373 MRI tests belonging to 150 subjects (age ≥ 60) were examined and analyzed in parallel with fourteen distinct features related to standard AD diagnosis. Four ML models, such as naive Bayes (NB), artificial neural networks (ANN), K-nearest neighbor (KNN), and support-vector machines (SVM), and the receiver operating characteristic (ROC) curve metric were used to validate the model performance. Each model evaluation was done in three independent experiments. In the first experiment, a manual feature selection was used for model training, and ANN generated the highest accuracy in terms of ROC (0.812). In the second experiment, automatic feature selection was conducted by wrapping methods, and the NB achieved the highest ROC of 0.942. The last experiment consisted of an ensemble or hybrid modeling developed to combine the four models. This approach resulted in an improved accuracy ROC of 0.991. We conclude that the involvement of ensemble modeling, coupled with selective features, can predict with better accuracy the development of AD at an early stage.

A "Verbal Thermometer" for Assessing Neurodegenerative Disease: Automated Measurement of Pronoun and Verb Ratio from Speech

Clinicians often use speech to characterize neurodegenerative disorders. Such characterizations require clinical judgment, which is subjective and can require extensive training. Quantitative Production Analysis (QPA) can be used to obtain objective quantifiable assessments of patient functioning. However, such human-based analyses of speech are costly and time consuming. Inexpensive off-the-shelf technologies such as speech recognition and part of speech taggers may avoid these problems. This study evaluates the ability of an automatic speech to text transcription system and a part of speech tagger to assist with measuring pronoun and verb ratios, measures based on QPA. Five participant groups provided spontaneous speech samples. One group consisted of healthy controls, while the remaining groups represented four subtypes of frontotemporal dementia. Findings indicated measurement of pronoun and verb ratio was robust despite errors introduced by automatic transcription and the tagger and despite these off-the-shelf products not having been trained on the language obtained from speech of the included population.

Predicting Montreal Cognitive Assessment Scores From Measures of Speech and Language

Purpose This study examined the relationship between measurements derived from spontaneous speech and participants' scores on the Montreal Cognitive Assessment. Method Participants (N = 521) aged between 64 and 97 years completed the cognitive assessment and were prompted to describe an early childhood memory. A range of acoustic and linguistic measures was extracted from the resulting speech sample. A least absolute shrinkage and selection operator approach was used to model the relationship between acoustic, lexical, and demographic information and participants' scores on the cognitive assessment. Results Using the covariance test statistic, four important variables were identified, which, together, explained 16.52% of the variance in participants' cognitive scores. Conclusions The degree to which cognition can be accurately predicted through spontaneously produced speech samples is limited. Statistically significant relationships were found between specific measurements of lexical variation, participants' speaking rate, and their scores on the Montreal Cognitive Assessment.

Artificial intelligence and neuropsychological measures: The case of Alzheimer's disease

One of the current challenges in the field of Alzheimer's disease (AD) is to identify patients with mild cognitive impairment (MCI) that will convert to AD. Artificial intelligence, in particular machine learning (ML), has established as one of more powerful approach to extract reliable predictors and to automatically classify different AD phenotypes. It is time to accelerate the translation of this knowledge in clinical practice, mainly by using low-cost features originating from the neuropsychological assessment. We performed a meta-analysis to assess the contribution of ML and neuropsychological measures for the automated classification of MCI patients and the prediction of their conversion to AD. The pooled sensitivity and specificity of patients' classifications was obtained by means of a quantitative bivariate random-effect meta-analytic approach. Although a high heterogeneity was observed, the results of meta-analysis show that ML applied to neuropsychological measures can lead to a successful automatic classification, being more specific as screening rather than prognosis tool. Relevant categories of neuropsychological tests can be extracted by ML that maximize the classification accuracy.

Changes in the Rhythm of Speech Difference between People with Nondegenerative Mild Cognitive Impairment and with Preclinical Dementia

This study explores several speech parameters related to mild cognitive impairment, as well as those that might be flagging the presence of an underlying neurodegenerative process. Speech is an excellent biomarker because it is not invasive and, what is more, its analysis is rapid and economical. Our aim has been to ascertain whether the typical speech patterns of people with Alzheimer's disease are also present during the disorder's preclinical stages. To do so, we shall be using a task that involves reading out aloud. This is followed by an analysis of the recordings, looking for the possible parameters differentiating between those older people with MCI and a high probability of developing dementia and those with MCI that will not do so. We found that the disease's most differentiating parameters prior to its onset involve changes in speech duration and an alteration in rhythm rate and intensity. These parameters seem to be related to the first difficulties in lexical access among older people with AD.

A Review of Automated Speech and Language Features for Assessment of Cognitive and Thought Disorders

It is widely accepted that information derived from analyzing speech (the acoustic signal) and language production (words and sentences) serves as a useful window into the health of an individual's cognitive ability. In fact, most neuropsychological testing batteries have a component related to speech and language where clinicians elicit speech from patients for subjective evaluation across a broad set of dimensions. With advances in speech signal processing and natural language processing, there has been recent interest in developing tools to detect more subtle changes in cognitive-linguistic function. This work relies on extracting a set of features from recorded and transcribed speech for objective assessments of speech and language, early diagnosis of neurological disease, and tracking of disease after diagnosis. With an emphasis on cognitive and thought disorders, in this paper we provide a review of existing speech and language features used in this domain, discuss their clinical application, and highlight their advantages and disadvantages. Broadly speaking, the review is split into two categories: language features based on natural language processing and speech features based on speech signal processing. Within each category, we consider features that aim to measure complementary dimensions of cognitive-linguistics, including language diversity, syntactic complexity, semantic coherence, and timing. We conclude the review with a proposal of new research directions to further advance the field.

An Automated Approach to Examining Pausing in the Speech of People With Dementia

Dementia is a common neurodegenerative condition involving the deterioration of cognitive and communication skills. Pausing in the speech of people with dementia is a dysfluency that may be used to signal conversational trouble in social interaction. This study aimed to examine the speech-pausing profile within picture description samples from people with dementia and healthy controls (HCs) within the DementiaBank database using the Calpy computational speech processing toolkit. Sixty English-speaking participants between the ages of 53 and 88 years (Mage = 67.43, SD = 8.33; 42 females) were included in the study: 20 participants with mild cognitive impairment, 20 participants with moderate cognitive impairment, and 20 HCs. Quantitative analysis shows a progressive increase in the duration of pausing between HCs, the mild dementia group, and the moderate dementia group, respectively.

Predicting dementia with routine care EMR data

Our aim is to develop a machine learning (ML) model that can predict dementia in a general patient population from multiple health care institutions one year and three years prior to the onset of the disease without any additional monitoring or screening. The purpose of the model is to automate the cost-effective, non-invasive, digital pre-screening of patients at risk for dementia. Towards this purpose, routine care data, which is widely available through Electronic Medical Record (EMR) systems is used as a data source. These data embody a rich knowledge and make related medical applications easy to deploy at scale in a cost-effective manner. Specifically, the model is trained by using structured and unstructured data from three EMR data sets: diagnosis, prescriptions, and medical notes. Each of these three data sets is used to construct an individual model along with a combined model which is derived by using all three data sets. Human-interpretable data processing and ML techniques are selected in order to facilitate adoption of the proposed model by health care providers from multiple institutions. The results show that the combined model is generalizable across multiple institutions and is able to predict dementia within one year of its onset with an accuracy of nearly 80% despite the fact that it was trained using routine care data. Moreover, the analysis of the models identified important predictors for dementia. Some of these predictors (e.g., age and hypertensive disorders) are already confirmed by the literature while others, especially the ones derived from the unstructured medical notes, require further clinical analysis.

Harnessing the Power of Machine Learning in Dementia Informatics Research: Issues, Opportunities, and Challenges

Dementia is a chronic and degenerative condition affecting millions globally. The care of patients with dementia presents an ever-continuing challenge to healthcare systems in the 21st century. Medical and health sciences have generated unprecedented volumes of data related to health and wellbeing for patients with dementia due to advances in information technology, such as genetics, neuroimaging, cognitive assessment, free texts, routine electronic health records, etc. Making the best use of these diverse and strategic resources will lead to high-quality care of patients with dementia. As such, machine learning becomes a crucial factor in achieving this objective. The aim of this paper is to provide a state-of-the-art review of machine learning methods applied to health informatics for dementia care. We collate and review the existing scientific methodologies and identify the relevant issues and challenges when faced with big health data. Machine learning has demonstrated promising applications to neuroimaging data analysis for dementia care, while relatively less effort has been made to make use of integrated heterogeneous data via advanced machine learning approaches. We further indicate future potential and research directions in applying advanced machine learning, such as deep learning, to dementia informatics.

The language profile of progressive supranuclear palsy

A progressive speech/language disorder, such as the non fluent/agrammatic variant of primary progressive aphasia and progressive apraxia of speech, can be due to neuropathologically verified Progressive Supranuclear Palsy (PSP). The prevalence of linguistic deficits and the linguistic profile in PSP patients who present primarily with a movement disorder is unknown. In the present study, we investigated speech and language performance in a sample of clinically diagnosed PSP patients using a comprehensive language battery, including, besides traditional language tests, a detailed analysis of connected speech (picture description task assessing 26 linguistic features). The aim was to identify the most affected linguistic levels in seventeen PSP with a movement disorder presentation, compared to 21 patients with Parkinson's disease and 27 healthy controls. Machine learning methods were used to detect the most relevant language tests and linguistic features characterizing the language profile of PSP patients. Our results indicate that even non-clinically aphasic PSP patients have subtle language deficits, in particular involving the lexical-semantic and discourse levels. Patients with the Richardson's syndrome showed a lower performance in the word comprehension task with respect to the other PSP phenotypes with predominant frontal presentation, parkinsonism and progressive gait freezing. The present findings support the usefulness of a detailed language assessment in all patients in the PSP spectrum.

Machine Learning for Predicting Cognitive Diseases: Methods, Data Sources and Risk Factors

Machine learning and data mining approaches are being successfully applied to different fields of life sciences for the past 20 years. Medicine is one of the most suitable application domains for these techniques since they help model diagnostic information based on causal and/or statistical data and therefore reveal hidden dependencies between symptoms and illnesses. In this paper we give a detailed overview of the recent machine learning research and its applications for predicting cognitive diseases, especially the Alzheimer's disease, mild cognitive impairment and the Parkinson's disease. We survey different state-of-the-art methodological approaches, data sources and public data, and provide their comparative analysis. We conclude by identifying the open problems within the field that include an early detection of the cognitive diseases and inclusion of machine learning tools into diagnostic practice and therapy planning.

Clinical, Anatomical, and Pathological Features in the Three Variants of Primary Progressive Aphasia: A Review

Primary progressive aphasias (PPA) are neurodegenerative diseases clinically characterized by an early and relatively isolated language impairment. Three main clinical variants, namely the nonfluent/agrammatic variant (nfvPPA), the semantic variant (svPPA), and the logopenic variant (lvPPA) have been described, each with specific linguistic/cognitive deficits, corresponding anatomical and most probable pathological features. Since the discovery and the development of diagnostic criteria for the PPA variants by the experts in the field, significant progress has been made in the understanding of these diseases. This review aims to provide an overview of the literature on each of the PPA variant in terms of their clinical, anatomical and pathological features, with a specific focus on recent findings. In terms of clinical advancements, recent studies have allowed a better characterization and differentiation of PPA patients based on both their linguistic and non-linguistic profiles. In terms of neuroimaging, techniques such as diffusion imaging and resting-state fMRI have allowed a deeper understanding of the impact of PPA on structural and functional connectivity alterations beyond the well-defined pattern of regional gray matter atrophy. Finally, in terms of pathology, despite significant advances, clinico-pathological correspondence in PPA remains far from absolute. Nonetheless, the improved characterization of PPA has the potential to have a positive impact on the management of patients. Improved reliability of diagnoses and the development of reliable in vivo biomarkers for underlying neuropathology will also be increasingly important in the future as trials for etiology-specific treatments become available.

Language in individuals with left hemisphere tumors: Is spontaneous speech analysis comparable to formal testing?

BACKGROUND

The relationship between spontaneous speech and formal language testing in people with brain tumors (gliomas) has been rarely studied. In clinical practice, formal testing is typically used, while spontaneous speech is less often evaluated quantitatively. However, spontaneous speech is quicker to sample and may be less prone to test/retest effects, making it a potential candidate for assessing language impairments when there is restricted time or when the patient is unable to undertake prolonged testing.

AIM

To assess whether quantitative spontaneous speech analysis and formal testing detect comparable language impairments in people with gliomas. Specifically, we addressed (a) whether both measures detected comparable language impairments in our patient sample; and (b) which language levels, assessment times, and spontaneous speech variables were more often impaired in this subject group.

METHOD

Five people with left perisylvian gliomas performed a spontaneous speech task and a formal language assessment. Tests were administered before surgery, within a week after surgery, and seven months after surgery. Performance on spontaneous speech was compared with that of 15 healthy speakers.

RESULTS

Language impairments were detected more often with both measures than with either measure independently. Lexical-semantic impairments were more common than phonological and grammatical impairments, and performance was equally impaired across assessment time points. Incomplete sentences and phonological paraphasias were the most common error types.

CONCLUSIONS

In our sample both spontaneous speech analysis and formal testing detected comparable language impairments. Currently, we suggest that formal testing remains overall the better option, except for cases in which there are restrictions on testing time or the patient is too tired to undergo formal testing. In these cases, spontaneous speech may provide a viable alternative, particularly if automated analysis of spontaneous speech becomes more readily available in the future. These results await replication in a bigger sample and/or other populations.

A Speech Recognition-based Solution for the Automatic Detection of Mild Cognitive Impairment from Spontaneous Speech

BACKGROUND

Even today the reliable diagnosis of the prodromal stages of Alzheimer's disease (AD) remains a great challenge. Our research focuses on the earliest detectable indicators of cognitive decline in mild cognitive impairment (MCI). Since the presence of language impairment has been reported even in the mild stage of AD, the aim of this study is to develop a sensitive neuropsychological screening method which is based on the analysis of spontaneous speech production during performing a memory task. In the future, this can form the basis of an Internet-based interactive screening software for the recognition of MCI.

METHODS

Participants were 38 healthy controls and 48 clinically diagnosed MCI patients. The provoked spontaneous speech by asking the patients to recall the content of 2 short black and white films (one direct, one delayed), and by answering one question. Acoustic parameters (hesitation ratio, speech tempo, length and number of silent and filled pauses, length of utterance) were extracted from the recorded speech signals, first manually (using the Praat software), and then automatically, with an automatic speech recognition (ASR) based tool. First, the extracted parameters were statistically analyzed. Then we applied machine learning algorithms to see whether the MCI and the control group can be discriminated automatically based on the acoustic features.

RESULTS

The statistical analysis showed significant differences for most of the acoustic parameters (speech tempo, articulation rate, silent pause, hesitation ratio, length of utterance, pause-per-utterance ratio). The most significant differences between the two groups were found in the speech tempo in the delayed recall task, and in the number of pauses for the question-answering task. The fully automated version of the analysis process - that is, using the ASR-based features in combination with machine learning - was able to separate the two classes with an F1-score of 78.8%.

CONCLUSION

The temporal analysis of spontaneous speech can be exploited in implementing a new, automatic detection-based tool for screening MCI for the community.

Machine learning applied to neuroimaging for diagnosis of adult classic Chiari malformation: role of the basion as a key morphometric indicator

OBJECTIVE The current diagnostic criterion for Chiari malformation Type I (CM-I), based on tonsillar herniation (TH), includes a diversity of patients with amygdalar descent that may be caused by a variety of factors. In contrast, patients presenting with an overcrowded posterior cranial fossa, a key characteristic of the disease, may remain misdiagnosed if they have little or no TH. The objective of the present study was to use machine-learning classification methods to identify morphometric measures that help discern patients with classic CM-I to improve diagnosis and treatment and provide insight into the etiology of the disease. METHODS Fifteen morphometric measurements of the posterior cranial fossa were performed on midsagittal T1-weighted MR images obtained in 195 adult patients diagnosed with CM. Seven different machine-learning classification methods were applied to images from 117 patients with classic CM-I and 50 controls matched by age and sex to identify the best classifiers discriminating the 2 cohorts with the minimum number of parameters. These classifiers were then tested using independent CM cohorts representing different entities of the disease. RESULTS Machine learning identified combinations of 2 and 3 morphometric measurements that were able to discern not only classic CM-I (with more than 5 mm TH) but also other entities such as classic CM-I with moderate TH and CM Type 1.5 (CM-1.5), with high accuracy (> 87%) and independent of the TH criterion. In contrast, lower accuracy was obtained in patients with CM Type 0. The distances from the lower aspect of the corpus callosum, pons, and fastigium to the foramen magnum and the basal and Wackenheim angles were identified as the most relevant morphometric traits to differentiate these patients. The stronger significance (p < 0.01) of the correlations with the clivus length, compared with the supraoccipital length, suggests that these 5 relevant traits would be affected more by the relative position of the basion than the opisthion. CONCLUSIONS Tonsillar herniation as a unique criterion is insufficient for radiographic diagnosis of CM-I, which can be improved by considering the basion position. The position of the basion was altered in different entities of CM, including classic CM-I, classic CM-I with moderate TH, and CM-1.5. The authors propose a predictive model based on 3 parameters, all related to the basion location, to discern classic CM-I with 90% accuracy and suggest considering the anterior alterations in the evaluation of surgical procedures and outcomes.

The acute mania of King George III: A computational linguistic analysis

We used a computational linguistic approach, exploiting machine learning techniques, to examine the letters written by King George III during mentally healthy and apparently mentally ill periods of his life. The aims of the study were: first, to establish the existence of alterations in the King's written language at the onset of his first manic episode; and secondly to identify salient sources of variation contributing to the changes. Effects on language were sought in two control conditions (politically stressful vs. politically tranquil periods and seasonal variation). We found clear differences in the letter corpus, across a range of different features, in association with the onset of mental derangement, which were driven by a combination of linguistic and information theory features that appeared to be specific to the contrast between acute mania and mental stability. The paucity of existing data relevant to changes in written language in the presence of acute mania suggests that lexical, syntactic and stylometric descriptions of written discourse produced by a cohort of patients with a diagnosis of acute mania will be necessary to support the diagnosis independently and to look for other periods of mental illness of the course of the King's life, and in other historically significant figures with similarly large archives of handwritten documents.

Connected Speech in Neurodegenerative Language Disorders: A Review

Language assessment has a crucial role in the clinical diagnosis of several neurodegenerative diseases. The analysis of extended speech production is a precious source of information encompassing the phonetic, phonological, lexico-semantic, morpho-syntactic, and pragmatic levels of language organization. The knowledge about the distinctive linguistic variables identifying language deficits associated to different neurodegenerative diseases has progressively improved in the last years. However, the heterogeneity of such variables and of the way they are measured and classified limits any generalization and makes the comparison among studies difficult. Here we present an exhaustive review of the studies focusing on the linguistic variables derived from the analysis of connected speech samples, with the aim of characterizing the language disorders of the most prevalent neurodegenerative diseases, including primary progressive aphasia, Alzheimer's disease, movement disorders, and amyotrophic lateral sclerosis. A total of 61 studies have been included, considering only those reporting group analysis and comparisons with a group of healthy persons. This review first analyzes the differences in the tasks used to elicit connected speech, namely picture description, story narration, and interview, considering the possible different contributions to the assessment of different linguistic domains. This is followed by an analysis of the terminologies and of the methods of measurements of the variables, indicating the need for harmonization and standardization. The final section reviews the linguistic domains affected by each different neurodegenerative disease, indicating the variables most consistently impaired at each level and suggesting the key variables helping in the differential diagnosis among diseases. While a large amount of valuable information is already available, the review highlights the need of further work, including the development of automated methods, to take advantage of the richness of connected speech analysis for both research and clinical purposes.

Predicting probable Alzheimer's disease using linguistic deficits and biomarkers

BACKGROUND

The manual diagnosis of neurodegenerative disorders such as Alzheimer's disease (AD) and related Dementias has been a challenge. Currently, these disorders are diagnosed using specific clinical diagnostic criteria and neuropsychological examinations. The use of several Machine Learning algorithms to build automated diagnostic models using low-level linguistic features resulting from verbal utterances could aid diagnosis of patients with probable AD from a large population. For this purpose, we developed different Machine Learning models on the DementiaBank language transcript clinical dataset, consisting of 99 patients with probable AD and 99 healthy controls.

RESULTS

Our models learned several syntactic, lexical, and n-gram linguistic biomarkers to distinguish the probable AD group from the healthy group. In contrast to the healthy group, we found that the probable AD patients had significantly less usage of syntactic components and significantly higher usage of lexical components in their language. Also, we observed a significant difference in the use of n-grams as the healthy group were able to identify and make sense of more objects in their n-grams than the probable AD group. As such, our best diagnostic model significantly distinguished the probable AD group from the healthy elderly group with a better Area Under the Receiving Operating Characteristics Curve (AUC) using the Support Vector Machines (SVM).

CONCLUSIONS

Experimental and statistical evaluations suggest that using ML algorithms for learning linguistic biomarkers from the verbal utterances of elderly individuals could help the clinical diagnosis of probable AD. We emphasise that the best ML model for predicting the disease group combines significant syntactic, lexical and top n-gram features. However, there is a need to train the diagnostic models on larger datasets, which could lead to a better AUC and clinical diagnosis of probable AD.

Linguistic Features Identify Alzheimer's Disease in Narrative Speech

BACKGROUND

Although memory impairment is the main symptom of Alzheimer's disease (AD), language impairment can be an important marker. Relatively few studies of language in AD quantify the impairments in connected speech using computational techniques.

OBJECTIVE

We aim to demonstrate state-of-the-art accuracy in automatically identifying Alzheimer's disease from short narrative samples elicited with a picture description task, and to uncover the salient linguistic factors with a statistical factor analysis.

METHODS

Data are derived from the DementiaBank corpus, from which 167 patients diagnosed with "possible" or "probable" AD provide 240 narrative samples, and 97 controls provide an additional 233. We compute a number of linguistic variables from the transcripts, and acoustic variables from the associated audio files, and use these variables to train a machine learning classifier to distinguish between participants with AD and healthy controls. To examine the degree of heterogeneity of linguistic impairments in AD, we follow an exploratory factor analysis on these measures of speech and language with an oblique promax rotation, and provide interpretation for the resulting factors.

RESULTS

We obtain state-of-the-art classification accuracies of over 81% in distinguishing individuals with AD from those without based on short samples of their language on a picture description task. Four clear factors emerge: semantic impairment, acoustic abnormality, syntactic impairment, and information impairment.

CONCLUSION

Modern machine learning and linguistic analysis will be increasingly useful in assessment and clustering of suspected AD.

Speaking in Alzheimer's Disease, is That an Early Sign? Importance of Changes in Language Abilities in Alzheimer's Disease

It is known that Alzheimer’s disease (AD) influences the temporal characteristics of spontaneous speech. These phonetical changes are present even in mild AD. Based on this, the question arises whether an examination based on language analysis could help the early diagnosis of AD and if so, which language and speech characteristics can identify AD in its early stage. The purpose of this article is to summarize the relation between prodromal and manifest AD and language functions and language domains. Based on our research, we are inclined to claim that AD can be more sensitively detected with the help of a linguistic analysis than with other cognitive examinations. The temporal characteristics of spontaneous speech, such as speech tempo, number of pauses in speech, and their length are sensitive detectors of the early stage of the disease, which enables an early simple linguistic screening for AD. However, knowledge about the unique features of the language problems associated with different dementia variants still has to be improved and refined.

Diagnosis of Cognitive Impairment Compatible with Early Diagnosis of Alzheimer's Disease. A Bayesian Network Model based on the Analysis of Oral Definitions of Semantic Categories

BACKGROUND

Early detection of Alzheimer's disease (AD) has become one of the principal focuses of research in medicine, particularly when the disease is incipient or even prodromic, because treatments are more effective in these stages. Lexical-semantic-conceptual deficit (LSCD) in the oral definitions of semantic categories for basic objects is an important early indicator in the evaluation of the cognitive state of patients.

OBJECTIVES

The objective of this research is to define an economic procedure for cognitive impairment (CI) diagnosis, which may be associated with early stages of AD, by analysing cognitive alterations affecting declarative semantic memory. Because of its low cost, it could be used for routine clinical evaluations or screenings, leading to more expensive and selective tests that confirm or rule out the disease accurately. It should necessarily be an explanatory procedure, which would allow us to study the evolution of the disease in relation to CI, the irregularities in different semantic categories, and other neurodegenerative diseases. On the basis of these requirements, we hypothesise that Bayesian networks (BNs) are the most appropriate tool for this purpose.

METHODS

We have developed a BN for CI diagnosis in mild and moderate AD patients by analysing the oral production of semantic features. The BN causal model represents LSCD in certain semantic categories, both of living things (dog, pine, and apple) and non-living things (chair, car, and trousers), as symptoms of CI. The model structure, the qualitative part of the model, uses domain knowledge obtained from psychology experts and epidemiological studies. Further, the model parameters, the quantitative part of the model, are learnt automatically from epidemiological studies and Peraita and Grasso's linguistic corpus of oral definitions. This corpus was prepared with an incidental sampling and included the analysis of the oral linguistic production of 81 participants (42 cognitively healthy elderly people and 39 mild and moderate AD patients) from Madrid region's hospitals. Experienced neurologists diagnosed these cases following the National Institute of Neurological and Communicative Disorders and Stroke/Alzheimer's Disease and Related Disorders Association (NINCDS-ADRDA)'s Alzheimer's criteria, performing, among other explorations and tests, a minimum neuropsychological exploration that included the Mini-Mental State Examination test.

RESULTS

BN's classification performance is remarkable compared with other machine learning methods, achieving 91% accuracy and 94% precision in mild and moderate AD patients. Apart from this, the BN model facilitates the explanation of the reasoning process and the validation of the conclusions and allows the study of uncommon declarative semantic memory impairments.

CONCLUSIONS

Our method is able to analyse LSCD in a wide set of semantic categories throughout the progression of CI, being a valuable first screening method in AD diagnosis in its early stages. Because of its low cost, it can be used for routine clinical evaluations or screenings to detect AD in its early stages. Besides, due to its knowledge-based structure, it can be easily extended to provide an explanation of the diagnosis and to the study of other neurodegenerative diseases. Further, this is a key advantage of BNs over other machine learning methods with similar performance: it is a recognisable and explanatory model that allows one to study irregularities in different semantic categories.