Search

Investigating automatic methods for the early detection of dementia and related conditions that cause cognitive impairment is an area of growing interest. Video processing could play a role by providing a non-invasive and low-cost alternative to current expensive assessments. For this to be successful it is crucial that approaches are robust to in-the-wild challenges. In this paper, visual cues, related to the eye blink rate (EBR), are investigated to quantify the early phase of neurodegenerative disorder (ND) and mild cognitive impairment (MCI) as well as functional memory disorder (FMD; problems with memory not related to neurodegenerative disorder). This paper aims to improve the detection of ND and MCI by investigating a novel approach to calculating the EBR that is more robust to in-the-wild challenges. An in-house dataset with 18 participants is used. The EBR is calculated from eye landmarks extracted using two libraries (Dlib and Openface). To mitigate issues observed in the noisy, in-the-wild recordings, a multiple threshold approach for EBR detection is proposed. It involves generating multiple thresholds for identifying a blink, where a threshold is used to determine whether an eye is open or closed. Several supervised machine learning approaches are used for automatic classification. The results show that accuracy measures of 89% and 78% are achieved using Dlib and OpenFace data, respectively, when distinguishing between three conditions with ND, MCI and FMD.

IntroductionRecent years have seen an almost sevenfold rise in referrals to specialist memory clinics. This has been associated with an increased proportion of patients referred with functional cognitive disorder (FCD), that is, non-progressive cognitive complaints. These patients are likely to benefit from a range of interventions (eg, psychotherapy) distinct from the requirements of patients with neurodegenerative cognitive disorders. We have developed a fully automated system, ‘CognoSpeak’, which enables risk stratification at the primary–secondary care interface and ongoing monitoring of patients with memory concerns.MethodsWe recruited 15 participants to each of four groups: Alzheimer’s disease (AD), mild cognitive impairment (MCI), FCD and healthy controls. Participants responded to 12 questions posed by a computer-presented talking head. Automatic analysis of the audio and speech data involved speaker segmentation, automatic speech recognition and machine learning classification.ResultsCognoSpeak could distinguish between participants in the AD or MCI groups and those in the FCD or healthy control groups with a sensitivity of 86.7%. Patients with MCI were identified with a sensitivity of 80%.DiscussionOur fully automated system achieved levels of accuracy comparable to currently available, manually administered assessments. Greater accuracy should be achievable through further system training with a greater number of users, the inclusion of verbal fluency tasks and repeat assessments. The current data supports CognoSpeak’s promise as a screening and monitoring tool for patients with MCI. Pending confirmation of these findings, it may allow clinicians to offer patients at low risk of dementia earlier reassurance and relieve pressures on specialist memory services.

Speech-based automatic approaches for detecting neurodegenerative disorders (ND) and mild cognitive impairment (MCI) have received more attention recently due to being non-invasive and potentially more sensitive than current pen-and-paper tests. The performance of such systems is highly dependent on the choice of features in the classification pipeline. In particular for acoustic features, arriving at a consensus for a best feature set has proven challenging. This paper explores using deep neural network for extracting features directly from the speech signal as a solution to this. Compared with hand-crafted features, more information is present in the raw waveform, but the feature extraction process becomes more complex and less interpretable which is often undesirable in medical domains. Using a SincNet as a first layer allows for some analysis of learned features. We propose and evaluate the Sinc-CLA (with SincNet, Convolutional, Long Short-Term Memory and Attention layers) as a task-driven acoustic feature extractor for classifying MCI, ND and healthy controls (HC). Experiments are carried out on an in-house dataset. Compared with the popular hand-crafted feature sets, the learned task-driven features achieve a superior classification accuracy. The filters of the SincNet is inspected and acoustic differences between HC, MCI and ND are found.

BACKGROUND: The early diagnosis of dementia is of great clinical and social importance. A recent study using the qualitative methodology of conversation analysis (CA) demonstrated that language and communication problems are evident during interactions between patients and neurologists, and that interactional observations can be used to differentiate between cognitive difficulties due to neurodegenerative disorders (ND) or functional memory disorders (FMD). OBJECTIVE: This study explores whether the differential diagnostic analysis of doctor-patient interactions in a memory clinic can be automated. METHODS: Verbatim transcripts of conversations between neurologists and patients initially presenting with memory problems to a specialist clinic were produced manually (15 with FMD, and 15 with ND). A range of automatically detectable features focusing on acoustic, lexical, semantic, and visual information contained in the transcripts were defined aiming to replicate the diagnostic qualitative observations. The features were used to train a set of five machine learning classifiers to distinguish between ND and FMD. RESULTS: The mean rate of correct classification between ND and FMD was 93% ranging from 97% by the Perceptron classifier to 90% by the Random Forest classifier.Using only the ten best features, the mean correct classification score increased to 95%. CONCLUSION: This pilot study provides proof-of-principle that a machine learning approach to analyzing transcripts of interactions between neurologists and patients describing memory problems can distinguish people with neurodegenerative dementia from people with FMD.

The diagnosis of Mild Cognitive Impairment (MCI) characterises patients at risk of dementia and may provide an opportunity for disease-modifying interventions. Identifying persons with MCI (PwMCI) from adults of a similar age without cognitive complaints is a significant challenge. The main aims of this study were to determine whether generic speech differences were evident between PwMCI and healthy controls (HC), whether such differences were identifiable in responses to recent or remote memory questions, and to determine which speech variables showed the clearest between-group differences. This study analysed recordings of 8 PwMCI (5 females, 3 males) and 14 HC of a similar age (8 females, 6 males). Participants were recorded interacting with an intelligent virtual agent: a computer-generated talking head on a computer screen which asks pre-recorded questions when prompted by the interviewee through pressing the next key on a computer keyboard. Responses to recent and remote memory questions were analysed. Mann-Whitney U tests were used to test for statistically significant differences between PwMCI and HC on each of 12 speech variables, relating to temporal characteristics, number of words produced and pitch. It was found that compared to HC, PwMCI produce speech for less time and in shorter chunks, they pause more often and for longer, take longer to begin speaking and produce fewer words in their answers. It was also found that the PwMCI and HC were more alike when responding to remote memory questions than when responding to recent memory questions. These findings show great promise and suggest that detailed speech analysis can make an important contribution to diagnostic and stratification systems in patients with memory complaints.

Background: Individuals living with long-term physical health conditions frequently experience co-occurring mental health problems. This comorbidity has a significant impact on an individual’s levels of emotional distress, health outcomes, and associated health care utilization. As health care services struggle to meet demand and care increasingly moves to the community, digital tools are being promoted to support patients to self-manage their health. One such technology is the autonomous virtual agent (chatbot, conversational agent), which uses artificial intelligence (AI) to process the user’s written or spoken natural language and then to select or construct the corresponding appropriate responses. Objective: This study aimed to co-design the content, functionality, and interface modalities of an autonomous virtual agent to support self-management for patients with an exemplar long-term condition (LTC; chronic pulmonary obstructive disease [COPD]) and then to assess the acceptability and system content. Methods: We conducted 2 co-design workshops and a proof-of-concept implementation of an autonomous virtual agent with natural language processing capabilities. This implementation formed the basis for video-based scenario testing of acceptability with adults with a diagnosis of COPD and health professionals involved in their care. Results: Adults (n=6) with a diagnosis of COPD and health professionals (n=5) specified 4 priority self-management scenarios for which they would like to receive support: at the time of diagnosis (information provision), during acute exacerbations (crisis support), during periods of low mood (emotional support), and for general self-management (motivation). From the scenario testing, 12 additional adults with COPD felt the system to be both acceptable and engaging, particularly with regard to internet-of-things capabilities. They felt the system would be particularly useful for individuals living alone. Conclusions: Patients did not explicitly separate mental and physical health needs, although the content they developed for the virtual agent had a clear psychological approach. Supported self-management delivered via an autonomous virtual agent was acceptable to the participants. A co-design process has allowed the research team to identify key design principles, content, and functionality to underpin an autonomous agent for delivering self-management support to older adults living with COPD and potentially other LTCs.

The ageing population has caused a marked increased in the number of people with cognitive decline linked with dementia. Thus, current diagnostic services are overstretched, and there is an urgent need for automating parts of the assessment process. In previous work, we demonstrated how a stratification tool built around an Intelligent Virtual Agent (IVA) eliciting a conversation by asking memory-probing questions, was able to accurately distinguish between people with a neuro-degenerative disorder (ND) and a functional memory disorder (FMD). In this paper, we extend the number of diagnostic classes to include healthy elderly controls (HCs) as well as people with mild cognitive impairment (MCI). We also investigate whether the IVA may be used for administering more standard cognitive tests, like the verbal fluency tests. A four-way classifier trained on an extended feature set achieved 48% accuracy, which improved to 62% by using just the 22 most significant features (ROC-AUC: 82%).

Objectives/AimsWe used our automated cognitive assessment tool to explore whether responses to questions probing recent and remote memory could aid in distinguishing between patients with early neurodegenerative disorders and those with Functional Cognitive Disorders (FCD).Hypotheses: pwFCD would have no significant differences in pause to speech ratio and measures of linguistic complexity compared to healthy controls. pwFCD would have significant differences in pause to speech ratio and measures of linguistic complexity compared to pwMCI and pwAD.MethodsWe recruited 15 participants with FCD, MCI and AD each as well as 15 healthy controls. Participants answered 12 questions posed by the ‘Digital Doctor’. Automatic processing of the audio-recorded answers involved automatic speech recognition including detecting length of pauses. Two questions probe recent memory, exploring knowledge of current affairs. Two probe remote memory, asking for autobiographical details.We analysed the data using: Pause to speech time ratio. Moving average type token ratio (MATTR): An automated measure of vocabulary richness. Computerised propositional idea density rater (CPIDR): An automated measure of propositional idea density.ResultsThere was a significant difference in the pause to speech ratio for recent memory questions for HC versus AD (P=0.0012) and MCI (pConclusionsThis study rejects both hypotheses. However, the data supports the application of linguistic measures to recent and remote memory questions in distinguishing those with MCI & AD from HC’s. Further work will investigate the utility of incorporating additional measures of lexical and grammatical complexity (word frequency, sentence structure). Longitudinal study will provide insights into which features may predict stability in FCD and HC’s and progression from MCI to AD, supporting the system’s promise as a monitoring tool.

Previous work on interactions in the memory clinic has shown that conversation analysis can be used to differentiate neurodegenerative dementia from functional memory disorder. Based on this work, a screening system was developed that uses a computerised ‘talking head’ (intelligent virtual agent) and a combination of automatic speech recognition and conversation analysis-informed programming. This system can reliably differentiate patients with functional memory disorder from those with neurodegenerative dementia by analysing the way they respond to questions from either a human doctor or the intelligent virtual agent. However, much of this computerised analysis has relied on simplistic, nonlinguistic phonetic features such as the length of pauses between talk by the two parties. To gain confidence in automation of the stratification procedure, this paper investigates whether the patients’ responses to questions asked by the intelligent virtual agent are qualitatively similar to those given in response to a doctor. All the participants in this study have a clear functional memory disorder or neurodegenerative dementia diagnosis. Analyses of patients’ responses to the intelligent virtual agent showed similar, diagnostically relevant sequential features to those found in responses to doctors’ questions. However, since the intelligent virtual agent’s questions are invariant, its use results in more consistent responses across people – regardless of diagnosis – which facilitates automatic speech recognition and makes it easier for a machine to learn patterns. Our analysis also shows why doctors do not always ask the same question in the exact same way to different patients. This sensitivity and adaptation to nuances of conversation may be interactionally helpful; for instance, altering a question may make it easier for patients to understand. While we demonstrate that some of what is said in such interactions is bound to be constructed collaboratively between doctor and patient, doctors could consider ensuring that certain, particularly important and/or relevant questions are asked in as invariant a form as possible to be better able to identify diagnostically relevant differences in patients’ responses.

This paper presents work on developing an automatic dementia screening test based on patients’ ability to interact and communicate — a highly cognitively demanding process where early signs of dementia can often be detected. Such a test would help general practitioners, with no specialist knowledge, make better diagnostic decisions as current tests lack specificity and sensitivity. We investigate the feasibility of basing the test on conversations between a ‘talking head’ (avatar) and a patient and we present a system for analysing such conversations for signs of dementia in the patient’s speech and language. Previously we proposed a semi-automatic system that transcribed conversations between patients and neurologists and extracted conversation analysis style features in order to differentiate between patients with progressive neurodegenerative dementia (ND) and functional memory disorders (FMD). Determining who talks when in the conversations was performed manually. In this study, we investigate a fully automatic system including speaker diarisation, and the use of additional acoustic and lexical features. Initial results from a pilot study are presented which shows that the avatar conversations can successfully classify ND/FMD with around 91% accuracy, which is in line with previous results for conversations that were led by a neurologist.

A recent study using Conversation Analysis (CA) has demonstrated that communication problems may be picked up during conversations between patients and neurologists, and that this can be used to differentiate between patients with (progressive neurodegenerative dementia) ND and those with (nonprogressive) functional memory disorders (FMD). This paper presents a novel automatic method for transcribing such conversations and extracting CA-style features. A range of acoustic, syntactic, semantic and visual features were automatically extracted and used to train a set of classifiers. In a proof-of-principle style study, using data recording during real neurologist-patient consultations, we demonstrate that automatically extracting CA-style features gives a classification accuracy of 95%when using verbatim transcripts. Replacing those transcripts with automatic speech recognition transcripts, we obtain a classification accuracy of 79% which improves to 90% when feature selection is applied. This is a first and encouraging step towards replacing inaccurate, potentially stressful cognitive tests with a test based on monitoring conversation capabilities that could be conducted in e.g. the privacy of the patient’s own home.

Objects and Aims Referrals to secondary care memory clinic has more than tripled. This has led to increased demand on diagnostic services. Conversation Analysis (CA) can help in the screening for dementia. In CA important features of the conversation between doctor, patient are analysed. In this project we created an avatar to automatically initiate conversation. Methods The avatar, a computerised head asked questions created by consulting neurologists to which patients responded verbally. Audio and video data was collected and a simple interface with two buttons (‘start/forward’ and ‘repeat’) on a keyboard was used. Audio recordings were transcribed and annotated by the CA expert. Results 24 participants were recruited. All finished the trial without problems. We received feedback from participants, many saying they felt the avatar was easy to talk to. The CA expert correctly classified in 6 of 7 cases of Functional Memory disorder (FMD) and 4 out of 6 cases of neurodegenerative dementia. Conclusion An Avatar has potential to be a low cost addition to memory screening

Speech and language based automatic dementia detection is of interest due to it being non-invasive, low-cost and potentially able to aid diagnosis accuracy. The collected data are mostly audio recordings of spoken language and these can be used directly for acoustic-based analysis. To extract linguistic-based information, an automatic speech recognition (ASR) system is used to generate transcriptions. However, the extraction of reliable acoustic features is difficult when the acoustic quality of the data is poor as is the case with DementiaBank, the largest opensource dataset for Alzheimer's Disease classification. In this paper, we explore how to improve the robustness of the acoustic feature extraction by using time alignment information and confidence scores from the ASR system to identify audio segments of good quality. In addition, we design rhythm inspired features and combine them with acoustic features. By classifying the combined features with a bidirectional-LSTM attention network, the F-measure improves from 62.15% to 70.75% when only the high-quality segments are used. Finally, we apply the same approach to our previously proposed hierarchical-based network using linguistic-based features and show improvement from 74.37% to 77.25%. By combining the acoustic and linguistic systems, a state-of-the-art 78.34% F-measure is achieved on the DementiaBank task.