Saturnino Luz
Professor of Digital Biomarkers and Precision Medicine
- Usher Institute
- College of Medicine and Veterinary Medicine
Contact details
Address
- Street
-
Centre for Medical Informatics
Usher Institute, Usher Building
The University of Edinburgh
5-7 Little France Road
Edinburgh BioQuarter ‒ Gate 3 - City
- Edinburgh
- Post code
- EH16 4UX
Background
Saturnino works in medical informatics, devising and applying machine learning, signal processing, natural language processing and dimensionality reduction methods in the study of behaviour and communication in healthcare contexts. His main research interest is the computational modelling of behavioural and biological changes caused by neurodegenerative diseases, with focus the analysis of vocal and linguistic signals in Alzheimers's disease. He has also employed these methods in the investigation of interaction in multidisciplinary medical team meetings, doctor-patient consultations, telemedicine and health promotion.
Qualifications
PhD (Informatics), MSc, BSc (Computer Science)
Responsibilities & affiliations
Saturnino is based in the Centre for Medical Informatics and is also a member of the Centre for Global Health.
He is also a member of the Scottish Dementia Research Consortium.
Postgraduate teaching
Saturnino is currently recruiting a PhD student and interested to hear from those considering a PhD in any areas relevant to his research.
Open to PhD supervision enquiries?
Yes
Current PhD students supervised
- Sofia de la Fuente Garcia: "Human-Robot Interaction for Dementia Prevention and Research with a Focus on Cognitive Well-Being". MRC Precision Medicine Doctoral Training Programme.
- Pierre Albert: "Interaction Analytics for Automatic Assessment of Communication Quality in Primary Care", MRC Precision Medicine Doctoral Training Programme.
- Minhong Wang: Agent state based modelling of pattern formation in pluripotent stem cells
Research summary
Saturnino's research seeks to harness the power of ubiquitous digital technology for the creation of novel biomarkers for objective, scalable and cost-effective measurement of physiology and behaviour within the broad field of precision medicine. This research has the potential to deliver meaningful impact on health care in Scotland and the UK, and to revolutionise care in low- and middle-income countries. While he has investigated methods and applications of digital phenotyping in several areas, his main focus has been on digital biomarkers of neurodegenerative diseases. He has conducted analyses of dementia data, including novel digital (behavioural) biomarkers that can be collected frequently, unobtrusively, and at scale through mobile and ambient technology. His lab have developed novel methods for the analysis of bioacoustical markers for detection and assessment of progression of Alzheimer’s dementia and other conditions. These models have achieved state-of-the-art categorisation results for Alzheimer's detection, reaching approximately 93% accuracy in monologue data. The Luz Lab's language-independent dialogue models reach 89% accuracy using acoustic features only.
Prof Luz has also led the development of methodology, shared data sets and resources for the assessment of voice, speech and language biomarkers. As shown by a systematic review he conducted recently, research in this area has grown considerably in the last few years. However, the field remains fragmented, and adequate assessment of the different approaches and ultimately translation to clinical practice, is hindered by a lack of shared data sets and poor standardisation of modelling and evaluation methods. To address this issue, he created, in cooperation with Prof Brian MacWhinney and colleagues at Carnegie Mellon University and Edinburgh, the first international shared machine learning task on dementia detection and assessment, the ADReSS (Alzheimer's Dementia Recognition from Spontaneous Speech) Challenge. ADReSS provided acoustically normalised, pre-processed, longitudinal spontaneous speech data sets, matched for gender and age, and a platform for evaluation of machine learning models for discriminative (Alzheimer’s detection and prediction of progression from mild cognitive impairment to dementia) and regression tasks (prediction of neuropsychological test scores). This and subsequent shared signal processing and machine learning tasks have attracted many participating teams from the world's top academic and industrial laboratories.
Research on digital biomarkers is particularly promising in relation to global health. Prof Luz's research in this field has opened new avenues for the deployment of low-cost devices for health monitoring in low- and middle-income countries (as well as in high income countries). He led, for instance, the advanced technologies work package of the EU-funded SAAM project, which investigated the use of data extracted from smart electrical meters and ambient sensors for monitoring the physical and mental wellbeing of older people living independently or in assisted living care, in low-income communities in Bulgaria. This work was done in cooperation with local community workers of the Bulgarian Red Cross and Caritas. The Edinburgh team developed ambient hardware for voice, temperature, gait and gesture data collection, which we incorporated into a mental wellbeing model. Adapted versions of this model have been used since in several predictive models for depression and mood assessment based on acoustic features extracted from speech. In relation to dementia, more specifically, these digital technologies have great potential to foster the development of low-cost, scalable and accessible tools for monitoring of cognitive function, dementia screening, and support for community-based prevention and care in low- and middle-income countries.
Current research interests
Saturnino is currently investigating the integration of digital and conventional biomarkers into predictive and explanatory models of neurodegenerative diseases. He is also interested in elucidating the neurological mechanisms underpinning behavioural changes that constitute early signs of Alzheimer's disease. He is currently leading the SIDE-AD project, "Speech for intelligent cognition change tracking and detection of AD (SIDE-AD)", funded by the Sony Research Award Programme (Focused Research Award).Past research interests
In early work, Saturnino pioneered the use of AI and computer-supported cooperative work (CSCW) research methods in the investigation of communication and interaction in healthcare settings. In particular, together with collaborators, he investigated communication in multidisciplinary medical team (MDT) meetings, doctor-patient consultations, telemedicine and health promotion in low- and middle-income countries. This research built on his research (with Bridget Kane) on the analysis of MDT meetings from a computer-supported cooperative work perspective, where we introduced the concept of an MDTM system as the core of coordinative processes in MDT work, encompassing pre- and post-meeting activities and their relationships with broader aspects of patient care. At the University of Edinburgh, he collaborated with the TeleScot research group, led by Prof Brian McKinstry, in the context of the INCA project ("Interaction Analytics for Automatic Assessment of Communication Quality in Primary Care", PI: Luz, 2016-2020) and the ViCo study (PI: McKinstry).Knowledge exchange
Highlights of Saturnino's knowledge exchange activities include:
- Four invention disclosures currently being assessed by Edinburgh Innovation for patenting and commercial exploitation
- Track record of productive engagement with industry and third-sector organisations for promoting societal and economic impact of my research lab’s output
- Membership of high profile expert evaluation panels, including EU and WHO
- Leading role in open research and open source software projects, with global impact
Prof Luz has led the development of novel methods for the analysis of time series, including a new method for feature extraction from high-dimensional, time-based data streams, called Active Data Representation (ADR) which we have employed in the analysis of speech, accelerometer and EEG signals, a method for prediction of cognitive decline through analysis of dialogue data (VOCALDIA), a method for mood assessment using multi-resolution cochleagram features, and a method for EEG signal analysis coupled with a machine learning model for REM sleep behaviour disorder prediction with potential applications to Parkinson’s disease detection.
Project activity
- Ongoing projects:
-
SIDE-AD, Speech for intelligent cognition change tracking and detection of Alzheimer's Dementia, Role: Edinburgh lead and PI. Funded by the Sony Research Award Programme (Focused Research Award). This project will carry out research and development of a speech-based application for the collection of brain health priorities which are specific and meaningful to the individual using the tool, and for the analysis of speech data in combination with routinely available clinical data. This will enable real world validation of speech-based digital biomarkers in individuals with early Alzheimer’s disease (AD). The research programme will employ machine learning and natural language processing technology to automate the assessment of the respondents’ speech patterns indicative of brain health status relevant to emergent AD-specific speech biomarkers. We will recruit patients from NHS Scotland Brain Health Services allowing the voice data to be analysed with respect to routinely available clinical and biomarker data collected on these patients, including neurodegenerative disease diagnosis; clinical outcomes (activities of daily living, mood/anxiety and sleep measures; neuropsychological outcomes (cognitive tests); neuroimaging outcomes and biomarker outcomes: genetic risk (ApoE4), amyloid and tau levels.
-
TAUKADIAL (TAiwan-UK Alzheimer's DIALogues), multilingual approaches for early Alzheimer's disease detection through analysis of spontaneous speech. Role: Lead PI. Funding Leverhulme Fund/British Academy. There is growing interest in using speech data for early detection based on computational linguistics methods. While various models have been proposed and applied in different language systems, progress has been hampered by (a) scarcity of speech datasets from clinical and preclinical AD patients; (b) lack of systematic multilingual studies aimed at identifying linguistic markers of AD that generalise across languages. This project will tackle these issues by developing dialogue protocols for collecting data from spontaneous speech conversations, and systematically comparing linguistic features derived from English and Chinese the detection of AD. Specifically, it will investigate network analysis methods that have been applied to English but not yet to Chinese data. In collaboration with colleagues from the Cardinal Tien Hospital, and the National Cheng Kung University, Taiwan, Prof Luz's lab is investigating these issues by developing dialogue protocols for collecting data from spontaneous speech conversations, and systematically comparing linguistic features derived from English and Chinese for the detection of AD. Specifically, this project will explore network analysis methods that have been applied to English but not yet to Chinese data.
-
OMC, Oslo Medical Corpus. Edinburgh lead investigator. Funded by the Norwegian Research Council (Collaboration with University of Oslo Medical School). This project has developed a corpus of medical and public health texts for use by medical humanities scholars and for medical education. The OMC has been used by the Univerity of Oslo’s Centre for Sustainable Healthcare Education (SHE) in a new course targeting the training needs of teachers in master’s degree programs and the professional program in medicine. The aim is to encourage and strengthen teachers’ qualifications in critical concept analysis by using the OMC to engage students in reflecting critically on key concepts such as health, illness, disease, sustainability, empowerment, health literacy, user participation, evidence, and expertise. This corpus and its tools have also been used in one PhD thesis and two master’s theses, and work is underway to develop an online teaching platform featuring OMC as the central resource for students to work on.
-
- Past project (recent):
- SAAM: Supporting Active Ageing through Multimodal coaching, EU Horizon 2020, Personalised Medicine; 2017-2020, €3,996,400. The SAAM project focused on innovative, technology-enabled approaches to support the aging population living at home, with a novel and practical emphasis on ambient sensing and learning of user needs and preferences, and effective coaching by leveraging the user’s social support networks. Over three years, the project’s partners will develop and test new methods allowing Europe’s aging population to remain active, independent and longer in their homes.
-
EMBEDDIA: "Cross-Lingual Word Embeddings for NLP in Less Represented Languages". Role: Edinburgh lead and PI. Funded by the EC Horizon 2020.
- INCA: Interaction Analytics for Automatic Assessment of Communication Quality in Primary Care. Health Research BoardGoK, The INCA project investigates technological mechanisms for electronic gathering and automated analysis of physician-patient interaction during medical interviews. It applies state of the art technology in speech processing, text analytics and social signal processing, and investigate the impact of models through which comprehensive, data-intensive communication analysis could be conducted. This interaction analytics research will use routinely collected audio-visual data from consultations between patients and trainee general practitioners.
- GoK, Genealogies of Knowledge: Evolution and Contestation of Concepts across Time and Space. AHRC, 2016-2020, £796,000. This project explores how our understanding of central scientific and cultural concepts has evolved since they first emerged. The project draws on a specially constructed electronic corpus of translations and recensions of key texts, both in the humanities and the sciences (with specific emphasis on medicine). Dr Luz leads the Natural Language Processing and Text Visualisation work packages.