Artificial Intelligence and Computer Vision for Neurodegenerative Diseases Assessment: Advancing Computer Science in Dementia and Neurodegenerative Disorders
24th November 2023, King's College, University of Aberdeen, UK
24th November 2023, King's College, University of Aberdeen, UK
BIO:
Dr. Wenhao Zhang is Associate Professor of Computer Vision and Machine Learning at the University of the West of England, Bristol. He is co-director of the Centre for Machine Vision, Bristol Robotics Laboratory. Wenhao specialises in machine learning, machine vision, and their applications in interdisciplinary fields, particularly in health technology and agriculture technology. Over the past five years, Wenhao has been extensively involved in more than ten funded projects, serving as principal investigator, co-investigator, or key researcher. Wenhao has current research interest in developing accessible, convenient, and pervasive eye tracking for early diagnosis of neurodegeneration and eye diseases.
Talk:
Evidence suggests that executive attention is one of the first non-memory domains to be affected in Mild Cognitive Impairment (MCI) that may precede dementia, and this may even occur in the preclinical stage, when certain biomarker changes are present but clinical symptoms have not yet developed. As a result, eye tracking technology has recently emerged as an area of interest in differentially detecting various neurodegenerative disorders. However, various challenges have prevented such technology from being clinically ready.
Research-grade eye trackers are often prohibitively expensive for many users. In comparison, consumer-grade eye trackers can be relatively inexpensive but often lack the desired accuracy and precision to capture subtle eye movements. In addition, eye tracking tests, such as those involving anti-saccades and smooth pursuit, can be counterintuitive and demotivating, resulting in inconsistent and sometimes conflicting results across studies in the literature. The analysis of eye tracking data frequently relies on conventional statistical techniques applied to human-engineered features such as saccade velocity and error rate. In contrast, the utilisation of machine learning and computer vision, particularly deep learning neural networks, remains relatively uncommon. Therefore, this presentation introduces an inexpensive and convenient approach to eye tracking, outlines the typical challenges encountered in various stages of eye tracking tests, and explores the potential opportunities and roles of machine learning and computer vision.
BIO:
Husnu Baris Baydargil is a postdoctoral research associate at the Institute for Basic Science in Daejeon, Republic of Korea. He finished his master's and Ph.D. at Kyungsung University Department of Electrical and Communication Engineering in Busan, Republic of Korea, focusing mainly on computer vision and neurodegenerative diseases such as Alzheimer's Disease. He is primarily interested in combining the power of state-of-the-art deep learning approaches and medical imaging to shed new light on early and more accurate identification and a streamlined progression monitoring of neurodegenerative diseases for better treatment options.
Talk:
In this era of ever-improving generative AI, synthetic medical images seem like a promising solution to data shortages. However, are they reliable and realistic enough to be used as auxiliary data in medical imaging? We delve into this pressing question, focusing on Alzheimer's disease as a case study. While synthetic images created using reconstruction methods by state-of-the-art generative models might look realistic at first glance, they often fail to capture the full complexity and the heterogeneity of real 3D MRI scans. We also investigate the inconsistencies in training these generative models regarding Alzheimer's disease severity and subtypes, which pose challenges. This can lead to errors when identifying different atrophy patterns, causing the deep-learning models to have impaired performance. We also find that using noise-generated, synthetic data causes models to perform even more poorly, massively overfitting or underfitting, depending on the generative model used. Our research serves as a cautionary tale, emphasizing the need for rethinking the training of generative models for optimized data generation. We hope that our findings are not just relevant for Alzheimer's but will also have implications for studying other neurodegenerative diseases, like Parkinson's and Multiple Sclerosis
BIO:
Luigi A. Moretti is a medical doctor (MD) by training with a multidisciplinary bent. His mindset has allowed him to build a career on the ability to combine the skills and perspectives of medicine and technology. After several first-hand experiences in the development of innovative digital products and an MSc in Health Technology, he is now sponsored by the University of the West of England (UWE) to conduct a research for the co-design, with patients and clinicians, of an affective computing-based solution for mental health, and more specifically for anxiety disorders. His aim is to collect objective, interpretable and continuous data in the medium/long term to support patient awareness and engagement, and to provide clinicians with an additional source of information to use in parallel with standard approaches (e.g. self-report surveys).
Talk:
Speech title: Emotion recognition can support the early detection of neurodegenerative diseases
Neurodegenerative diseases (NDDs), such as Alzheimer's disease, frontotemporal dementia, dementia with Lewy bodies, and Huntington's disease, inevitably lead to impairments in higher-order cognitive functions, and this includes emotional changes, mostly, but not only, in emotional perception [1] and psychological behaviour [2]. Anxiety is a common symptom in several NDDs, with prevalence estimates ranging from 5% to 21% for anxiety disorders and from 8% to 71% for anxiety symptoms [3]. In addition, studying emotion patterns in patients with NDDs may also help to better understand the biological changes brought about by the diseases. Meanwhile, studying emotion variations in patients with known biological neurogenerative changes may help to better understand the physiology of emotion generation and patterns [4]. Emotion recognition can support the early detection of neurodegenerative diseases, as this must be a multidisciplinary and multimodal approach, in fact it is suggested "an urgent call for refinement of assessment and intervention tools with full consideration of the multifaceted nature of these disorders in a naturalistic and timely manner, by adopting behavioural quantification. It is now known that a continuum of non-specific neurodegenerative manifestations constitutes the prodromal stage of neurodegenerative diseases" [5].
BIO:
I am a seasoned professional in psychology and neuroscience, boasting a successful track record in research and teaching. My journey began with a Bachelor's degree in Psychology from the University of Padua in 2001, followed by the attainment of my professional psychologist certification in 2002. In 2010, I achieved a Ph.D. in Experimental Neurobiology from the University of Bari. I currently serve as an Associate Professor in General Psychology, where I bring extensive experience in both academic instruction and research.
My research predominantly revolves around neuroscience, emotions' genetic underpinnings, and psychiatric disorders. I have authored numerous scientific publications and actively contributed to national and international research initiatives.
In recent years, my research has pivoted towards investigating cognitive decline and aging trajectories. I also explore the application of assistive technologies in addressing both normal and pathological cognitive decline. I play an active role in national and international research projects within these domains. My clinical background as a psychologist and psychotherapist bolsters my research efforts. Additionally, I collaborate closely with the Department of Computer Science to integrate artificial intelligence into my research, allowing me to harness AI's potential in advancing our understanding of cognitive decline and aging.
I have earned recognition as a prominent figure in the field of neuroscience, consistently making noteworthy contributions to the study of cognitive processes and aging through innovative research approaches.
Talk
Title: Artificial Intelligence-Driven Advancements in Understanding and Enhancing Cognitive Well-Being During Aging
Paolo Taurisano, Daphne Gasparre, Chiara Abbatantuono, Alessia Monaco, Vincenzo Dentamaro, Donato Impedovo
Introduction
The process of aging is something that happens toeveryone and involves various changes in the body and mind. Thesechanges occur gradually over a person's lifetime and can affectdifferent aspects of their health. For example, the aging process canlead to a decline in physical strength, changes in the heart and bloodvessels, and alterations in how the brain functions. Specifically, when itcomes to the brain, aging can impact cognitive abilities such as memory, attention, problem-solving, and how quickly information is processed.
The Role of Cognitive Decline in Aging
The Impact of Cognitive Decline on Aging Cognitive decline is a natural part of the aging process and refers to a decrease in complex cognitive abilities, such as memory, concentration, executive function, and decision-making skills. This decline can take different forms, ranging from mild changes that are considered normal for aging, known as "mild cognitive impairment," to more severe forms associated with neurodegenerative disorders like Alzheimer's disease and other dementias. Cognitive decline is a major public health issue as it can have a detrimental effect on the quality of life and independence of older individuals.
Our Research Focus
Our research group is focused on addressing the important issues of aging and cognitive decline. We are dedicated to developing new and innovative approaches to primary prevention. We understand the importance of fully understanding these concepts in order to create effective strategies for preventing and intervening in the negative effects of aging and cognitive decline. Our goal is to promote cognitive well-being in aging populations.
Integration of Assistive Technologies and AI
We are currently working on integrating advanced technologies, including artificial intelligence (AI), into our research efforts. Our focus is on understanding the complexities of the aging process and cognitive decline, and AI can provide valuable tools for investigating these phenomena and developing innovative therapies. To achieve this, we are utilizing wearable devices driven by AI that can continuously monitor various physiological, behavioral, and environmental factors. By combining these devices with AI analysis, we can gain unique insights and create personalized solutions to improve the cognitive well-being of individuals in aging populations.
Conclusion
In conclusion, our research group takes an interdisciplinary approach by combining expertise from the fields of health, AI, and device engineering. This approach shows potential in tackling the difficulties associated with aging and cognitive decline. By seamlessly integrating advanced technologies and having a deep understanding of the aging process, we aim to redefine primary prevention. Our goal is to provide personalized, data-driven solutions that enhance the quality of life, autonomy, and cognitive well-being of older individuals. Ultimately, our work contributes to promoting healthier and more fulfilling aging experiences.
BIO:
Liziane Bouvier is an assistant professor at McGill University (Montreal, Canada) and a trained speech-language pathologist. She has completed a Master in Speech-Language Pathology and a PhD in Biomedical and Clinical Sciences at Université Laval (Quebec City, Canada). She has also completed a postdoctoral fellowship at University of Toronto (Toronto, Canada). Her primary research goals are to improve the understanding of and contribute to clinical advances in neurological diseases causing motor speech disorders. Her research program includes the identification of acoustic biomarkers for the early diagnosis and monitoring of bulbar dysfunction in ALS, as well as the development of tools for the assessment of motor speech disorders, particularly in Canadian French speakers. Her research is strongly influenced by her linguistics and clinical trainings, from her research questions to the development of AI models
Talk
TITLE: Clinical and linguistic considerations in the development of AI models for amyotrophic lateral sclerosis.
ABSTRACT: Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease marked by the degradation of upper and lower motor neurons. Approximately 80% of the patients experience complications with the bulbar system, leading to speech, swallowing, and salivation deficits. Speech decline is one of the earliest indicators of bulbar motor system involvement, making acoustic biomarkers a promising solution for early detection and monitoring of the disease. While showing encouraging results, the current studies present some limitations for clinical application: 1) predominant focus on a binary classification between symptomatic bulbar ALS and healthy controls (i.e., presence or absence of disease); 2) sparce application to features easily collected in clinical settings (e.g., timing measures); 3) lack of physiological correlates of the selected features; 4) most importantly, an overwhelming concentration on English-speaking patients. This presentation will review important criteria for interpretability of AI models in health care settings, focussing on clinical and linguistic considerations. The presentation will by supported by examples from a recent study from our lab applying machine learning models to the early detection of bulbar symptoms in Canadian French speakers with ALS. We provide a transparent and clinically understandable model. Our results highlight an important clinical consideration, which is the need to select different features according to aim of the model and the stage of the disease. Future research should prioritize larger sample sizes to corroborate these findings. The use of linguistically and clinically adapted models should deepen our insights into early bulbar dysfunction detection in Canadian French speakers with ALS .
BIO:
Mah Parsa is a postdoctoral researcher at CRIM focusing on developing and deploying machine learning based speech and language assessment methods for neurodegenerative disease (e.g., Alzheimer's disease) and psychiatric disorders (e.g., schizophrenia).
Talk
TITLE: AI-based assessments of speech and language impairments in dementia
Recent advancements in artificial intelligence (AI) have transformed the landscape of early cognitive impairment detection linked to dementia. This shift has spurred clinicians to embrace AI-driven systems, especially those that analyze and learn speech and language patterns, to swiftly and accurately pinpoint individuals with dementia. This presentation outlines an approach to crafting an AI-based assessment for identifying patients with dementia from their speeches and languages and showcases initial findings while shedding light on AI developers' challenges in deploying these systems within dementia care settings.