Privacy-Preserved Human Motion Analysis for Healthcare Applications
Lead Research Organisation:
University of Glasgow
Department Name: School of Computing Science
Abstract
Human motion analysis is a powerful tool in healthcare applications as it has shown to be effective in providing disease progression markers in neurodegenerative conditions such as Alzheimer's, Parkinson, Amyotrophic Lateral Sclerosis, Huntington's disease and dementia. On the other hand, deep learning in human motion analysis has shown impressive results in human pose tracking in real-time. This technology can empower patients to have an active role in managing their condition(s), which is a significant objective in a growing e-Health (digital Health) era. Opportunities in digital health initiatives have increased through the response to the pandemic and it has become evident of the need for an intelligent system to detect abnormal changes in patient gait patterns and subsequently alert carers. This technology can also prevent further deterioration (multimorbidity) due to the associated risk of falls and mood disorders.
However, translating recent advances in computer vision in home care is challenging for three major reasons: data privacy, lack of large healthcare labelled data and reduced data quality. This project proposes that data privacy and ethics should be encoded in the algorithms early in the pipeline so that systems are resilient to attacks and do not compromise real-time interaction. We argue that this approach could also improve the performance of the machine learning models with small datasets by focusing on the most relevant features in a data-driven way. Furthermore, we propose that coupling this technology with synthetic data generation can significantly boost the development of ambient sensing technologies for human motion tracking in healthcare applications and develop technology viable for the UK market.
However, translating recent advances in computer vision in home care is challenging for three major reasons: data privacy, lack of large healthcare labelled data and reduced data quality. This project proposes that data privacy and ethics should be encoded in the algorithms early in the pipeline so that systems are resilient to attacks and do not compromise real-time interaction. We argue that this approach could also improve the performance of the machine learning models with small datasets by focusing on the most relevant features in a data-driven way. Furthermore, we propose that coupling this technology with synthetic data generation can significantly boost the development of ambient sensing technologies for human motion tracking in healthcare applications and develop technology viable for the UK market.
Publications
Gu X
(2024)
Beyond Supervised Learning for Pervasive Healthcare.
in IEEE reviews in biomedical engineering
Jacobs E
(2024)
Threat Perception Captured by Emotion, Motor and Empathetic System Responses: A Systematic Review
in IEEE Transactions on Affective Computing
Lai-Tan N
(2023)
Toward Personalized Music-Therapy: A Neurocomputational Modeling Perspective
in IEEE Pervasive Computing
Malek-Podjaski M
(2022)
Adversarial Attention for Human Motion Synthesis
Malek-Podjaski M
(2023)
Adversarial Attention for Human Motion Synthesis
| Description | Industrial CASE Account - University of Glasgow 2024 |
| Amount | £392,000 (GBP) |
| Funding ID | EP/Z531042/1 |
| Organisation | University of Glasgow |
| Sector | Academic/University |
| Country | United Kingdom |
| Start | 09/2024 |
| End | 09/2029 |
| Description | Student travel award |
| Amount | £1,000 (GBP) |
| Organisation | British Machine Vision Association |
| Sector | Charity/Non Profit |
| Country | United Kingdom |
| Start | 11/2023 |
| End | 11/2023 |
| Title | Adversarial Attention for Human Motion Synthesis |
| Description | We have developed a novel method to synthesise human motion based on Adversarial Attention. Human motion synthesis is a crucial research problem within deep learning and computer vision, since acquiring human motion datasets is highly time-consuming, challenging, and expensive. In our work, we showed that we can generate synthetic human motion over both short- and long-time horizons using adversarial attention. Furthermore, we show that we can improve the classification performance of deep learning models in cases where there is inadequate real data, by supplementing existing datasets with synthetic motions. |
| Type Of Material | Computer model/algorithm |
| Year Produced | 2023 |
| Provided To Others? | No |
| Impact | Our manuscript titled 'Adversarial Attention for Human Motion Synthesis' has been accepted for publication to IEEE Symposium Series on Computational Intelligence. |
| URL | https://eprints.gla.ac.uk/307484/ |
| Title | Multi-Scale Cross Contrastive Learning for Semi-Supervised for Image Segmentation |
| Description | We developed knowledge distillation frameworks that aim to transfer knowledge from one network to another with the aims to provide (i) more efficient learning representations, (ii) compact networks (ii) the ability to exploit both labelled and unlabelled datasets in a semi-supervised learning framework. One article has been presented to BMVC2023: 'Multi-Scale Cross Contrastive Learning for Semi-Supervised Medical Image Segmentation'. Another article on knowledge distillation of human pose estimation is currently under preparation. |
| Type Of Material | Computer model/algorithm |
| Year Produced | 2023 |
| Provided To Others? | Yes |
| Impact | One article has been presented to BMVC2023: 'Multi-Scale Cross Contrastive Learning for Semi-Supervised Medical Image Segmentation'. This has attracted a lot of attention and inspired further research as reflected with the citations to the article. |
| URL | https://github.com/kathyliu579/MCSC |
| Title | Optimizing Vision Transformers for Image Segmentation |
| Description | We have designed a compact and accurate Transformer network (CS-Unet), which introduces convolutions in a multi-stage design for hierarchically enhancing spatial and local modelling ability of Transformers in semantic segmentation. This is based on a novel Convolutional Swin Transformer (CST) block which merges convolutions with Multi-Head Self-Attention and Feed-Forward Networks for providing inherent localized spatial context and inductive biases. Experiments demonstrate CS-Unet without pre-training outperforms other counterparts by large margins with fewer parameters. |
| Type Of Material | Computer model/algorithm |
| Year Produced | 2023 |
| Provided To Others? | Yes |
| Impact | Related work has been also accepted at the 2023 IEEE International Conference on Acoustics, Speech, and Signal Processing on optimising vision transformers for image segmentation. |
| URL | https://github.com/kathyliu579/CS-Unet |
| Description | NHS Lanarkshire |
| Organisation | NHS Lanarkshire |
| Country | United Kingdom |
| Sector | Public |
| PI Contribution | I have engaged with Patient and Public engagement groups to support Dr Ana Talbot who is an Older Peoples Consultant at NHS Lanarkshire. She has been awarded a two years NHS Scotland Innovation Fellowship to collaboratively work on 'Preventative/ Proactive Technological Approaches to Frailty, Falls and Syncope'. The aim of this fellowship is to develop digital technology to facilitate efficient completion of a multi-factorial assessment/ intervention that could reduce future risk of falls. For those without significant injury it converts an unscheduled care attendance into planned care. The patients could be signposted to leisure services for strength and balance training, one of the most effective strategies for falls risk reduction. |
| Collaborator Contribution | She provides expert clinical knowledge on falls prevention in older patients. She has provided links for patient and public engagement groups. She has also participated in workshop organisation. |
| Impact | This is a multi-disciplinary collaboration that involves clinicians (Older People Consultants) The outcome was related to engagement events: workshop organisation on 'AI in Pervasive Well-Being and Healthy Ageing' Received feedback via public and patient engagement |
| Start Year | 2022 |
| Description | Patient and Public engagement |
| Form Of Engagement Activity | A formal working group, expert panel or dialogue |
| Part Of Official Scheme? | No |
| Geographic Reach | Regional |
| Primary Audience | Patients, carers and/or patient groups |
| Results and Impact | I have engaged with Patient and Public engagement groups to support Dr Ana Talbot who is an Older Peoples Consultant at NHS Lanarkshire. She has been awarded a two years NHS Scotland Innovation Fellowship to collaboratively work on 'Preventative/ Proactive Technological Approaches to Frailty, Falls and Syncope'. The aim of this fellowship is to develop digital technology to facilitate efficient completion of a multi-factorial assessment/ intervention that could reduce future risk of falls. For those without significant injury it converts an unscheduled care attendance into planned care. The patients could be signposted to leisure services for strength and balance training, one of the most effective strategies for falls risk reduction. |
| Year(s) Of Engagement Activity | 2023 |
| Description | Workshop on 'AI in Pervasive Well-Being and Healthy Ageing' |
| Form Of Engagement Activity | Participation in an activity, workshop or similar |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Postgraduate students |
| Results and Impact | We organised a workshop on 'AI in Pervasive Well-Being and Healthy Ageing' of around 100 participants. The workshop was supported by the UKRI CDT in Social Intelligent Artificial Intelligence and funding from the Impact Acceleration and Knowledge Exchange university funds. The workshop encompassed how AI can accelerate the adoption of novel sensing approaches in monitoring Wellbeing and Disease (http://www.questhealthcare.co.uk/ ). The workshop was designed for rapid Impact Acceleration and Knowledge Exchange by bringing together academics, industry, and NHS related organisations with the goal to highlight opportunities for collaborations. We plan to: • Promotes our technologies and Brain-based AI tools for adoption in NHS and industry • Strengthen our current industrial partnerships and seek to establish new ones • Discuss state-of-the art research in the area of pervasive sensing and healthy brain imaging and how to exploit these to enable innovative solutions. |
| Year(s) Of Engagement Activity | 2023 |
| URL | https://www.gla.ac.uk/events/conferences/ai-healthy-ageing/ |