Skip to content
Home » Blog » Multimodal AI for Brain Disorders

Multimodal AI for Brain Disorders

The impact of AI applications for brain disorders in fields like Neurology or Psychiatry is still limited in comparison to Radiology or Cardiology. The diagnosis and care processes for conditions such as depression or dementia are complex, involve multiple tests, data types and specialists; and they are, in some cases, very dependent on the subjective ability of the patient to report on their condition. Some of these challenges have become at the same time an opportunity to improve the healthcare processes by leveraging AI.

Multimodal Machine Learning solutions can assist specialists and provide them with new tools and insights to make the healthcare pathway more efficient and objective. In this post, we introduce one of the projects we are developing at PickleTech focused on such improvements: HALO. Starting from a cohort of elite athletes, we leverage multimodal AI and the development of biomarkers, aiming to provide athletes and preparation teams with objective measures of their mental health.


The relation between brain conditions such as depression, dementia or strokes are multiple and quite involved. From the medical perspective, the nature of this association still remains to be fully understood, but the scientific evidence is growing, e.g. [1]-[2]. Looking at the likely biological mechanisms behind -see [1] for dementia and depression- it is no surprise there are also relations in their diagnosis and treatment processes. For instance, speech patterns and AI are being researched and analyzed as possible biomarkers to improve the diagnosis of dementia [3], depression [4]-[5], or Parkinson [6].

Beyond the medical associations, there are also common characteristics in the healthcare pathways of such brain conditions. Their diagnosis is complex to confirm, involves varied tests, and combines several data types. For depression in particular, there is a strong dependence on the subjectivity of the patient. The patient’s journeys are in addition affected by taboos, and personal fears, existing along the process of noticing the symptoms, seeking and accepting care, and living with the condition and hopefully treatment afterwards.

The existence of multiple data features informing of the disease is probably one of the reasons for the slower appearance of innovation techniques that impact the patient journey. But at the same time, this has set the perfect context for multimodal AI to provide the tools and insights that can revolutionize healthcare.  Multimodal AI aims to build models that can process and relate information from multiple modalities: numeric, text, audio, images, physiology etc.

Properly devised, multimodal AI can assist health practitioners providing new objective measures and biomarkers, as well as tools for resource intensive clinical processes. AI has the potential to enhance the capability of our clinicians to treat brain conditions, speeding and making the healthcare pathway more efficient.

In this post, we focus on introducing one of such cases we are developing at PickleTech. The case of the mental health of athletes.

Competitive Minds

We are used to perceive elite athletes as almost superpowered human beings with an unbreakable willpower and mental strength. Fear, doubt, lack of confidence, or demotivation, are hardly ever used when talking about sports. Athletes have to be the toughest, or face the stigma of weakness. While the focus of research and advances on athlete performance have centered on injuries and physical performance, leading to impressive advances, the extent and reach of mental health studies in sports have been comparatively limited.

This picture has already started to change. We have seen several cases of athletes openly speaking about mental health issues: Michael Phelps, Naomi Osaka, Andrés Iniesta, Simone Biles, Ricky Rubio, Kilian Jornet, Kevin Love or Tom Dumoulin, to mention some notorious cases. The limited research on mental health conditions, such as burnout, depression, anxiety, panic disorders, eating disorders, and substance abuse, draw a worrying picture. Depending on the study, see e.g. [7] or [8], numbers talk about a prevalence of depression and anxiety symptoms of 40-50% among elite athletes in team sports, and certainly non negligible numbers when it comes to burnout, eating disorders, or substance abuse, varying depending on the study: from 5-25%. Providing tools for a systematic approach has become imperative to assess and protect the mental health of athletes.

At the same time, from our conversations at PickleTech with elite teams and athletes in several sports, like football or cycling, we hear a recurrent consensus: the mental component of the athlete plays a major role in its preparation and its mental health has a huge impact on its performance. This is many times perceived by the public as well. Even if not openly shared with them: athlete career progressions that suddenly stop or fade away; severe performance drops when changing teams; or when playing with a club vs. with the national team; home vs. visiting; when facing high pressure phases like competition finals or top level championships, just to mention a few. There is a common agreement that mental health issues severely impact the athlete’s performance. Mental health and injuries are strongly connected as well.

Image by RCS/LaPresse.

The economical impact of injuries and performance improvements in sports like football or basketball are estimated at the multi-million level per year. But while generically accepted, the effects of mental health conditions on performance are not properly quantified nor integrated in modern training solutions. Apart from attempts at addressing this problem involving indirect features that alone lack a contextual interpretation, such as for instance heart rate variability (HRV), quality of sleep, or other wellbeing data.

In order to protect the health of athletes and push modern solutions on athlete performance and preparation to the next level, we must evaluate and integrate mental health variables within them. We need a comprehensive picture of the athlete’s performance, including not only physical features, but also information related to mental health.

HALO

PickleTech’s multimodal AI project HALO aims to provide athletes and preparation teams with new objective measures and tools for their preparation. HALO focuses on two main challenges. First, it helps athletes prevent and treat earlier situations of mental fatigue and burnout, and also allows the detection of symptoms related to pathologies like anxiety and depression. The aim is to provide early insights and warn athletes of their need to seek special care. Athletes and preparation teams need to normalize their treatment as for any other physical condition they face.

Second, the solution enables the integration of the athletes’ own perceptions and feelings in their preparation for top-level sports competition, aiming to increase their performance. The golden question in many sports is still how much and what type of training to complete to seek a stimulus that leads to a supercompensation and then improved performance. Load and adaptation. There is continuous research and many recent innovations to provide more data to respond to it: e.g. the use of HRV, which is known to be related to parasympathetic activity. But mixing an athlete’s perception in the response is far from being solved, and still relies heavily on the subjective ability of the athlete to interpret it. HALO aims to advance further in this direction.

Following advances on the use of AI to find patterns for the characterization of perceptions based on e.g speech variables [5], visual expressions [10], HRV [11], on top of other assessments [12], HALO leverages multimodal Machine Learning in close collaboration with mental health specialists. This enables the combination and analysis of physiological signals, speech features, personal perceptions, and other emotional and behavioral expressions, in order to find patterns and biomarkers that signal certain pathologies like anxiety or depression.

Beyond the acute need for a tool adapted to athletes needs, focusing the development on sports provides several other advantages in the long term vision of addressing the mental health pandemic in wider populations. Athletes are used to monitor aspects of their sportslife to seek health and performance improvements. Focusing first on athletes reduces the initial resource intense demand to fulfill the certification processes needed for medical devices.

Athletes enable the definition of a more homogeneous cohort: they share common traits regarding lifestyle, behavior, health levels, or competitive perspectives. This makes it easier to find and study biomarkers related to actual mental health conditions; rather than fooling the AI learning process with other cohort confounding factor differences. This same homogeneity enables us to focus on studying the relation of these biomarkers with events and environmental variables of interest to athletes. This is, what are the biomarker relations to specific events in their lifestyle: competition milestones, contracting periods, pre-season ramp ups, holidays, national team competitions etc.

This makes HALO a unique opportunity to improve the lives of athletes, while developing a methodology applicable in the future to any amateur practitioner.

Powered by Data, Driven by Science

Multimodal Machine Learning solutions can assist health practitioners treating brain conditions. Projects like HALO provide new objective measures and biomarkers, as well as tools that help clinicians to speed up and make the corresponding healthcare pathways more efficient.

At PickleTech, we work developing tailored solutions to improve competitive aspects related to Health, Sports, and DeepTech. We believe Data Science and advances in Machine Learning coupled with domain knowledge and experimentation have the potential to provide new tools to better understand, monitor, and systematically improve the competitive performance of organizations.


*Opening image by Pexels stock image.

[1] Depression and Risk of Developing Dementia, Amy L. Byers, PhD, MPH and Kristine Yaffe, MD: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3327554/

[2] Are Depression and Dementia a Common Problem for Stroke Older Adults? A Review of Chosen Epidemiological Studies, Karolina Filipska, Adam Wiśniewski, Monika Biercewicz, Robert Ślusarz: https://pubmed.ncbi.nlm.nih.gov/32277403

[3] Artificial Intelligence Tools to Evaluate Language and Speech Patterns in Alzheimer’s Disease, Anna Favaro, Seneca Motley, Quincy M Samus, Ankur Butala, Najim Dehak, Esther S Oh, Laureano Moro-Velazquez: https://pubmed.ncbi.nlm.nih.gov/36537479/

[4] Automated depression analysis using convolutional neural networks from speech, Lang Hea. Cui Cao: https://www.sciencedirect.com/science/article/pii/S153204641830090X

[5] Robust Speech and Natural Language Processing Models for Depression Screening, Y. Lu, A. Harati, T. Rutowski, R. Oliveira, P. Chlebek, E. Shriberg: https://ieeexplore.ieee.org/document/9353611

[6] A Hybrid U-Lossian Deep Learning Network for Screening and Evaluating Parkinson’s Disease, Rytis Maskeliūnas, Robertas Damaševičius, Audrius Kulikajevas, Evaldas Padervinskis, Kipras Pribuišis, Virgilijus Uloza: https://www.mdpi.com/2076-3417/12/22/11601

[7] Symptoms of Common Mental Disorders in Professional Football (Soccer) Across Five European Countries, Gouttebarge V, Backx FJ, Aoki H, Kerkhoffs GM: https://pubmed.ncbi.nlm.nih.gov/26664278/

[8] Sport psychiatry: a systematic review of diagnosis and medical treatment of mental illness in athletes, Reardon CL, Factor RM: https://pubmed.ncbi.nlm.nih.gov/20942511/

[9] Machine Learning for Medical Devices, PickleTech: https://pickletech.eu/blog-medical-devices/

[10] Deep learning in mental health outcome research: a scoping review, Chang Su, Zhenxing Xu, Jyotishman Pathak, Fei Wang:  https://pubmed.ncbi.nlm.nih.gov/32532967/

[11] Artificial Intelligence in education: Using heart rate variability (HRV) as a biomarker to assess emotions objectively, J. W. Y. Chung, H. C. F. So, M. M. T. Choi, V. C. M. Yan, T. K. S. Wong: https://www.sciencedirect.com/science/article/pii/S2666920X21000059

[12] The PHQ-9: validity of a brief depression severity measure, Kroenke K, Spitzer RL, Williams JB: https://pubmed.ncbi.nlm.nih.gov/11556941/

[13] Data Science for Emergency Medicine, PickleTech: https://pickletech.eu/blog-emergency/

[14] The Drag Side of the Moon, PickleTech: https://pickletech.eu/blog-drag/

[15] Medical Accelerator Data Science, PickleTech: https://pickletech.eu/blog-particle-accelerators/

EN