Where did the original idea for this project come from? What led you to it?
Adrika Gupta: NeuroSense emerged from an observation that initial symptoms of mental decline usually start with behavioral differences in speaking, moving, and everyday routine; however, this type of monitoring is hardly done constantly in real-life settings. We were curious to find out if AI can make early detection possible while providing sufficient privacy for the patients. Thus, we created a privacy-oriented edge AI solution based on EEG, speech, and mobility features aimed at detecting behavioral changes compared to individual patient’s behavior pattern. The idea behind our invention wasn’t to diagnose anything, but to do research and raise awareness about the necessity to seek medical advice.
What is a digital twin? How would you describe it, and how does it work?
Sangyan Hari Pushkar: Digital twin refers to a model of a physical entity, which is updated based on data from the real world. Digital twins can be utilized in health care applications for modeling behavioral or physiological patterns, which can be tracked for change. In NeuroSense, the digital twin refers to the customized behavioral model, where information such as gait and speech is obtained alongside EEG readings to learn the individual's baseline patterns, which can be compared in the future to detect behavioral changes. The use of digital twins enables tracking of changes that are unique to an individual as opposed to generalizations that may exist among people within the entire population.
What exactly does your project detect, and what kind of data does it work with? What can it capture, and over what time frame?
Sangyan Hari Pushkar: The NeuroSense system is built to track incremental drifts in behaviour patterns that might indicate an early onset of cognitive impairment. NeuroSense uses multimodal data which include EEG waves, speech characteristics and gait/movement patterns. When analyzing speech characteristics, NeuroSense considers pause patterns, speech rate, and fluency levels. When considering gait and movement patterns, NeuroSense looks at movement stability and consistency. In terms of EEG analysis, NeuroSense is able to detect any changes in patterns in the form of brainwave patterns in alpha, theta, and delta bands. Unlike the single snapshot approach, NeuroSense is meant to be used for a prolonged period of time, typically lasting for weeks or even months. The purpose is to observe subtle differences from individual behavior patterns.
How long before the onset of Alzheimer’s disease do certain types of symptoms appear in a person’s speech or movements? Can such symptoms be recognized by their close ones? Can an average person notice them?
Aadrika Gupta: Neuroscience and research on brain health reveal that certain behavioral changes that are caused by Alzheimer's disease tend to manifest themselves several years before the official diagnosis of the disease. Such behavioral changes might be reflected in more frequent pauses while speaking, decreased fluency, problems with finding appropriate words, changes in balance during walking, slower patterns of movement, and other behavioral changes. However, it should be acknowledged that such symptoms tend to emerge gradually and not abruptly; hence, they might be mistaken for the effects of aging, stress, tiredness, and other temporary factors. At times, family members and close friends might detect the emergence of the mentioned symptoms, but there would be no systematic evidence that could require immediate medical attention. As a result, a common individual will most likely miss those signs of the onset of the disease in the first place; hence, one of the reasons behind our project was to try to figure out whether continuous behavioral monitoring via speech, movement, and brain activity would help detect any gradual drift in time in an efficient and anonymous manner.
In what ways are current hospital approaches inadequate? Why are they insufficient?
Aadrika Gupta: The current hospital strategies may prove to be inadequate as patients are usually observed in clinical sessions which are relatively short. Cognitive problems are known to evolve slowly for months or years together; hence, certain behavioural patterns can remain unobservable within the short duration of an observation session. Apart from this, continuous surveillance away from hospitals is still not common, and some systems require cloudbased processing, which may pose certain privacy issues. Our concept behind NeuroSense involved exploring the possibility of using edge AI that ensures the privacy of the users.
What would this look like in practice? How could a person use or apply the device? Who would evaluate the results, and how? Or would the device itself notify the person that they should see a doctor?
Aadrika Gupta: In terms of implementation, NeuroSense will be built to operate as a continuous behaviour-monitoring system for application in everyday situations rather than hospital settings alone. The user can utilize NeuroSense through wearable devices that will collect information on speech, motion, and periodic EEG data collection. This collected data will then be analyzed through AI processing on the edges, and behavioral trends can be represented via an interpretable dashboard. Rather than offering an immediate diagnosis, the system will observe behavioral changes over time compared to the user’s usual pattern. The main aim of this platform is early awareness. In cases where clear behavioral patterns of drift emerge, the system could alert the user to seek advice from a doctor. Yet ultimately, the interpretation and clinical diagnosis will still fall into the hands of doctors.
You mention in the project that the system also aims to work with EEG. Will the person’s brain activity actually be measured? For how long and how often would they need to be “connected” in order to obtain valid and reliable data?
Mayur Mundada: Indeed, our project considers utilizing EEG signal readings as part of multimodal behavioral tracking. However, it does not aim at implementing the constant and professional-level monitoring of the brain waves by conducting EEG readings in hospitals. Currently, at the stage of creating a prototype, we investigate whether there is an opportunity to track certain behavioral changes on the basis of alpha, theta, and delta bands of EEGs. The specific period needed to do that will need further validation. Now, the main point is to consider the possibilities of implementing the project without imposing a requirement for the users to be constantly connected to our system
How financially demanding would this continuous “monitoring” be for the patient? Is it expected that the tool would be covered by insurance, or would patients have to pay for it themselves? And if so, what price range are we talking about?
Aadrika Gupta: One of the objectives of NeuroSense is to ensure that continuous behavioral monitoring is made accessible and affordable as much as possible. The architecture of the technology includes edge AI and consumer-friendly sensors, which makes us less dependent on costly hospital equipment or even cloud-based processing systems. As far as our current level of research is concerned, it is too early to establish any specific prices at this stage. But the idea is to develop the platform with the help of wearable technologies that can make monitoring more accessible. In the coming years, if the system proves its utility with sufficient validation research, healthcare institutions or even insurance companies might be able to offer them as preventive healthcare tools for cognitive disorders.
The data is divided into three categories—the first serving the patient and their family, the second intended for doctors, and the third for scientific purposes. How could this data be used in the future? Could it contribute to further research on Alzheimer’s disease?
Sangyan Hari Pushkar: Yes, we believe that such data would be useful in future research for Alzheimer’s and other cognitive disorders, especially if they are collected longitudinally and anonymized. For the patient and their family members, the data will be able to show behavioral patterns over time and assist in detecting signs of cognitive problems early on. Clinically, the data can provide more insight into the patient's behavior between visits, concerning their speech, motor function, or any EEG behavioral drift. Research-wise, larger datasets and anonymization of the information can allow researchers to better understand behavioral patterns over time and assist in creating future cognitive monitoring algorithms using artificial intelligence. It would allow for conducting research without compromising the privacy of the individuals involved in the study.
The project also includes a dashboard that helps the individual maintain a routine. Are these recommendations based on measured data and personalized for the specific user, or are they general recommendations for maintaining good health and improving cognitive reserve?
Aadrika Gupta: As of now, with respect to the prototype we have, the dashboard generates generic recommendations with regards to health habits, cognitive activities, and well-being. But as a long-term plan, the recommendation engine will be designed to generate personalized recommendations by leveraging the behavior trends seen in data from the individual themselves. For instance, the future generation will take into account shifts in behavior, speech behavior, and cognitive trend indicators while ensuring that there is no breach of privacy at any point.
In your prototype, you also include a Cognitive Score Trend, and in the section for researchers even a live EEG stream. Is it possible, for example, to study how different activities affect brain activity— which ones are beneficial and which ones may place unnecessary strain on it?
Mayur Mundada: Perhaps, yes. An intriguing point about longitudinal tracking is that it might aid in studying correlations between behavioral patterns and activities, as well as variations in the EEG trends over time. At present, the Cognitive Score Trends and real-time EEG streams are primarily designed for visualizing and analyzing trends from a research perspective rather than clinical assessment. Nevertheless, with bigger data and clinical validation, applications such as NeuroSense can be used to examine how various factors, including sleep, exercise, stress, or cognition, affect behavioral and brain activity patterns.
Could the application potentially function as a kind of diary that a patient keeps, from which data could then be analyzed to determine what is “good” for the brain?
Aadrika Gupta: Yes, that would definitely be one of the possible ways ahead for our platform. Combining constant behavioral monitoring with everyday activity routines will enable the system to operate in a way similar to a cognitive health diary. Eventually, after collecting sufficient data, it might become possible to determine any particular behavioral pattern associated with increased focus or stability for an individual user, leading towards more customized suggestions about their routine, physical activity, and cognition-related exercises. That said, drawing conclusions from the data collected will still require much more extensive data collection, and so far, we can consider our prototype more like a research-oriented tool.
And if such data were collected on a larger scale and in greater volume, could it then be evaluated collectively and used, for example, to derive recommendations for the prevention of neurodegenerative diseases, or recommendations on what to avoid?
Sangyan Hari Pushkar: In principle, yes. Large sets of data pertaining to behaviour and EEGs collected in a manner that respects privacy and anonymity might be useful for analysing trends in relation to cognitive wellness and neurodegeneration. Ultimately, such an analysis might facilitate the formation of more precise recommendations about how individuals can engage in certain activities or behave in a way that mitigates their risks of developing conditions like Alzheimer’s disease. However, such an outcome will necessitate much more extensive clinical studies to verify any findings. Currently, our research is centred around creating a research-based platform for such inquiries to be conducted properly.
Could the system’s use be expanded in the future to help prevent other neurological conditions—not only neurodegenerative ones, but also autoimmune diseases (such as early-stage multiple sclerosis), or even to help predict the timing and intensity of epileptic seizures?
Mayur Mundada: Yes, we think that the architecture can possibly be expanded in the future for diseases other than neurodegenerative diseases. As NeuroSense uses multiple modalities of behavioral and EEG-based data, other neurological disorders which see changes in brain functions or movements and behavior over time can similarly benefit from this kind of approach as well. This is because EEG-based monitoring has become an increasingly significant topic within epilepsy research, while movements and behaviors might also be relevant in case of disorders like multiple sclerosis. With disease-specific data sets, systems like NeuroSense might help monitor such cases early on. But at present, we are focusing only on early cognitive behavioral monitoring of neurodegenerative disorders, so any expansion will need considerable research efforts.
How do you plan to develop the project in the future? Will the financial support associated with winning the CLARA YOUNG INNOVATOR AWARD 2026 help in its implementation?
Aadrika Gupta: For the future, we would like to further advance NeuroSense by performing larger validation studies, improving our multimodal models, and integrating EEG, speech, and behavior more effectively. We would also like to focus on personalizing NeuroSense, making the system more interpretable and ensuring deployment on edge devices that can preserve user privacy. It is crucial that we work with researchers and clinicians to test and assess the effectiveness of NeuroSense in reallife applications and determine its practical utility in long-term cognitive assessments. The encouragement received from the CLARA Young Innovator Award will certainly contribute towards advancing our project. We will have access to resources for prototyping, testing, and conducting further research, particularly in domains like edge computing and multimodal analysis. Most importantly, the award itself will motivate us to keep innovating in the realm of AI for healthcare and neuroscience.
