- Industry
- Healthtech: long-term care, wearables and AI
- Stage
- Start-up, from first prototype to market
- Reon’s role
- Co-founder and CTO
- How long
- Clinically validated in 2020, on the market in 2021
The client and the situation
Mentech, a Dutch start-up building emotion AI for long-term care.
People with an intellectual disability or dementia often cannot tell anyone when they are distressed, and behavior labeled as challenging is often a sign of stress that went unnoticed. Stress can lead to incidents and escalations.
Care organizations work with heavy workloads and staff shortages, and they had few tools to act early.
The problems
- The product had to be clinically validated and developed under medical-device rules before care organizations could rely on it.
- COVID closed care institutions in the middle of the validation studies.
- As users grew, an expensive external database filled up quickly and backpressure caused system errors.
The journey
- ConceptReon co-founded Mentech and, as CTO, led HUME from concept and hardware prototyping.
- BuildReon defined the system architecture, from sensors in washable textile to real-time stress classification and a care platform on AWS, and grew the engineering team to 30 engineers.
- 2019Mentech won a Gerard & Anton Award.
- 2020HUME was clinically validated, with the studies moved to remote data collection after COVID closed care institutions.
- 2021HUME became commercially available and went into use in Dutch long-term care.
- ScaleWith AWS partner DoiT, the platform moved from one monolithic database to services matched to each type of data: database costs fell by 60% while the user base grew fivefold.
The solution: technology and approach
- Electrodermal activity and heart-rate sensors on a custom circuit board in washable textile, worn as a sock, with a wrist variant
- Firmware in C/C++ on the Zephyr real-time operating system, with Bluetooth Low Energy to a hub
- Movesense sensors as a source of physiological data
- Real-time signal processing and stress classification with multi-task learning, in Python with PyTorch and TensorFlow
- A care platform on AWS with secure data ingestion, alert routing and a dashboard for care professionals, showing stress as a green, orange or red signal
- Development under ISO 13485 quality management and IEC 62304 software lifecycle requirements
What they got
- HUME is MDR-certified and clinically validated, and is used by more than 30 care organizations in the Netherlands, Belgium and Austria.
- Reon reports more than 90% accuracy on real-world data.
- HUME's research and IP include two patents (2022 and 2023) and more than 15 peer-reviewed publications.
- In one published case, a young woman's daily escalations stopped after more than six months of using HUME.
- Mentech won a Gerard & Anton Award in 2019.
Sounds like your situation?
- You are building a medical or care product that combines hardware, data and AI.
- Clinical validation and medical-device rules stand between your prototype and your first customers.
- Your platform has to scale with its users without its cloud costs growing as fast.
Related services and use cases
Sources
- Reon Smits’s profile on CTO Grand Rounds
- HUME project - reonsmits.nl (opens in a new tab)
- HUME - mentechinnovation.eu (opens in a new tab)
- Gerard & Anton Award winner Mentech Innovation (2019) - ioplus.nl (opens in a new tab)
- Corona bridging loan helps Mentech through validation (2020) - ioplus.nl (opens in a new tab)
- HUME gives vulnerable people a voice (2023) - ioplus.nl (opens in a new tab)
- Mentech showcase - movesense.com (opens in a new tab)
- Improvements in stress-detection technology - AHFE Open Access (opens in a new tab)
This case describes Reon Smits’s own work before Grand Rounds, as Reon’s profile and the public sources above describe it. It is not a Grand Rounds engagement and not a promise of the same result for another company.