Stay organized with collections Save and categorize content based on your preferences.

NICO.LAB: Revolutionizing stroke care with StrokeViewer, a cloud-based AI solution

About NICO.LAB

Established in 2015 following the MR CLEAN trial, Dutch health tech company NICO.LAB uses the latest advances in AI technology to empower physicians in the acute setting to provide every patient with the right treatment in time. Their StrokeViewer tool uses highly accurate, clinically validated, and FDA-certified algorithms to reduce time to treatment and therefore improve patient outcome.

Industries: Healthcare, Technology
Location: Netherlands

Tell us your challenge. We're here to help.

Contact us

NICO.LAB built its StrokeViewer tool on Google Cloud, using TensorFlow to build ML models and powerful, scalable processing on Compute Engine to reduce image analysis to under three minutes.

Google Cloud results

  • Processes images in under three minutes with custom-configured GPUs on Compute Engine that launch quickly
  • Ensures easy implementation in any hospital, thanks to a cloud-based, scalable infrastructure that requires no additional hardware
  • Accelerates time to treatment and improves patient outcome, verified by recent studies

Reduces the time delay to treatment by 29%

Around one in four people will suffer a stroke in their lifetime. Beyond the huge financial burden, with strokes costing healthcare providers €60 billion in Europe and $72 billion in the US, strokes carry an enormous human cost, often leading to death or permanent disabilities. With an aging population, stroke numbers continue to increase, creating mounting pressure on healthcare professionals. In response, Dutch company NICO.LAB developed StrokeViewer, a solution that empowers physicians on every step of the stroke workflow. It connects human and artificial intelligence to ensure every patient is provided with the right treatment, in time.

When a stroke patient arrives at the hospital, the clinician first needs to determine if it’s an ischemic stroke, where there’s a chance for treatment. The blood clot in the brain needs to be located as quickly as possible. “Every additional minute between stroke and treatment affects the patient’s outcome,” explains Merel Boers, co-founder and CEO at NICO.LAB. “Every hour spent waiting for treatment equates to a loss of four healthy living years.”

“NICO.LAB was created to transform cutting-edge academic research into products that can be successfully integrated within clinical contexts. Google Cloud technology gives us all the tools we need to deliver StrokeViewer on a software-as-a-service basis: speed, accuracy, and easy scaling.”

Renan Sales Barros, co-founder and CTO, NICO.LAB

To help clinicians access the information they need to make a diagnosis as quickly as possible, NICO.LAB developed StrokeViewer, an artificial intelligence solution that accurately analyzes the key biomarkers of stroke. In a scenario such as this, where every second counts, a vendor wants a cloud provider that offers powerful processing that can easily scale to accommodate additional hospitals and patients. Following a proof of concept, NICO.LAB chose to build the infrastructure for its StrokeViewer tool on Google Cloud.

“NICO.LAB was created to transform cutting-edge academic research into products that can be brought to market,” says Renan Sales Barros, co-founder and CTO at NICO.LAB. “Google Cloud technology gives us all the tools we need to deliver StrokeViewer on a software-as-a-service basis: speed, accuracy, and easy scaling.”

“We’ve found that Compute Engine has the fastest start-up times for VMs, as well as offering a low minimum usage. Billing for VMs is per-second, with a one-minute minimum, so it’s perfect for our use case, where processing might only take a minute.”

Renan Sales Barros, co-founder and CTO, NICO.LAB

Building an infrastructure ready to launch in seconds

When fast processing times are critical, the reliability of cloud services becomes all the more important. NICO.LAB’s upload times can be affected by variable hospital broadband speeds, so to reach its target image delivery time, it needs to maximize the efficiency of its image processing. “Because the images StrokeViewer uses are big, we don’t want to keep the VMs on all the time. That would be too expensive,” explains Renan. “So we need VMs that are both really powerful and can launch as quickly as possible.”

CT scan images are transferred to StrokeViewer via a secure connection from the hospital’s infrastructure into the cloud using DICOM, an encrypted image protocol used in most hospitals. “The images vary between 50 megabytes and 6 gigabytes,” explains Renan. “The biggest images we use are 4D and consist of 1,000 slices over 30 time steps.”

If the hospital has legacy systems that don’t support secure DICOM transfers, NICO.LAB uses Google Cloud VPN to easily set up a secure private network to complete the upload. “Once it’s uploaded, the data never leaves the cloud,” says Renan. “We can select the location of the data center used, so that really helps with managing the data storage requirements of hospitals in the EU and the US.”

NICO.LAB then uses Google Kubernetes Engine (GKE), Pub/Sub, Cloud Storage, and Filestore for its optimized data pipeline. The uploaded image arrives through NICO.LAB’s DICOM service provider hosted on a Kubernetes cluster on GKE, which scales according to need. The raw image data is saved using Cloud Storage, with some attributes saved to a Firestore database. From there, Pub/Sub is used to trigger when any unprocessed images should be analyzed by the TensorFlow models.

To run its algorithms, NICO.LAB uses powerful GPUs on Compute Engine, utilizing a range of custom configurations, depending on the size and parameters of the image. “We’ve found that Compute Engine has the fastest start-up times for VMs, as well as offering a low minimum usage,” says Renan. “Billing for VMs is per-second, with a one minute minimum, so it’s perfect for our use case, where processing might only take a minute.”

Identifying stroke biomarkers using machine learning

StrokeViewer uses a number of algorithms to look for significant image features that are biomarkers for different kinds of stroke, so each patient can then be given the best treatment. They include asymmetries in the brain and elements that could indicate the presence of a blood clot, such as interrupted blood flow. To build its machine learning models, NICO.LAB uses TensorFlow.

“The goal is to identify significant features and highlight them, so the clinician can inspect them, then make an informed decision and provide the right treatment,” says Merel. “The tool works the same in the middle of the night as it does at 9 AM in the morning, giving the radiologist an extra pair of eyes, reducing the influence of subjectivity, and helping to lower the number of missed opportunities for treatment.” The differences in the scanning equipment of various hospitals, together with the relatively small datasets available for model training, present challenges for training accurate AI models. Despite this, NICO.LAB’s hemorrhage detection and quantification algorithm has a sensitivity of 94% and specificity of 95%, while blood clot detection has a sensitivity of 94.9%. StrokeViewer holds the promise to detect 15 extra strokes on every 100 patients.

“Deploying StrokeViewer on Google Cloud means hospitals don’t have to install any additional hardware, and it’s compliant with all current hospital systems,” says Merel. “We aim to provide as many stroke patients as possible with the right treatment globally, therefore, using such a scalable cloud infrastructure is a huge advantage as we’re planning to launch all over the world.”

Merel Boers, co-founder and CEO, NICO.LAB

Supporting clinicians to improve patient outcomes

Currently, it takes hospitals an average of almost two hours for CT scan images to be acquired, processed, and inspected, and in remote locations where images need to be manually transported to a comprehensive stroke center, that time increases to several hours. With the fast processing support provided by Google Cloud, StrokeViewer can deliver those images in under three minutes. “Clinical evidence shows that reducing the time to treatment by one hour can shift 10% of patients from a clinically poor to a clinically good outcome,” says Merel. “That represents a profound impact on quality of life.” Recent preliminary results have shown StrokeViewer saves a hospital in the Netherlands on average 44 minutes from the time the patient arrives at the hospital to receiving treatment.

StrokeViewer received FDA approval recently and CE certification in 2018, and NICO.LAB has been implemented in several hospitals in the Netherlands, with launches in an additional six European countries, the US, and Australia planned for 2021. Thanks to the flexible, scalable infrastructure built on Google Cloud, implementing the solution in new hospitals will be a straightforward process for the physicians and the hospitals’ IT departments. “Deploying StrokeViewer on Google Cloud means hospitals don’t have to install any additional hardware, and it’s compliant with all current hospital systems,” says Merel. “We aim to provide as many stroke patients as possible with the right treatment globally, therefore using such a scalable cloud infrastructure is a huge advantage as we’re planning to launch all over the world.”

To improve its processing times further, NICO.LAB uses the Cloud Healthcare API, which supports the DICOM protocol used by StrokeViewer. They are also planning to start using Cloud TPU. “We are always trying to optimize and achieve faster processing times,” Renan explains. “Thanks to the latest hardware solutions offered by Google Cloud, we can do things now that weren’t possible even a few months ago.”

Looking ahead, NICO.LAB plans to continue developing products to reduce the impact of stroke by focusing on cutting-edge technology. “We strongly believe that cloud platforms are the way forward in creating scalable products that can rapidly be brought to market,” says Merel. “With remote maintenance and compatibility, the potential impact on the number of lives we can save or improve is really significant.”

Tell us your challenge. We're here to help.

Contact us

About NICO.LAB

Established in 2015 following the MR CLEAN trial, Dutch health tech company NICO.LAB uses the latest advances in AI technology to empower physicians in the acute setting to provide every patient with the right treatment in time. Their StrokeViewer tool uses highly accurate, clinically validated, and FDA-certified algorithms to reduce time to treatment and therefore improve patient outcome.

Industries: Healthcare, Technology
Location: Netherlands