CEREBRA: Kazakh startup taps AI for early stroke detection

AI has the power to revolutionize the healthcare industry.

CEREBRA: Kazakh startup taps AI for early stroke detection
CEREBRA Founder and CEO Doszhan Zhussupov. Image credit: CEREBRA

While artificial intelligence is generating a lot of controversy these days due to the popularity of ChatGPT and AI art, a tech startup in Kazakhstan is using AI-powered software for early stroke detection. And the founder has a personal reason for making this his mission.

"Why exactly stroke among other brain diseases? Because my father died of a stroke attack, and it was kind of a personal thing and not just for me. It is the life story of many people in the world. Even on my team, the Head of ML's father got hit by a hemorrhagic stroke. The ML architect lost his father due to a stroke attack as well," CEREBRA Founder and CEO Doszhan Zhussupov told Digital Life Asia.

Origin story

Zhussupov said that when they started their journey in healthcare, they realized that Kazazkstan was facing shortage of qualified neuroradiologists, which creates a major challenge in the early diagnosis of strokes. He said this is even the case in the US, where the demand is so great that the country will have a projected shortage of up to 42,000 radiologists by 2033.

"In addition, we found that there was a miscommunication between neurosurgeons and neurologists, and doctor burnout and overwhelming emergency conditions. We also discovered that the human eye has low sensitivity in recognizing emerging anomalies on CT scans. So, we decided to apply AI to address these challenges and improve the early diagnosis of strokes. Our AI technology can accurately analyze brain scans and detect even subtle signs of stroke that might be missed by the human eye. This will enable doctors to make more informed decisions and ultimately save more lives," he said.

It was in 2018 that Zhussupov started what would become CEREBRA.

"I had a technical team and money from selling my shares in a medtech startup called Med365.kz," he said.

While he initially had three teams working on applying machine learning and AI in three healthcare areas – pulmonology, ophthalmology, and neuroradiology. The first challenge they ran into was convincing hospitals to share data with them. After overcoming that and doing parallel research in these three areas for four to five months, they ran into a second obstacle: money.

"We were hit by the severe reality that we cannot afford doing a lot of research simultaneously. So, we decided to cut two projects – pulmonology and ophthalmology – and focus all our remaining resources and energy just on one topic within the neuroradiology field – brain stroke. The main reason neuroradiology won lies in the active feedback we were getting from many neurosurgeons in Almaty who wanted to be our early adaptors and expected us to deliver proper and useful products for their needs. And that was how CEREBRA.kz was born," he said.

Powerful algorithms

How does CEREBRA facilitate the early diagnosis of strokes through its algorithms?

When a stroke patient is brought to the hospital, a CT scan is conducted. Unfortunately, sometimes the CT scan shows no symptoms of a stroke, and so the patient suffers later on because of inaction and treatment without medical opinion. This leads to a high risk of disability and death.

With CEREBRA, however, the CT scan is analyzed by the AI-powered software's algorithms. The software is available as a web platform and a smartphone app for iOS and Android. Not only does it enable quick analysis of brain damage according to the ASPECTS (Alberta Stroke Program Early CT Score) scale, but also the automation of medical guidelines. According to CEREBRA, the software can identify a brain stroke within 10 minutes.

"Our ASPECTS service is based on the segmentation of ischemic stroke lesions. First, we segment regions which belong to the ASPECTS atlas (regions of middle cerebral arteria). The Unet-like model was trained to tackle such a task. Then, we detect the regions which are affected by ischemic stroke using the segmentation model – Unet. After, the relative volume of ischemic stroke lesion is presented in each region of the atlas and if it exceeds the thresholds we can say that a region is affected. Finally, we calculate 10 minus the number of affected atlas regions which is the ASPECTS score," Zhussupov said.

For the product's ICH (intracerebral hemorrhage) service, CEREBRA aims to process original (non-coregistered) images using 3D UNet.

"After inference, the service produces individual jsons for each plain (axial, sagittal and coronal) which contain the polygons of hemorrhage on the plain," he said.

Meanwhile, Zhussupov said the heatmap service aims to highlight the regions of high interest with any abnormalities for the doctor.

"We trained the WGAN (Wasserstein GAN) model on normal brains (without any disease). Then we ask the model to generate an image based on the left side of the brain slice image. Then we flip the right side to be the left side and repeat the inference on it. After that, we can create a generated image by concatenating both generated sides. Finally, we differentiate between the generated image and the actual one, and after normalization, we see regions with the bigger difference as red ones, whereas regions without any abnormalities are shown as blue ones," he said.

AI and society

Asked for his thoughts on the impact of AI on society and the fears it is generating, Zhussupov replied: "AI is a really powerful and yet scary technology. So, it could do both harm and benefit people. Thus, I share thoughts of many techpreneurs that AI must be carefully regulated. But I am 100% sure it will change society and the normal state of living for good and it will happen during the next 10 years. Even in our field – AI in medical imaging. We expect comprehensive, cloud-native, AI to be born during the next 5 to 6 years. That AI will be able to detect 80% of known diseases better than any expert could do. Probably, it will be in the form of a library of different standalone AI solutions combined into one powerful tool."

Zhussupov is quite bullish on AI as an example of technology as a powerful tool for good.

"I strongly believe in the potential of AI, particularly in healthcare. AI has the power to revolutionize the healthcare industry by enabling faster and more accurate diagnosis, improving patient outcomes, and ultimately saving lives. Our AI-powered stroke diagnosis system has already shown promising results, and we believe it has the potential to make a real difference in the lives of patients and healthcare providers. By combining advanced technology with medical expertise, we can create innovative solutions that address the most pressing healthcare challenges of our time," he said.

CEREBRA has already gone through three rounds of funding and successfully closed all of them. The pre-seed was for US$300,000, seed was for US$1 million, and the startup has just closed its pre-series A for 2 million USD.

For this year, CEREBRA plans to expand abroad, with the US as its main target. Zhussupov said that the company is also already working with Saudi Arabia and Uzbekistan.

What advice would Zhussupov like to give aspiring startup founders?

"No one ever believed deep tech in healthcare was possible in Kazakhstan. But we did it. We went from absolute outsiders to one of the best healthtech startups in Asia (according to Medtronic and J&J) in three years. So, the only rule I learned that works is 'Be patient and just don't give up. You must stay in the game no matter what,'" he said.

Not only did Zhussupov stay in the game. With CEREBRA, he and his team are changing the game.