Walter De Brouwer, co-founder of Doc.ai, talks about his vision to revolutionize the way we handle medical data and his very personal motivation to fight for this change.

In 2005, Walter De Brouwer and his wife Sam were sitting in a hospital room. Their 5-year-old son had had an accident and suffered a severe brain injury. There was nothing the De Brouwers could do, except to trust the doctors and wait. Waiting for their son to wake up, hoping that he would wake up at all.

Walter De Brouwer is not a man to sit around and do nothing. Originally from Belgium, he is a polymath who has founded no less than 11 companies. He led two IPO’s in the technology sector in Europe and also worked for MIT’s One Laptop Per Child initiative.

After the dot-com bubble burst, De Brouwer briefly moved away from technology and ran a bank in Brussels, Belgium. He jokingly explains he realized that “the only people who didn’t lose money in the crash were bankers”, and thus decided to become one.

Healthcare was ripe for disruption

The accident had been a life-altering event that turned the world of the De Brouwers upside down. All they could do was to sit and wait. Wait for doctors to give them information, for their son to get better.

Their son was in the hospital for almost a year. The routine of visiting hospitals and doctors and become the De Brouwers’ new normal. Waiting, asking questions without getting satisfying answers, more waiting.

But Walter De Brouwer didn’t despair. Where others might have drowned themselves in sorrow, De Brouwer did what he knew best: assess the situation and find a way to improve it. “I thought [the healthcare] industry was really ripe for disruption,” he tells me.

The most frustrating part of the De Brouwer’s hospital experience was the lack of structured medical data about their son’s condition. There was either no data available at all, or it was in cryptic doctor reports that were difficult to understand. Doctors would often spend a long time looking for data somewhere in their files and then had no time to explain the reports.

Redistristributing power

That gave De Brouwer an idea. Why not shift the responsibility of keeping patient data from doctors back to patients? This would empower patients who felt poorly informed and at the same time assist doctors who were swamped with data-keeping tasks. Patients would have access to their data, while doctors could focus on interpreting that data and deriving treatment decisions from it.

“The best keepers of [patient] records are the patients themselves. And the best interpreters of this information are the doctors,” De Brouwer says. “So, let’s not do each other’s work.”

The De Brouwers’ company doc.ai is working on a way to achieve this task. doc.ai wants to enable people to collect all their available medical data on their mobile device. AI offers personalized health insights based on the data and scientists can be allowed to access it to develop new predictive models.

This all, De Brouwer explains, is built on blockchain technology in order to leverage features such as privacy by design, security by decentralization and integrity by immutability.

Specifically, doc.ai wants patients to have access to their medical raw data, such as blood and genetic tests, instead of high-level reports.

Today, doctors have to rely on reports that patients bring with them from other doctor appointments. A full set of raw clinical data would enable each doctor to gain more detailed and accurate insights about the patient and examine the data from her own perspective.


Blockchain for medical records

A pillar of doc.ai’s technology stack is blockchain. Invented in 2008 for the cryptocurrency bitcoin, blockchain has since proven itself a promising concept for solving a variety of problems.

A blockchain is a chain of records, where each new record contains a hash of the previous one. This design makes blockchain immune to retroactive manipulations of the block data since all subsequent blocks would have to be altered as well.

For electronic medical records, blockchain could guarantee authenticity. Together with strong encryption, the technology can make storing medical data on any electronic device very secure.

Edge computing could cure our fixation on the cloud

The choice of device is another of De Brouwer’s revolutionary ideas. Instead of storing data in the cloud as it is common today, De Brouwer wants patients to store their medical data on their own devices, such as their smartphone. Nobody but the patients themselves would then be able to grant or remove access to the data, returning power over their medical data back to the patients.

De Brouwer sees huge potential in edge computing — using end-user devices such as smartphones for complex computations. He argues that cloud computing was driven by the need for computational power and storage space, but now that edge devices are catching up it’s time to reverse the trend.

Mobile device storage is steadily increasing. Even if medical records take up 10 gigabytes of data, they could be easily stored on most smartphones without limiting the user.

De Brouwer is certain that similar improvements are coming for edge device computation, enabling machine learning applications on smartphones. More and more smartphones include powerful GPUs. Huawei is already offering the first smartphones with dedicated AI chips. De Brouwer also mentions TensorFlow Light, a TensorFlow version for mobile devices, as an example of software support for machine learning of edge devices.


Edge training is still a distant dream

But there are still challenges to overcome. “The main problem is training [machine learning algorithms] on [mobile] devices,” De Brouwer admits. “You just cannot do it today.” Trained neural networks are fast and small in memory footprint, but training them takes huge amounts of data and computational power.

And even if they would be, most users would hardly want to spend their battery charge on training an AI algorithm. But Walter De Brouwer is an optimistic man. He admits to the impossibility of training on edge devices, but he quickly counters: “My question is: will it still be the same big challenge in 2020? Probably not. The technology will be there.”

Instead of talking about a task’s infeasibility, De Brouwer prefers to prepare for the day when it becomes feasible: “It is very important to start the learning curve now,” he says.

AI for every task

Doc.ai uses AI technology for a vast variety of tasks related to collecting and interpreting patient data. Computer vision algorithms try to approximate information such as age, height, and body mass index from a selfie.

Natural Language Processing models read data from medical test results. Yet other algorithms analyze the medication a patient takes and correlates this information with test results and other factors.

De Brouwer says his company has tried many different machine learning architectures and has ultimately settled on Recurrent Neural Networks (RNNs). In particular, doc.ai uses LSTM-based networks for natural language processing tasks.

LSTM stands for Long-Short-Term-Memory and is based on the idea that each cell in a network has a memory. Instead of just computing an output signal, an LSTM cell has multiple gates that control its behavior. Based on the signals on those gates, a cell can use either the input signal or its memory to calculate an output signal.

LSTMs have become important building blocks of many AI-based technologies today. Google, Apple, Amazon, and others are using it in their speech recognition and translation services.

Re-envisioning clinical data trials

De Brouwer emphasizes that digital medical records open the door for a myriad of useful applications. He mentions clinical data trials as an example: “Today, a clinical trial [might cost] USD $57 million and take 2 years. What about a data trial where people already have their data and could agree to participate in no time?”

This development could lead to significantly shorter times for drugs and treatments to reach the market, as well as reducing the price for research.

With their medical data at their disposal, patients could participate in trials from their phones. De Brouwer envisions a marketplace for data trials, where patients can choose the trials they want to participate in and get paid to do so.

The data would be anonymized on their own device, so that sensitive information would never even enter the cloud. “That’s an income for people from their medical data,” he says. “If [my medical data] is anonymized and I do the encryption and I do the opt-in, I don’t have any problem with [sharing my data].”

Another area where De Brouwer sees huge potential for doc.ai is research for rare diseases. “One of our focuses will be to enable data trials to help with rare diseases,” he tells me.

He believes that gathering and structuring data from rare disease cases can make it more cost-efficient to do research on them. “It’s too expensive for a human to work on a rare disease that only 100 people in America have,” he explains. “That’s where AI can help dramatically.”

Spreading love and knowledge

doc.ai’s mission is to simplify medicine for both patients and doctors. De Brouwer believes that letting AI handle the tasks of structuring, safekeeping, and maintaining our health records may be the key to make healthcare more affordable and enable it to focus on its core task again: to make people well.

He is certain that removing complexity and frustration will benefit doctors and patients alike: “There is one thing that people want more and more: love. You have to spread love.”

With a bag full of revolutionary ideas, Walter De Brouwer is on a crusade to simplify medicine for people, so nobody has to feel as powerless as he did in 2005.