Close this search box.

Encord: Stringing a Powerful Cord of Futuristic Data-Centric Computer Vision

Eric Landau | Encord

Innovations do not have to be always solving huge problems. Rather powerful innovations are those which solve everyday issues for those who are the most important social contributors like healthcare givers. This is the story of such an innovative company which is stringing a powerful cord between technology and healthcare professionals by using futuristic data-centric computer vision.

Doctors in the Division of Nephrology, at Stanford Medicine, were facing a complex problem. They had to use three different kinds of software to identify, annotate, and count podocytes and glomeruli in microscopy images. Similarly, King’s College London had been using clinicians to annotate pre-cancerous polyp videos, had prohibitively high costs to produce large datasets. Likewise, Memorial Sloan Kettering Cancer Center staff were finding detecting and classifying vena cava filters in complex label structures (ontologies) rendered existing and open-source tools unusable.

Simply, in the words of King’s College’s Director for Gastroenterology and Endoscopy, this has been like mental torture to them. As the director and many more heads of such institutions were living in the agony thinking they had to endure it lifelong, came along Encord and using the power of advanced technologies like AI and ML, first its Co-founder and CEO, Eric Landau convinced Stanford to use Encord’s annotation tools and SDK to automate segmentations, count, and calculate sizes of segments. The result was astonishing, as with Encord, Stanford researchers reduced experiment duration from an average of 21 to 4 days (80%) while processing 3X the number of images.

Eric also convinced Dr Bu Hayee, the Director from King’s College to deploy Encord’s micro-model module to increase clinician labelling efficiency and automate 97% of produced labels. Again the result was outstanding with the highest expense clinician seeing 16X labelling efficiency improvement, cutting down the model development time from one year to two months and a 6.4X average increase in labelling efficiency for GI videos.

Furthermore, MSK adopted Encord to build custom label structures for pulmonary thrombosis projects, thus making them entirely feasible for the company. Today, Machine learning and data operations teams of all sizes use Encord’s collaborative applications, automation features, and APIs to build models and annotate, manage, and evaluate their datasets.

In Eric’s words, “We work with some of the world’s leading AI companies, enterprises, hospitals, and research institutions.”

Elaborating on his own journey Erics shares that during his undergrad, he studied physics. He began a PhD but left to take a job in finance. After almost a decade of working in finance, Eric was ready for a new challenge. He joined an entrepreneur network here in London, which is where he met Ulrik. Ulrik told Eric about his seed idea for Encord, stressing the need to solve the data bottleneck by automating data annotation in medical AI and beyond.

Erics reveals, “I thought it was the best startup idea I had heard, and it was a lot better than the ones I had had, so we founded the company and began solving the problems associated with putting together and managing high-quality training data for machine learning models.”

In an interview with Insights Care, Eric spoke at length about the inception, the journey and their future plans for Encord. Herein are the highlights of that exclusive interview.

Eric, please brief our audience about Encord and kindly tell us the source of inspiration for venturing into the medical imaging niche.

We are an early-stage startup that provides a computer-vision-first platform for annotating and managing training data. When my co-founder Ulrik was studying at Imperial College London, he worked on a project where he was visualizing scale image datasets for medical AI. Through this work, he realized the difficulties that doctors face when labelling these images. Hand labelling medical images take an enormous amount of time, so Encord started off working in the medical imaging field because he understood how suboptimal the data annotation process was, and how it could be done a lot more efficiently if doctors had a technological solution for it.

Can you elaborate upon the core values on which Encord is built, what is its vision and mission?

At Encord, we value curiosity and a growth mentality. Our team prioritizes humility– being willing to learn from mistakes– and resourcefulness–being scrappy and creative when solving new problems– which enables them to develop great products and deliver high-quality customer support.

Encord’s mission is to accelerate the pace of data-centric AI in medicine so that healthcare institutions and medical professionals can reap the benefits of this technology. AI has the potential to make healthcare and medical treatments increasingly proactive rather than reactive.

However, until recently, the AI world has been very model-centric, so healthcare professionals haven’t been able to use AI to its full potential. By taking a data-centric approach to AI and focusing on improving the quality and quantity of the training data used to train medical AI models, we can eliminate potential demographic biases and make model predictions as realistic as possible to what happens in a clinical setting. Because we want medical professionals to be able to use AI in practice in a way that’s consistent with the values of the medical profession, Encord has focused on delivering an efficient, user-friendly platform that allows for all the security, privacy, and integrity expected when handling patient data.

Kindly elaborate upon the services that make Encord stand out as one of the most advanced medical imaging solution providers in the modern healthcare industry.

We are technologically focused. We knew this problem needed a more sustainable, long-term solution than the current solutions which were built upon improving manual labelling processes or increasing the amount of manual labellers. Rather than continue to throw a lot of doctors at the problem, we wanted to create a technological solution that used doctors’ time as efficiently as possible. The micro-model technology we developed enables doctors to automate data labelling with just a handful of manually labelled images. We are the only company that provides an automated solution to the problem of annotating training data.

We are also very hands-on from a company-to-consumer perspective. We don’t throw one-size-fits-all solutions at a problem; instead, we really try to understand each individual customer’s needs to make sure they get the solution most likely to work for their existing issues.

What is your opinion on healthcare providers’ aligning their offerings with newer technological developments, especially when it comes to catering to the dynamic needs of the healthcare space?

We are technologically forward-thinking, so we encourage healthcare providers to align their offerings with newer technological developments. That said, those technological solutions also have to work and deliver real-world value, so the technology shouldn’t just be used because of the hype surrounding it. When it comes to machine learning, this is where the power of data-centric AI really comes into play.

A healthcare provider could use a very sophisticated model, but if the model hasn’t been trained on a vast amount of data that is reflective of the patient population, then the technology and its outputs will be fundamentally flawed.

Healthcare providers need to put processes in place that validate the technology and make sure it delivers valuable, real-world patient care. Conditional upon that, they shouldn’t be shy about or afraid of using technology because it can be immensely useful in delivering patient care.

What advice would you like to give to the budding entrepreneurs and enthusiasts aspiring to venture into the medical imaging niche that you are catering to?

Don’t assume you know the answer to the problem or even the problem itself. You need to talk to a lot of people to understand the problem and its solution. Don’t be shy about talking to people.

Talk to doctors, nurses, AI professionals, and as many other stakeholders as you can. Really listen to them to understand their experience and the challenges they’re encountering. A lot of entrepreneurs will speak first and say, “This is what I think the problem is,” but that’s not the right approach.

First, you listen, then you synthesize the information you’ve learned in all of these conversations to understand the problem, and finally, you begin to formulate a solution from that starting point.

How do you envision scaling Encord’s services and operations in 2022 and beyond?

We’re going to continue the growth of the team and expand geographically, with a particular focus on the U.S. market where we are working to establish a stronger presence. We also have exciting new healthcare-focused partnerships in the works, with some of the world’s best academic institutions and industrial AI partners.

Encord’s Hall of Fame

Encord was named by CB Insights as one of the 100 most promising private AI companies of 2022. In this interview, King’s College Hospital’s Director for Gastroenterology and Endoscopy Dr Bu Hayee discusses how Encord’s tool eliminated the “mental torture” of manual data labelling.

In this NVIDIA post, Fareedc Iqbal, the CEO of SurgEase, a company that offers telepresence technology for gastroenterology, said, “Encord’s software has been instrumental in aiding us in solving some of the hardest problems in endoscopic disease assessment.”

Encord’s technology has been used by Stanford University, King’s College London, and Memorial Sloan Kettering. Case studies with links to academic papers are available on Encord’s website



Copyright 2023 © Insightscare Magazine ( a Digital Ink brand ) All rights reserved.