A startup out of Taiwan has developed an AI platform capable of rapidly developing customised AI models that can analyse video and gigapixel images in seconds.
What do you do when you’re faced with the prospect of losing decades of on-the-job experience and specialist knowledge through the retirement of one of your top employees?
Have you considered capturing their knowledge in an AI model?
At least this is what a startup out of Taiwan is proposing with their Auto AI High Performance Computing (HPC) image analysis platform.
Preparing for the brain drain as a senior staff member retires is an inevitability faced by all organisations. From incentivising the transition through part-time contracts to developing a ‘culture of sharing’, there are many tactics a company can employ to mitigate the loss. When managing with senior doctors, such practices haven’t always proven effective for hospitals.
Ai Explore, an image and video analysis startup have developed an Auto AI platform which can build customised AI models fast and efficiently. Their super-fast and adaptive deep learning framework requires as few as 30 annotated images and can be continuously improved.
“I’m currently talking to a children’s hospital in Thailand which has a very experienced doctor who is about to retire. This platform will allow the hospital to capture his knowledge so that it can continue to benefit their patients,” said Prof Ching-Wei Wang, founder of Ai Explore.
“A traditional AI company will only focus on single recognition models, for example, they will try to create a model that can only recognise breast cancer tissue or lung cancer tissue,” says Wang. “This is because their AI engine isn’t efficient and relies on huge data sets which need to be annotated, only then can an engineer manually create a model. This often results in a single AI model taking over a year to be developed”.
For Ai Explore, it’s all about being fast. Providing the AI infrastructure for real-time analysis of gigapixel images, a new AI model can be developed by a doctor in roughly two days.
Realising that the effectiveness of each model is reliant on the expertise of its creator, Ai Explore place their platform in the hands of “doctors and clinical scientists who have domain knowledge”.
It’s all about the know-how
Last year saw Ai Explore went toe-to-toe with some of the world’s leading AI teams at IEEE International Symposium on Biomedical Imaging (ISBI). Competing teams were tasked with developing an AI model which could accurately detect and segment lung carcinoma using only 200 annotated images to train their models. Landing a top 10 finish, the team from Ai Explore defeated over 400 teams.
They managed to replicate this feat with another top 10 finish in a liver cancer segmentation challenge at MICCAI 2019. This time defeating a whopping 900 teams.
To date, Ai Explore’s Auto AI has outperformed reputable names such as Pfizer, Leica, 12 Sigma and the University of Tokyo.
Wang remains coy when pressed for details pertaining to their Auto AI platform. When asked how her team manages to outperform much larger companies, she stipulated that it’s “the know-how, not the scale” that matters.
Part of what makes Ai Explores platform so fast is their expertise in data management.
Wang said, “when you use Google maps, you will notice some delay due to the large amount of data it’s using, this is the same for microscopic image data. Today, large international brands such as Microsoft and Leica still struggle to display these images, especially when a user starts to zoom in. This is because their data management systems aren’t efficient enough”.
AI as a commodity
Trust is a fundamental part of any patient‐doctor relationship. Comparatively, AI has yet to win over the hearts and minds of the general public with a recent study showing “Americans, on average, expect that high-level machine intelligence will have a harmful impact on balance”.
What does this mean for AI in healthcare and the public’s ability to trust its accuracy?
Wang takes the stance that, since the AI models are developed using a doctor’s expertise, patients will simply need to consider the same factors they currently do when choosing a hospital.
“When we first started building AI models for breast cancer, we were communicating with two hospitals. We found that when the knowledge of the doctors varied between the hospitals, this directly impacted the models they developed”.
It’s true that not all hospitals are made equal. Your ability to access top quality healthcare facilities is often predicated on the affluence of your community.
As a means to mitigate the disparity in knowledge between hospitals, Ai Explore has established a profit-sharing scheme which incentivises hospitals to sell their models. This could prove to be a great leveller in patient care. Models developed by a nation’s finest doctors could become widely accessible as hospitals buy and sell AI models as a commodity.
At a time when Medical AI is pushing for adoption, Wang stressed the importance of creating reliable solutions. Wang describes how a hospital in Taipei recently “declined IBM Watson” after a trial period. “They felt the AI model was better suited for Europeans or Americans”. This was because the “suggested dosage would have been potentially lethal for an Asian patient”.
Moving beyond digitalisation
When it comes to healthcare and the push towards digitalisation, Wang believes it’s all about finding the right carrot.
“Doctors can be reluctant to adopt new technologies as they don’t have the right motivation. If it’s just pure digitalisation then it’s just extra work. But if you tell Doctors that you’ll have an AI system that will reduce your workload, then they’ll do it”, says Wang.
Having previously adopted the digitalisation of medical records over 20yrs ago, it’s safe to say that Taiwan isn’t afraid of being a trailblazer. Today, Taiwan are taking it one step further, “the Taiwan government doesn’t just encourage digitalisation, they encourage AI, it’s a big difference,”says Wang
Taiwan’s government have implemented a novel way of allocating funds. Hospitals will now need to adopt efficiency saving technologies to attain points which can be cashed in for additional funding.
Digitalisation for the sake of digitalisation will no longer cut it, hospitals will need to prove that they are using technology to speed up the diagnosis process, reduce staffing costs and to increase quality of care.
This push towards automation has aided Ai Explore in deploying their Auto AI platform in a hospital to automate cervical cancer detection.
Annual cervical cancer screening is provided to all Taiwanese women aged 30 years or older. The samples obtained through the screening process go through a two-stage process of examination performed by a clinical scientist and a pathologist. On average, these specialists are able to check 50-80 samples a day, taking roughly 6-8 minutes on each sample.
Specialists can struggle from both fatigue and time pressure, resulting in mistakes. The advantages of using AI to automatically review microscopic images is immediately apparent.
That’s before you even consider that the AI can operate 24/7.
More than a medical solution
Ai Explore’s platform has now reached a point that it can be quickly adapted to meet new challenges. This has led the startup to modify their solution for different industries.
“We’ve been approached by NEC in Japan to develop a licence plate solution that can work in heavy rain and at night,” said Wang. They’ve also been working on a concealed objects solution: “In Thailand they have terrorist concerns and are fearful of people throwing bombs. They’ve requested a solution which could reveal concealed weapons”.
If you’re interested in seeing their platform in action, check out their videos on YouTube. Ai Explore keep their YouTube channel up-to-date and can be found here.
Wang believes that HPC is making conventional AI model development a thing of the past.
“Large Companies are still hiring 100s of AI engineers and relying on manpower, almost like an AI engineering factory. These engineers set up their own system and models independently then when they leave, the knowledge leaves with them,” says Wang. “If you’re still using the conventional approach then you won’t be able to compete”.