A chat with Denti.AI founder Dmitry Tuzoff

A chat with Denti.AI founder Dmitry Tuzoff

“The market is telling us that they don’t want the AI to compete with humans.”

I had the pleasure of interviewing Denti.ai founder Dmitry Tuzoff. Denti.AI is a cloud service using deep learning technology to analyze dental images and help patients, dentists, and insurers. The company collaborates with research institutions around the world (including Canada, India, Switzerland, China, and Russia, amongst others) to develop technology with proven effectiveness for commercialization.

What is Denti.ai’s value proposition?

Denti.AI focuses on automatic analysis of dental images to help dentists. We started from x-ray images to panoramic images. The aim is to visualize the complete dental record, this means including all image information such as cavities, lesions, cysts to help reveal cavity pathologies.

A common challenge faced by dental health practitioners is that currently about 30% of cavity pathologies go undetected, and hence they go untreated. Dentists miss out on the opportunity to treat the patient earlier on and effectively. Patients, in turn, face much higher bills if things go detected, and often end up dealing with implants with easily 5-10 times the cost. Sometimes it may end up being too late to do anything at all. Using AI, we can lower the cost by up to 5 times if the analysis is performed automatically. The AI can detect more potential pathologies.

So the value proposition is to combine the expertise of the specialist with the efficiency of our automatic software. The key benefits are:

  • Increased availability of basic automatic service
  • Increased number of pathologies detected earlier
  • Visualization of information for the patient
  • Provide data ownership, directly on the cloud, and can be forwarded to any dentist
  • For the dentist – this means more leads and a better chance of conversion to treatment

To give you an example, a dentist spends a few minutes on average analyzing and describing a single radiograph. This may not seem like much, but over time this adds up to 25-30 business days a year. With Denti.AI, interpretation of one dental image will only take 4 seconds. This includes visualization of information about earlier treatments and new pathologies and generation of comprehensive tentative reports.

Furthermore, Denti.AI will point out any deviations from the norm. Its operations and the likelihood of finding pathologies are not affected by how tired the dentist is, or how well he or she feels. In effect, dentists will have access to a second opinion, based on the experience of professional radiologists who participated in the training of our algorithms. The more users the system has, the better it will “learn” about all sorts of subtleties that no single human professional can master in full.


What form of AI works underneath the hood for Dental.ai?

Our solution is based on deep learning, a form of machine learning that is commonly used to process images, video, audio or similar perception tasks. The most well-known application of deep learning is probably driverless cars. The math is actually similar across applications, essentially you break it down into a task of object detection and then another task for recognition. For detection of pathological sites, we focus on areas of potential pathologies – to advise the dentist on which areas to investigate. We do not provide medical advice, but rather we provide a form of computer-aided diagnostics.

What data do you use to train your AI?

We invested a lot in the data and the size of our private dataset is orders of magnitude higher than what is available to most researchers. This allows us to deliver high recognition and prediction accuracy and bring practical value to our users, both professionals and researchers, now. And the more users we have, the better the quality gets.

In order to train our deep learning AI, we collaborated with a company that specializes in data labeling. They helped us obtain the labeled data to train with professional radiologists and clinicians. Once the model was trained, we tested the performance against human experts for validation purposes.

What is your go-to market strategy?

Clinics around the world use our software to perform the computer-aided-diagnostics. The challenge is that there is no single standard for diagnosis. Different countries have different local standards for diagnostics – Eastern Europe has very different diagnostic standards from North America, for instance.

Our approach is to not rely on one school or standard, but to provide localized modules for each region. What this means is we train different models using localized sets of training data to cater to different needs. I have heard that IBM is criticized for training Watson as a one-size-fits-all product that tries to impose North American standards over EU standards – who knows whether that is intentional or not? Regardless, it is really important for our application to be seen as catering to local standards.

What’s the competitive landscape like?

We target the general dental practice. We see some equipment vendors are also trying to do what we are doing. I believe in 1-2 years there will be even more products in the space.

We compete by:

  1. Focusing on the needs of general practice dentists, we understand their requirements better. Unlike specialist companies such as CEPHX, which focuses on orthodontics, or Orca which focuses on dental radiology.
  2. We develop direct relationships with clinics and dental practitioners, rather than through equipment vendors. We believe a direct relationship is important.

The market’s response has been encouraging. Most dentists and radiologists are wowed. We are seeing great response from Canada, Hong Kong, India, and Russia. We are holding focus groups in the UK and USA. The market is telling us that they don’t want the AI to compete with humans. Overall we sense there is a level of reservation from the market before fully embracing AI technology. We are piloting projects, and working on the feedback. We are getting good responses. Many clients simply don’t know enough about the AI technology and its potential.

Where do we go from here? What is the future of AI?

Right now we are seeing a lot of expectations in the fields of cancer research and cardiology. I believe the time for people to figure out a proven business case for AI in the medical sciences is yet to come. There are many barriers to overcome – such as ethical issues, conservatism and lack of knowledge of what AI is and isn’t.

For us, we are able to learn from all of this. We are watching the trends in consumerization of the technology. Some startups provide telemedicine, medical diagnosis. Even in orthodontics, we see a trend of consumerization.

We also see a trend of establishing a direct relationship between patient and the technology. We start with the people who can interpret the analysis, then reach out to the patient. Together with the clinics, I can see a day where the patient uses our AI to do field-based screening of dental diagnosis, right through an app. We are already seeing lots of research in fields such as blood tests and telemedicine. In these areas there is evidence that field-based apps can be really effective. Especially in helping broaden the availability of medical assistance to underserved populations.

Thanks for your thoughts Dmitry! I really enjoyed this interview and wish you best of luck!

If you enjoyed this blog post, please remember to share or like!

The post "A chat with Denti.AI founder Dmitry Tuzoff" first appeared in https://aiartisan.wordpress.com on March 3, 2018

要查看或添加评论,请登录

Maurice Chang的更多文章

社区洞察

其他会员也浏览了