Face recognition technology hunts for rare diseases
FDNA Inc. helps doctors to diagnose rare diseases with a camera of a cellphone
Machine learning algorithm in rare diseases medicine
Approximately 90% of all medical data are stored in the form of images [1]. The technical progress did a lot to come inside of a human body and go down to the level of cells and even molecules to make a diagnosis. However, a lot of information could be received from a channel, which historically had a major role in medicine – from visual inspection of human body. And here AI could help also.
Image recognition technology, one of practical applications of machine learning, is not just an important megatrend for FDNA Inc., it’s simply the reason of existence. The founders of the company, sold their previous startup, Face.com to Facebook [2] and then turned with the same technology to medicine and rare diseases in particular with an application named Face2Gene. The underlying idea is to analyze the photos of patients’ faces and assess the probability of rare diseases. The expertise of the founders is a great aid in that task, however, company operates in a highly competitive environment. For instance, IBM bought a Merge Healthcare, coming to the industry in a search of application for IBM Watson. This “winner takes it all” setting influences the product development process.
Product innovation approach: high focus
The company is razor focused on the main solution in allocation of the limited financial and managerial resources. Although there is a number of ideas how to apply image recognition in medicine (for instance X-Ray or MRI scans analysis), management is not distracted Such concentration allows company to overtake competitors and make a strong bet for the product to become an industry standard. Currently, around 200,000 patients [3] are evaluated by clinicians using Face2Gene and company claims that it is used by 60-70% of geneticists [4].
The logic of new products selection could be considered as another design decision for innovation process. New projects selected to reinforce position of the main product and have low feasibility risks. For instance, one of the applications that FDNA offers is Forum, which allows collaborating with other doctors on undiagnosed cases on a platform resembling a social network [5]. That instrument gives the doctors tool for effective communication and discussion around diagnosis and creates a professional community around the main product. At the same time project uses the technologies and solutions already existing on the market and available for quick implementation and scaling, thereby it doesn’t require a lot of effort and could even be outsourced to a third party.
Innovative companies often face difficulties integrating a number of innovative solutions that were generated on a bottom level of organizations. The nature of innovations assumes the iterative approach to the development and hinders top-down design. The integration of those separate initiatives into a united whole could be time and resources consuming process. However, the is a recipe to alleviate that problem by making a common platform and declaring common principles and interfaces. FDNA successfully applies this approach by developing all products on the basis of one platform Face2Gene. [5]
Next steps: data hunger
There is a number of risks and opportunities that could be decisive in the future of the company.
Usually, machine learning algorithms need data to raise the accuracy of prediction. That’s why access to the maximum number of photos is a key success factor. Mechanism of gathering such data and transparent rules for partners is an important infrastructure need for the company. Today the product is free, what means that customer doesn’t pay money for the benefit of using that. However, customer provides data which is more important in the context of fierce competition and relative availability of sources of capital.
The data source in the developed countries will likely be a battlefield for the data between digital health startups. However, the range of sources could be extended by reaching to potential suppliers in developing countries. Although their current focus is on application of already known solutions in high-impact directions such as cardio-vascular or infection diseases, the interest to rare diseases will grow. Joining the already existing educational programs of non-governmental and pharmaceutical companies is a greenfield investment in new sources of data.
The transformation of a service to a platform with access of third parties to the engine via API could be an important direction in development of the company. The funnel of new initiatives will not be limited only by ideas generated internally. An example of such step is Amazon Rekognition tool that allows using the face recognition technology to users of AWS [6].
Question
Startups often work on a competitive edge of technologies, that requires great dedication and focus on the main product. How to prevent negative impact on motivation of employees from such constraints on creativity?
References
[1] “Why IBM Just Bought Billions of Medical Images for Watson to Look At”, Mike Orcutt, TechnologyReview, August, 2015
[2] “10 Face Recognition Acquisitions”, 10 Face Recognition Acquisitions
[3] “Your face could reveal if you have a rare disease”, Joao Medeiros, Wired, 30 April 2017
[4] “FROM FACEBOOK FACIAL RECOGNITION TO GENETIC DISEASE IDENTIFICATION”, Eytan Halon, The Jerusalem Post, 16 October 2018
[5] Company website: http://www.fdna.com/face2gene-for-geneticist-healthcare-providers/
[6] Amazon product page https://aws.amazon.com/rekognition
I can see your point, that focusing on one existing product dissuades employees of a startup from dreaming up solutions to new problems, but for FDNA this focus makes sense. The information it stores is extremely sensitive. The inputs to this algorithm, patient images, contain classified medical information and are probably hard to anonymize. With such a high-risk product, I think specialization is key.
This is a great overview of the company. I’m interested in learning more about how the product actually works and where it adds value – rare diseases are generally diagnosed with genetic testing and I would imagine that patients who use this product would still need to undergo confirmatory genetic testing. Rare diseases are also by definition rare, so I’m curious to see whether they can acquire enough images to improve accuracy and enough patients to stay sustainable.