What can AI healthcare startups learn from the recent IBM Watson-Anderson fallout?

share this article
Share on facebook
Share on linkedin
Share on twitter
In this article

Four years ago, MD Anderson, global pioneers in cancer care announced collaboration with IBM Watson to power ‘Moon Shots’ – a mission aimed at ending cancer by using the former’s cognitive computing system. The ambitious system, ‘IBM Watson powered Oncology Expert Advisor’ would integrate MD Anderson’s knowledge of clinicians and researchers to advance the goal of treating patients with the most effective, evidence-based and safe standard of care.

However, few months ago, Anderson announced a fallout of this project.

Watson Health’s broad offerings

With a mission to “improve lives and enable hope”, IBM is attempting to solve diverse healthcare challenges to improve diagnosis outcomes as well as engage in medical research. Their vast product portfolio includes precision medicine in genomics, ‘Watson for Drug Discovery’, health patient engagement, oncology and care management and automated diagnosis in care delivery and research.

Has it created any real difference on the ground? It’s tough to ascertain, yet.

With investments to the tune of USD $4 billion already, Watson Health is an extremely comprehensive project, in terms of application offerings and integration in the existing systems. However, Watson Health’s complexity has so far only acted as a bottleneck and not an enabler in adoption or proving noteworthy utility of any kind. One plausible reason, is that the entire project is much ahead of its time.

Yet, one can’t deny that the healthcare sector in emerging and developed markets has tremendous opportunity to optimize care delivery, outcomes and costs using cognitive computing. And keeping this in mind, IBM’s bet on this sizeable market makes sense.

But, does Anderson deal’s fallout and Watson’s complex nature of system strongly indicate failure of cognitive computing/AI in Healthcare? Or simply put, is the need of the hour different?

With it’s broad-based end-to-end solutions, covering a plethora of diseases and aiming to replace doctors, Watson Health could be a play in distant future. But today, we strongly believe, incremental-focused niche solutions are need of the hour. Intelligent engines independently focusing on discrete problem spaces that exist within the healthcare delivery value chain, would be more impactful in near future, with them successfully solving problems bottom up and acting as a fundamental baseline in creating a broader, connected offering.

Watson’s most prominent offering is providing a second opinion to a physician’s diagnosis on a multitude of diseases by aggregating data from zillions of clinical journals and correlating it with patients’ data and suggesting the best alternative treatments. All this comes at an additional cost and does not, by any means, replace the need of a physical doctor. The adaptability thus has been difficult

On the other hand, smaller innovative startups are showing better signs of adaptability through focused solutions-

  • Israeli startup, Medymatch is building deep and advanced cognitive analytics, in stroke diagnosis emergencies. Medically, when a stroke patient arrives to the hospital, doctors have an extremely small window to make a right diagnosis, often resulting in 30 per cent misdiagnoses. Medymatch’s offerings reduce the 20-minutes reading time to just 2 minutes, significantly increasing efficient and accuracy of diagnoses.
    asd
  • Another startup, US-based Arterys – a SaaS analytics platform to revolutionize medical imaging, visualizes and precisely quantifies blow flow support to cardiovascular disease diagnostics using MRI and cloud computation, which currently, is an extremely tedious task as segmentation and measurement need to be performed manually, leading to no reports at all.

Medymatch and Arterys establish effectively that different clinical streams have a wide variety of problem spaces and highly-targeted solutions not only provide better value proposition but also ease of integration and adaptability by the end-users i.e., consumers.

The Indian Opportunity

In developing markets like India, a major, untackled bottleneck is lack of trained doctors for the exploding population, resulting in increasing disease burden. In a market like this, efficiency improvement not only means improved diagnosis outcomes but also increase in healthcare access by increasing physicians’ and providers’ efficiency, indicating double-impact potential.

Ranging from building intelligence in point-of-care devices to enhancing productivity of physicians and other care providers, there are plenty of white spaces waiting to be solved. We believe, AI/Cognitive Computing has significant potential to enable faster access and greater impact on the lower-end of healthcare delivery, much better than higher-end precision medicine and Genomics.

Telemedicine, for long, was considered the holy grail for increasing last mile access to healthcare. Yet, there are holdups – lack of trained on-ground assistants’ inhibit effectiveness of telemedicine models. Innovative, cognitively trained systems that could aid telemedicine and enhance care through PoC (Point of Care) service delivery would be game changers in these markets.

Imagine this situation – a fetal monitoring device available in a remote corner of India, whose utility is dependent on a trained on-ground worker to integrate with an end-to-end system – if the same device also had the ability to perform first-level screening (example – detecting hypoxia using fetal heart rate and uterine pressure), a timely signal sent to the closest care provider could avoid life-long disability, death even. AI-based systems can effectively achieve this.

Imminent white spaces

Areas such as Radiology and Pathology could immensely benefit with AI, which can assist the radiologist with workflow solutions that can enhance the efficiency of diagnosis enabling a single radiologist, to cater to multiple patients. The beauty isn’t mere automation but plummeting misdiagnoses and improving accuracy. Of course, this depends on precise training of the neural networks, which in turn depends on accurate and specific data.

India-based SigTuple enables automated analysis of peripheral blood smear, urine and semen sample, retinal scans and chest X-rays. It’s device, ‘Shoonit’ provides peripheral blood smear analyzer solution and generates reports on the differential blood count for pathologists’ review.

Another startup, Predible Health, is leading the radiology space with building solutions around improving efficiency and accuracy of radiologists and physicians. Predible’s deep learning algorithms are targeting advanced insights in the Oncology space, powered by a neutral cloud computation platform.

What will it take for startups to succeed in AI/Cloud Computing?

We see four key attributes that are essential to succeed in this space –

  1. Data Strategy – Data is the king. Having the right data acquisition and processing strategy is critical to building a sustainable business model and ensuring optimum accuracy. For this, procurement, ensuring variability in data and annotations are key. Given the volume and variety of data availability, India can be a strategic place to train these systems. The challenges could lie in sourcing, locating data, digitization and standardization but, there are numerous private and public initiatives that are riding the digitization wave and bringing it up to consumable levels.
    asd
  2. Business Model – Healthcare industry is a price-sensitive industry, a physician driven-market. Hence, effective models should have two differentiators – no increase in costs for the end consumers and a compelling value proposition for care providers to part away with a certain percentage of their revenue. Licensing models, pay-per-use, monetizing unit efficiency improvement are the most viable models.
    asd
  3. Integration Capabilities – In cases of workflow solutions and SaaS, ability to swiftly integrate with existing software and hardware systems from giants such as GE and Philips, especially in the workflow solutions, will define the scalability and expansion of any product.
    asd
  4. Regulatory Approvals – Lack of definition of regulations is a clear challenge. Currently, only FDA approved Deep Learning based algorithm is of Arterys. While most other players are positioning themselves as CDSS (Clinical decision support system) and don’t require FDA, it will be important to keep a tap on evolving regulations (both FDA and in India) well in advance to avoid later hiccups, as momentum in this domain is increasing and so are the definitions of regulations taking shape.

Players should try to leverage the increasing amount of data that the Indian patient base generates to train these systems. Leveraging lower cost in training and deployment and taking solutions from developing to developed markets is an exciting possibility, waiting to be tapped.

At Unitus Ventures, we foresee AI/Cognitive Computing to be defining the future years and are keen on inviting potential fundable early-stage startups. Please have a look at our criteria and make an application, today. We look forward to hearing from you!

Subscribe to

Our monthly roundup of funding initiatives, meetups
and latest trends in startup tech.

Subscribe to

Our monthly roundup of funding initiatives, meetups
and latest trends in startup tech.