Google’s announcement of the culmination of three years of work and a subsequent move into skin disease analysis at their developer conference earlier this week has led to a number of people approaching me to ask what does this mean for Skin Analytics?
Spoiler alert: it’s great for us.
With a market cap of over $1.5 trillion and a history of aggressively disrupting traditional industries, on the surface of it Google’s announcement presents a challenge to other AI skin cancer companies. Let me explain why Google’s work in skin cancer is ultimately a very good thing for patients and for Skin Analytics.
What do patients do next?
We work in a nuanced and complicated field and it’s fair to say that Google’s solution works very differently from Skin Analytics. While we have always focused on supporting clinicians to make better decisions for their patients, Google’s product will be launched directly to the public in Europe through Google search.
The key question is, what do you do with Google’s output? The way we think about it is to ask ourselves what we did when we asked Google if we had coronavirus. We booked a test and spoke to our GPs.
Google’s initiative has a huge scope and reach to help people identify when to worry about a skin lesion. But it’s not designed to provide a definitive answer. The next step is to visit a doctor and that is where Skin Analytics technology sits.
What Google is doing in their own words
Our AI-powered dermatology assist tool is a web-based application that we hope to launch as a pilot later this year, to make it easier to figure out what might be going on with your skin. Once you launch the tool, simply use your phone’s camera to take three images of the skin, hair or nail concern from different angles… The AI model analyzes this information and draws from its knowledge of 288 conditions to give you a list of possible matching conditions that you can then research further.
Critically they clarify the intended use:
The tool is not intended to provide a diagnosis nor be a substitute for medical advice as many conditions require clinician review, in-person examination, or additional testing like a biopsy. Rather we hope it gives you access to authoritative information so you can make a more informed decision about your next step.
Of course, what Google is doing right now isn’t necessarily what Google may do in the future and they may well decide to take a more active role with healthcare providers.
It’s all about the clinical evidence
In this case, to share the difference in what we do and what Google skin disease search might do, let’s have a quick look at the evidence Google has published.
Their most recent paper showed results for 26 skin conditions and one group for the long tail of other skin conditions. These conditions ranged from things like acne or tinea to cancers such as basal cell carcinoma and melanoma.
The paper shows a net overall improvement in accuracy for primary care clinicians and nurse practitioners in diagnosing skin conditions. However, this doesn’t hold for the specific cancers within those diverse 288 skin conditions.
Cancers were acknowledged as a limitation of the paper with Google’s authors stating:
… AI-associated improvements for malignant neoplasms were lower than those across all cases, and future work is needed to further improve the AI tool for malignant neoplasms.
Can you assess skin disease without assessing skin cancer?
In the interest of brevity, let’s just look at the most aggressive skin cancer, melanoma. This cancer is responsible for the vast majority of skin cancer deaths around the world and has long been the primary focus of health systems.
The graphic below is from Google’s supplementary information for the paper. I am happy to break this down into more detail if anyone would like (just leave a comment) but to make my point, the last thing I need to tell you is that Google’s algorithm spits out a hierarchy of options that the lesion could be. So it may return up to three different answers about your what your skin condition is.
The chart above tells us that for melanoma Google’s algorithm would find less than 15 out of 100 melanomas as the first output label.
It also tells us that every type of clinician they assessed (nurse practitioner, primary care clinician, dermatologist) would identify fewer melanomas using the AI assistant, than without.
If instead we accept a melanoma classification from any of the top three suggestions, Google’s algorithm would find a bit less than 60 out of 100 melanomas. However, clinicians still perform worse with the AI assistant than without.
It should be noted that for melanoma and most other conditions the clinicians benefit from the AI assistant to varying degrees. But this highlights the difference between what Google’s tool is trying to do and what it is good at.
At Skin Analytics, we specialise in finding skin cancers and support clinicians with very high sensitivity services. To be accepted within the clinical community we have to find a high percentage of the melanoma we see.
Our own JAMA paper released in 2019 showed that we achieved a sensitivity of over 95% in a powered prospective study run in seven UK hospitals. It is just one of five studies underway or out for publication and a further two under development.
Then why is this good for patients and Skin Analytics?
Well, dermatology demand is higher than our workforce’s capacity to service it. What Google is doing is creating a more useful way for people to gauge their level of concern and provide the motivation for seeking help.
If we look at the time it takes to find and treat cancer, by far the vast majority of this time passes while between our ears we are deciding whether to seek help.
Google’s tool provides a real time answer to that question and will result in more patients seeking help. We at Skin Analytics will continue to implement high sensitivity services to support those clinicians better serve their patients.
We might even take a call from Sundar to explore how we can benefit each other…