Science for All .#1

What's the connection between Alphabet, diabetic retinopathy and a wearable lens?

Simple. A lens has been made by Calico Health Sciences, subsidiary of Alphabet Inc.The same lens can detect diabetic retinopathy better than a certified and experienced ophthalmologist.

Now you must be wondering, HOW ON EARTH  DOES THIS THING WORK?

Before you gain an effective understanding of how Calico Inc's lens works. We need to understand the approach that many research teams successfully employed before this. 

The most prominent and successfully employed method,

before this ,was the method of automatic detection of diabetic retinopathy using SIFT or Scale Invariant Feature Transform. SIFT is a computer algorithm which basically helps spot specific and localized features in high detail. 

I won't get into the finer details of this study as it requires some patience and concentration to assimilate all the detailed information in the study. 

Here's the link if you're interested:

http://ai-d.org/pdfs/Silberman.pdf

They were two difficulties with this process. One, this process wasn't smart enough to pick up intricacies that make the difference between actually having diabetic retinopathy and being misdiagnosed as diabetic retinopathy. For example, when fundus photography(detection technique for diabetic retinopathy) was used to detect diabetic retinopathy,this system couldn't really distinguish between aneurysms and hemorrhages.

The second problem was that the system although automated could not learn.   Although many attempts at making use of machine learning in this system by multiple researchers were successful, they failed when medical practitioners provided them with real data. 


Now, this is where Calico Inc's lens makes a leapfrog improvement over this approach. 

It can learn.

What the heck does that mean? 

It means that if the Calico's lens' computer is exposed to a sufficiently large pool of data, it will gain the ability identify the actual ability to diagnose diabetic retinopathy accurately.

Turns out that Calico has already that.They have also published the results.

Figure 1. Examples of retinal fundus photographs that are taken to screen for DR. The image on the left is of a healthy retina (A), whereas the image on the right is a retina with referable diabetic retinopathy (B) due a number of hemorrhages (red spots) present.

This is the description of the approach adopted by Calico INC according to the Google Research Blog:

"Working closely with doctors both in India and the US, we created a development dataset of 128,000 images which were each evaluated by 3-7 ophthalmologists from a panel of 54 ophthalmologists. This dataset was used to train a deep neural network to detect referable diabetic retinopathy. We then tested the algorithm’s performance on two separate clinical validation sets totalling ~12,000 images, with the majority decision of a panel 7 or 8 U.S. board-certified ophthalmologists serving as the reference standard. The ophthalmologists selected for the validation sets were the ones that showed high consistency from the original group of 54 doctors."

Performance of both the algorithm and the ophthalmologists on a 9,963-image validation set are shown in Figure 2. 

This implies that this algorithm can perform as well as ophthalmologists and at times,even better.

This research was done by Google's research division. But Calico's job here is to make proper use of this research and develop a wearable device that acts as powerful tool against this disease. Calico has developed a prototype using the research done by the Google research team.


This is how the lens's prototype looks like.

With this kind of progressive synergy in healthcare and tech, there is an immense degree of optimism surrounding the future of the fight against disease. However, one thing remains unclear. 

How is Calico going to protect the sensitive data that is going to be handled by this device?

Stay tuned for more updates. I will be publishing a second part to this article covering further minutiae !


Pranesh Umashankar,
Editor