‘Modernising healthcare with the help of AI, intertwining clinical practice and research critical’

Ashish Modi: We are an R&D unit of ZEISS India. We are called Center for Application Research in India. We are heavily focused in developing products in the area of ophthalmology. AI is poised to change the way traditional medicine is practiced. 

We started working with IISc in 2020 to look at some of the key eye care problems that could be solved with AI. After getting very interesting results, we felt that instead of doing a project-based collaboration, why not do something long term. Thus, we jointly agreed to set up a lab focused on AI for eye care.

Rajesh Sundaresan: A few years back, the principal investigator of this particular project, Professor Chandra Sekhar Seelamantula, worked on signal processing for optical coherence tomography (OCT). It is a method used to look at the shape of the retina, and internal structures using infrared optical signals. Now, many autonomous vehicles use this to do perform depth mapping.

Under a programme by the Government of India called IMPRINT, Prof Seelamantula combined this with artificial intelligence techniques. He took the ideas of optical coherence tomography and combined it with the advanced techniques in artificial intelligence to create data sets for glaucoma, and having experts annotate these data sets so that AI models could be trained on it. Since ZEISS was also interested in such technologies, the two have now come together on AI in for eyecare.

More about the lab?

Ashish Modi: We are setting up a high-end IT infrastructure lab for both Master’s and Ph.D students. We will be sponsoring these students. As of now, two masters students have enrolled. We are looking for two PhD students. Over the next three years, we are expecting 8 to 10 researchers to be working in the lab. We want to see how this flourishes and more problem statements come up and then we can build on top of that.

Are there specific areas of research that you are interested in?

Rajesh Sundaresan: This particular group is looking at what is called fundus imaging. What that involves, apart from glaucoma, is the potential early screening of another disease called diabetic retinopathy. All of these require imaging the retina.

Another possible application in the future is a potential window into understanding the blood vessel structure in the eye. This can be analyzed to understand some corresponding changes taking place in the brain. 

So, these are three things that I can think of – glaucoma, diabetic retinopathy, and the things related to early identification of cognitive impairment in those who are aging. 

These are the important applications for screening.  

What we would also like to see is large scale screening, and thereby equitable health. If handheld screening, and imaging devices can move to villages where experts are few in number. Currently, it could enable that. Where AI technology can be used to screen images in an automated way and reduce the load on healthcare professionals.

The next thing is data. AI tools need data which is high quality gold standard data. For that we need to work with hospitals, standardise and gather this data, annotate it and then train the AI. We are at the first level of creating the gold standard data.

Would the researchers also receive help in identifying markets and productizing? 

Ashish Modi: We would provide exposure to researchers in identifying problem statements and market possibilities so that they get some direction in terms of where, if they put their efforts, will be more RoI from a commercialization perspective.

How does all of this fit into the larger IISc research context?

Rajesh Sundaresan: IISc is starting a medical school and hospital. 

Clinical research is somewhat separated from practice in India. And the COVID-19 experience taught us the benefits of bringing these two together. Intertwining these two is important and the hospital at IISc is one such experiment. 

IISc is one of the institutions where biological research, technological research and bioengineering take place. So, with the medical school and the hospital, I think the circumstances are great for this. 

We have to think of healthcare, diagnostics, and delivery of healthcare in a modern way. AI can enable the acceleration of better healthcare today. There is a concerted effort to bring together AI in healthcare with the hospital and the medical school and the technology and science that is there in the Indian Institute of Science. 

We are building an ‘AI in healthcare’ collective which is a consortium of several people, hospitals and institutions. One of its activities is AI in eyecare. 

Also, the Indian Council of Medical Research and IISc came together to create a database. It’s called the Medical Imaging Data Sets Platform or MIDAS. 

The purpose of this is to work with hospitals, ICMR, and researchers, to understand what their needs are, and then codify that into a set of standards in terms of image gathering, annotations that are needed and so on. 

Given health data is very personal, how do you go about ethical concerns regarding these data?

Rajesh Sundaresan: Before we embark on data collection, we have to follow some rigorous standards related to ethical differences. We have to lay out a plan for the processes we follow, how the data will be used, what kind of sharing mechanisms will be employed and so on and then share it with the ethics committee. They have to clear the process by which the data is obtained and used. Only if the participant is willing to share it, we collect the data. We have to ensure that anonymity is protected even after they give consent. So, we make sure that all the images are de-identified. Once the de-identification has happened and there is sufficient guarantee that you can’t remap it to the individual, it is put in another database which is for research purposes. So, there are protocols we follow for this.

What are some of the challenges that you anticipate? 

Rajesh Sundaresan: Clean data is one. We are trying to address that through our gold-standard data collection. The other thing is the fast pace at which it is evolving and the resources that are needed to be able to keep pace with the fast transitions in the ecosystem. We would need a significant amount of infrastructure to be able to do the training. 

Resources in the research space is another challenge. There are a lot of individuals who want to go into applications of AI. But when it comes to healthcare, it is one of those domains where the technology that comes in today was developed 10 years back. So, we would need individual researchers to commit to this long timeframe.

Another important challenge is the deployment of AI. Let us say we develop a new OCT device, but an unskilled person in a remote part of the country is being asked to use it for imaging. Will they do it or will they prefer continuing the existing norms? Will AI be accepted is a question that also we need to understand and think about.

Ashish Modi: AI has the potential to make an impact in clinical aspects, but how does it weigh from an economic perspective? I think it’s going to be a big challenge as to how we can commercialize it and make it available for everyone. We have to figure out ways to make it financially viable and to commercialize it effectively.