Dr Karuna Ramesh, views from a seasoned Pathologist

MBBS, DCP, MD, Ph.D, DML&E (Medical Law and Ethics) FICP

Years of experience:
Teaching 37 years
Service 37 years
Research 30 years
Administrative 30 years

Career Journey:

  • Research & Ethics committee chairperson – St. John’s Medical college ( Oct.2018 to date)
  • Pathologist at Rainbow children’s hospital – DNBE program (February 2019 to date)

Fields of Specialisation:

  • Haematology Anaemia, Thalassemia and Malignancies including myelodysplastic syndromes, Geriatric haematology, 
  • Blood bank, Medical ethics, Gynaec pathology, Medical education


As a head of pathology lab in your experience, what are your thoughts on SigTuple’s technology, Shonit, as a potential product in your lab?

“The product is good! Not only have I seen the demo now, but I’ve seen the product once when they were using at Humain at CARE diagnostics. The team was very comfortable handling the workflow. I saw  the performance of Shonit and it is quite good. One thing I felt is that this AI solution is totally dependent on morphology. But as a pathologist, I also interpret data with some clinical inputs. Hence apart from the AI diagnosis, we would also need to correlate with clinical inputs before diagnostic reports can be issued.”

As a product positioned to assist a pathologist, how confident are you with Shonit?

I have seen and done validations for Shonit in the past as well along with 50 slides we’re currently doing. So I’ve seen around 200 slides so far in various instances. I’m 85% comfortable and confident to allow Shonit to predict the patient results with accuracy, similar to auto approval procedure.

A small green rectangle to divide sections of the document

What would be the instances when you’d like to review an AI result such as from Shonit PBS?

As mentioned, the  15% in my 85% confidence can be built when I see more validation done using pediatric samples. Pediatric population’s normal range is totally different and certain aspects of the morphology is slightly different like nRBCs are part of the normal population, hence the apprehension. Also, the solution is still technician dependent on the smearing procedure. You will always find poor slides where bare nuclei or degenerated cells are seen. In such cases, I use my judgement and count upto 200 cells instead of the regular 100 to ensure accuracy in reporting, But with the case of Shonit, there is no such  judgement involved yet.

Since you’re a certified NABL auditor and head the ethics committee, how do you view AI based solutions like Shonit in terms of ethics and auditing? 

These are two different issues  and we’ll deal with it one by one. First auditing, they accept AI, without any issues. They accept and also look at the limitations when dealing with patient samples. For example, they will check if there is some kind of QC available. Next, if the report or values are only AI based, or if the QC is not available, then they would check the validation done for AI values against a reference and such reports would be checked once a month or bi weekly. In many places and abroad, though I have not seen personally, AI is already implemented and many reports are auto approved. 

Now from the ethics point of view, since this is a patient sample, privacy and confidentiality is key. In analysers, sensitive information is protected. But in AI, such information is a grey zone. Hence there is no easy answer to ethics when it comes to AI. There is a paper on this called “Ethics and AI” where such suggestions have been brought out and a panel of ethics committee members were asked to comment their views on this. 

*It was also clarified that Shonit PBS doesn’t have any such ethics issues since it operates every similar to a reference analyser. 

A small green rectangle to divide sections of the document

How do you think confidence can be built for an AI based product amongst pathologists and technicians?

Technicians are very comfortable with such products which automate their processes. Anything which is manual usually has a lot of errors in terms of transcription etc. Good technicians are used to such changes in the lab, for example Hemat analysers have undergone many changes over the years and people have always adopted better technology. Image based analysers like Shonit and flow cytometry based hemat analysers will inherently have some differences between their values since the methodology is different, but other than that, I don’t think adopting AI based solutions will have any problems. 

Technicians always understand limitations, for example we faced such issues when moving from manual DC and 3 part analyser (first versions of sysmex etc) and then eventually 5 part analysers came about. So until that came, 3 part analysers were used in conjunction with manual DC. So limitations are always understood. 

For Shonit, when 7 part readouts are given, it will have a good advantage over a 5 part analyser, when it performs and picks up IGs, Atypical cells and monocytes correctly. This has to be validated carefully. But the advantage of a 7 part analyser is in huge labs like St John’s which process 2000 samples a day. There patients with malignancies also come, hence the DC on these cells are very important. An accurate 7 part analyser will prove very valuable.

In terms of costs incurred in a lab, how do you see Shonit with AI 100 reduce costs? 

This can be looked at in two ways, from a personnel point of view, there is a change since the smearing and staining can be done by just two technicians in shift. But when you look at the cost of running an analyser, the charges are always calculated per cycle. The reagents and consumables need to be bought and are very expensive. Hence analysers are always used in rental models instead of buying them. So in that sense, Shonit PBS will prove to be cost effective since it doesn’t involve extra consumables, each for detecting WBC, RBC and platelet types. Here the analyser is the only investment. Hemoglobin I understand is still in R&D but it will be critical to have in the future, since a PS report alone doesn’t complete the package. For many patients, anemia detection becomes key in diagnosis.

A small green rectangle to divide sections of the document

What feedback would you give on Shonit when thinking about its implementation pan India? 

I think the major hurdle would be in handling belts with specific health problems. In the north eastern states, detecting Malarial parasites and quantifying hemoglobin is key. The regular analysers are able to flag malarial parasites in the form of a peak and such samples are then manually reviewed under a microscope. So Shonit’s ability to flag malarial parasites especially in malarial belts across the coastal region would be beneficial. 

A small green rectangle to divide sections of the document

What are your thoughts on the tele reporting feature which AI 100 with Shonit enables, especially during a pandemic scenario? 

Shonit that way is  a very well positioned product. Nowadays, only technicians go to the lab. Transfer of patient data happens through the backend channels connected to a doctors personal laptop. So routine reports can be approved without much of a problem but samples which need manual review, or which are dengue positive, (especially during Bangalore’s dengue season) become problematic. Such special cases need Tele review, because analyser results and manual DC don’t match well. So right now they just take pictures of the microscopic images and send via WhatsApp. But with Shonit there is a clear advantage in how this can be done hassle free.

A small green rectangle to divide sections of the document

While validating products like Shonit, what should validations extensively cover? 

As for regulatory and performance perspectives, the AI based solution should always be validated against a known standard. The frequency of these validations should be defined and the metrics should be established. Traceability and troubleshooting of data should be easy and clean. 

In terms of statistics, there is no specific requirement, correlation and RSQ is a very common way of clinical results evaluation, but it should be defined very clearly at the beginning of the validation. Like for example, CV for hemoglobin tests cannot be more than 0.5%, so such products have very low margins but for platelets a wider CV is accepted. 

A small green rectangle to divide sections of the document