Bringing AI to AIIMS to help identify cancer therapies

iOncology AI aims to sequence the genomes of 3,000 cancer patients who are currently seeking treatment at AIIMS, aiming to address a fundamental question: Does a patient’s genetic makeup correlate with the efficacy of diverse cancer therapies?

So far, the AI model has been tested on the pathology, radiology and clinical history of 1,500 breast and ovarian cancer patients from AIIMS and was found to be over 75 per cent accurate when compared to the diagnosis provided by the clinicians.

Genomic data is a powerful tool for medical researchers and doctors. It helps them understand how variations in DNA affect our health. Through genomic sequencing, they decipher our genetic makeup and spot alterations in our genes. These changes are key to understanding how diseases such as cancer develop.

Globally, there is an increasing interest in the use of AI for cancer treatment as it can help doctors – from developing new therapies to diagnosing patients at early stages of the disease and selecting appropriate treatments. (Graphic design by Abhishek Mitra)

“This is an evolving platform, the more data it gets, the more accurate it will become,” said Dr Ashok Sharma, the principal investigator of the project and additional professor specialising in epigenetics in the department of biochemistry at AIIMS. The system is now being validated in five district-level hospitals in Madhya Pradesh, Uttar Pradesh, and Rajasthan.

How does the AI tool work?

The platform, developed with Pune’s Centre for Development of Advanced Computing, allows doctors to store cancer patients’ blood test and lab reports, scans and ultrasound reports and their history. Along with collecting and collating data, the AI-enabled platform is also capable of reading the different types of datasets and helping the doctors make decisions.

“After studying the clinical data and genomic make-up of, say, a couple of thousand cancer patients, the tool will be able to help the doctors in selecting the appropriate treatment for the next patient,” said Dr G K Rath, former head of the National Cancer Institute, Jhajjar, and one of the key coordinators of the project. This tool, he explains, can help in targeted treatments in resource-constrained settings as well where genomic studies might not be possible.

Genomic data is a powerful tool for medical researchers and doctors. (Graphic design by Abhishek Mitra)

“While the system cannot replace a doctor, it can guide them. When a scan or histopathology report is uploaded on the platform, the trained AI will be able to automatically flag the ones that have certain abnormalities. Or, take for example, an extremely small tumour that a radiologist might miss at first. The AI will flag it. This can help in early detection of cancers in some cases,” said Dr Sharma.

This data can help oncologists create their own prevention strategies and standard protocols. “For example, haemoglobin levels of people in India are less as compared to the Caucasian population. Our treatment protocols can be adjusted to account for this,” said Dr Rath.

Can the tool address all types of cancer?

The research will focus on five types of cancers that are the commonest, deadliest or have immunotherapies available — breast cancer, ovarian cancer, head and neck cancer, colorectal cancer and two types of blood cancers. At present, the platform is trained only for breast and ovarian cancer. It already has various models on CT scan prediction, ultrasound image, histopath prediction, mammogram image prediction and detection and classification of the tumour.

While this work is still in the future, the team has already laid the groundwork. (Graphic design by Abhishek Mitra)

“These parameters were decided upon after extensive discussions with clinicians to figure out what they really look at to decide whether a patient has cancer, what stage the cancer is at, and what should be the treatment protocol,” said Dr Sharma. As for patient confidentiality, “There are separate logins for everyone, so a radiologist will be able to see only the scans that have been uploaded by them along with an anonymised analysis of other scans. A clinician will be able to see the clinical history, scans, as well as lab reports of their own patient. This is essential for maintaining patient confidentiality while benefiting the doctors and researchers.”

AI and cancer care: The big shift

Globally, there is an increasing interest in the use of AI for cancer treatment as it can help doctors – from developing new therapies to diagnosing patients at early stages of the disease and selecting appropriate treatments. Studies have shown that certain AI-based models were able to identify individuals at high risk of developing pancreatic cancers up to three years earlier. This is game-changing considering that most get diagnosed only when the cancer has advanced or metastasised.

Similar results have been observed for breast and lung cancers as well. Take, for example, the AI tool being developed by the Harvard Medical School, which is similar to AIIMS, but is specific for colon cancers. The model is being trained to accurately predict survival, provide insights for treatment response based on tumour pathology, radiology images and genomic data.

“It’s not just the genes that you inherit but also the epigenetic switches that turn them on and off that can influence the outcome of a therapy. The same therapy might work wonders for some and not others. We want to see what are the genetic factors that determine whether a treatment works or not,” said Dr Sharma.“We already know of certain genes and their impact on cancer from global research, our work will focus on finding these genetic markers for the Indian population,” he added.

While this work is still in the future, the team has already laid the groundwork. While the prototype iOncology AI has brought the much-needed technology to the clinicians, Dr Sharma remains optimistic that his project on co-relating genetic makeup and cancer treatment will take AIIMS to the forefront of cutting edge research.