Now doctors use artificial intelligence for early cancer detection: Experts

Experts say tech still nascent, diversified dataset for correct risk profiling currently unavailable

The development of artificial intelligence (AI) models in oncology and radiology for early cancer detection is gradually gathering steam, with several Indian health care companies initiating research studies to integrate them into screening efforts.

Apollo Cancer Centre recently inaugurated India’s first AI-driven precision oncology centre in Bengaluru. This centre is working on providing precise and timely oncology care, leveraging AI for accurate diagnosis, insights, cancer risk assessment, and treatment protocols.

Similarly, Apollo Radiology International announced its tie-up with Google recently to develop AI models for the early detection of cancer.

While AI integration can help in cancer detection due to its ability to analyse medical images with enhanced precision, some experts believe that a permanent solution might still be far away.

AI can lead to improved study of scans

Speaking on the potential role of AI in oncology, Joydeep Ghosh, senior consultant medical oncologist at Apollo Cancer Centres, Kolkata, said that AI algorithms can scrutinise radiological scans, such as mammograms (for breast cancer) and computed tomography and positron emission tomography scan images (for lung cancer), detecting subtle abnormalities that may indicate the presence of cancer at its nascent stage.

“This early identification, in turn, will enable doctors to initiate timely interventions, significantly enhancing the chances of successful treatment and recovery," he added.

AI algorithms can analyse vast datasets, including genetic information and patient histories, to identify genetic markers and risk factors associated with specific cancers.

AI models for medical imaging and diagnostics are already being used for tuberculosis (TB) screening.

An example is Qure.ai, a Mumbai-based startup, which has been providing its AI algorithm-based device for TB screening to several health facilities since February 2020.

Commenting on the uptick in the adoption of AI models, Ghosh said that some centres are in the process of incorporating AI into mammography reporting for breast cancer, but it is too early to make it standard practice.

AI models still a trial-and-error-based method

Despite advances in AI algorithms, doctors acknowledge that it is still in its infancy as a diagnostic tool, as the development of successful AI models may still take a long time.

Commenting on the current state of AI models in oncology, Nitesh Rohatgi, senior director of medical oncology at Fortis Memorial Research Institute, said that one of the biggest issues with the integration of AI models with cancer detection would be data collection.

“AI algorithms require a diversified dataset for correct risk profiling of potential cancer patients, which currently is not available anywhere in the world," he said.

Terming AI as a trial-and-error method, Aparna Dhar, director of hereditary, precision oncology and genetic counselling at Max Super Specialty Hospital, Delhi, said that while AI capabilities can improve the accuracy and speed of cancer diagnosis, one must understand that there can always be a blind spot due to a lack of data.

“With improvements in data processing by algorithms, AI can automate procedures and streamline clinical workflows, reducing the risk of error," she added.

Rohatgi added that while AI currently may not add any value in oncology, it is very much possible that there will be an increase in AI usage in detection in the next five to seven years.

Tech opportunity

> AI algorithms can scrutinise radiological scans, such as mammograms (for breast cancer) and CT and PET scan images (for lung cancer)

> It can detect subtle abnormalities that may indicate the presence of cancer at a nascent stage

> AI can analyse vast datasets, including genetic information and patient histories 

> But doctors say the tech is still in infancy as a diagnostic tool and development of successful AI models may take a long time