India's breast cancer crisis: Can AI make mammograms more accurate?

AIIMS has transferred its AI-based mammography interpretation technology to BPL Technologies. The move could support earlier breast cancer detection by helping radiologists flag suspicious findings.

In Short

• Most breast cancer cases in India are detected only at advanced stages
• Uneven access to specialist imaging experts limits timely screening beyond metros
• AI tools can flag subtle abnormalities and ease radiologists' heavy reading workload

India has been seeing a sharp rise in breast cancer, but the bigger challenge may be detecting the disease early enough. More than 60% of breast cancer cases in India are estimated to be diagnosed at stages III or IV, when the disease has already advanced. At the same time, access to specialist breast-imaging expertise remains uneven, particularly outside major cities.

This makes accurate interpretation of mammograms a critical gap in the country's cancer-care system.

Now, an artificial intelligence (AI)-based mammography interpretation model developed at the All India Institute of Medical Sciences (AIIMS), New Delhi is moving a step closer to wider clinical use.

The premier health institute has transferred the technology to BPL Technologies, a Delhi-based medical device maker, opening the way for further development and validation of a tool designed to help radiologists identify suspicious findings on mammograms.

Dr Nikhil Tandon, director of the institute, flagged the importance of translating research generated within academic institutions into solutions that address real-world healthcare needs.

So, how exactly can AI help read a mammogram, what has AIIMS developed, and could this technology help India detect breast cancer earlier?

MAMMOTH CHALLENGES

Breast cancer is the most common cancer in India with over 2.3 lakh new cases diagnosed every year, the highest among all malignancies. A recent analysis published in The Lancet Oncology found that India's breast cancer incidence rate rose from 13 per 100,000 women in 1990 to 29.4 per 100,000 in 2023 – a 127% increase.

The age-standardised mortality rate also rose from 8.9 to 15.5 per 100,000 during the same period.

A mammogram is essentially an X-ray image of the breast. It can reveal tiny abnormalities, including suspicious masses or clusters of calcifications, before a woman develops a noticeable lump or other symptoms.

But reading a mammogram is not always straightforward. Some cancers can be extremely subtle. Dense breast tissue can also make abnormalities harder to see. India additionally faces a major shortage and uneven distribution of specialist imaging expertise, with access to advanced screening concentrated in larger urban centres.

Reviews of breast-cancer screening in India have highlighted limited access, shortage of skilled personnel and challenges associated with breast density as major barriers.

This is where AI is being explored not as a replacement for the radiologist, but as an additional pair of eyes.

An AI mammography system is trained using large numbers of breast images. It learns patterns associated with cancer and other abnormalities and can flag areas on a mammogram that deserve closer attention. Depending on the system, it can also help prioritise cases and support a radiologist's interpretation.

That could matter enormously in India. A radiologist in a busy hospital may have to read hundreds of images, while a smaller centre may not have a breast-imaging specialist at all. An AI system can rapidly analyse images and draw attention to suspicious areas, potentially reducing the chance that a subtle finding is overlooked.

But AI does not make the diagnosis on its own. A flagged mammogram still needs clinical assessment, additional imaging or biopsy where appropriate, and ultimately a pathologist's confirmation of cancer.

WHAT AIIMS HAS BUILT

This week, the institute transferred its AI-based mammography interpretation technology to BPL Technologies Limited.

The model was developed by teams at AIIMS and is designed to support the interpretation of mammography examinations, particularly in settings where specialist breast-imaging expertise may be limited.

Now, say the experts, the model needs to be tested on diverse datasets, including Indian patients and different imaging environments, before its performance can be judged in routine practice.

It also opens the possibility of integrating the technology into mammography systems or clinical workflows used by hospitals and diagnostic centres, subject to the necessary validation, regulatory requirements and clinical evidence.

The development comes amid growing evidence from countries with established mammography screening programmes that AI can improve the performance of breast-cancer screening.

One of the strongest pieces of evidence comes from Sweden's MASAI randomised trial, involving more than 105,000 women. In the trial, AI-supported mammography detected 6.4 cancers per 1,000 women screened, compared with 5.0 per 1,000 with standard screening. That represented about a 29% higher cancer-detection rate. The AI-supported approach also reduced the screen-reading workload by 44.2%.

The findings became more interesting in subsequent analysis. AI-supported screening had a sensitivity of 80.5%, compared with 73.8% with standard double reading, while specificity was virtually identical at 98.5% in both groups.

The rate of cancers appearing between screening rounds – known as interval cancers – was also non-inferior with AI support.

- Ends