How deep tech is revolutionising mammogram screening

 

PUBLISHED ON 

SCROLL DOWN

Despite the importance of regular breast cancer screening, mammograms are not perfect. They can sometimes produce false negatives, missing cancers that are actually present, and false positives, which flag harmless findings and cause unnecessary tests and anxiety.

The risk is greater for Asian women, many of whom have dense breast tissue that looks white on a mammogram, just like tumours. There may be a three-week wait for results as radiologists need longer to interpret the images. If only technology could overcome these limitations.

In 2023, doctors at the National University Hospital (NUH) began trialling the use of FxMammo, an artificial intelligence (AI)-powered software designed to help them detect breast cancer on mammogram images faster and more accurately.

FxMammo is the flagship product of FathomX, a medtech company spun off from the National University of Singapore (NUS) and the National University Health System (NUHS).

A/Prof Mikael Hartman, Head and Senior Consultant, Division of General Surgery (Breast Surgery), Department of Surgery, NUH, developed FxMammo in collaboration with A/Prof Feng Mengling from the NUS Saw Swee Hock School of Public Health (SSHSPH). Together, they co-founded FathomX to commercialise this innovative solution.

“Unlike traditional methods, today’s mammogram images are digital, which opens up new possibilities for more efficient analysis,” said A/Prof Hartman. He also holds professorships at the NUS Yong Loo Lin School of Medicine (NUS Medicine) and SSHSPH.

“With advanced AI tools and some powerful number-crunching, coupled with a sea of data for training, we developed FxMammo with the hope that this technology would give us an edge in beating breast cancer.”

“Breast cancer is a silent epidemic. In the last 35 years, we’ve seen a doubling by generations. It means a mother’s risk is twice that of her mother’s, and the daughter’s risk is twice that of her mother’s.”

– A/Prof Mikael Hartman

FxMammo is an AI assistant capable of advanced mammographic image analysis. It can spot subtle variations and anomalies in dense breast tissue that are often overlooked by the human eye.

Once suspicious areas are identified, the software generates a detailed report with a heat map and cancer probability score. This map highlights abnormal lesions for radiologists to perform further review, enabling accurate and reliable diagnoses.

Most mammography screenings utilise double reading, where two radiologists review the same image independently to reduce errors and improve accuracy.

Early results suggest that FxMammo is as good as two radiologists in detecting breast cancer, if not better. FxMammo can also serve as a “triage tool”, said A/Prof Feng. The bulk of cases would be easily resolved within a couple of minutes, freeing up radiologists to focus on more complex cases.

“It usually takes two radiologists more than 16 minutes to read a mammogram, but FxMammo can help cut this time by at least 20 per cent."

– A/Prof Feng Mengling

Crucially, these time savings do not come at the expense of reliability.

“FxMammo is as accurate as a trained radiologist,” said A/Prof Hartman. “Using a metric called ‘area under the curve’, where 1.0 is perfect, a radiologist scores about 0.9, while FxMammo’s score is between 0.91 and 0.93.”

Using FxMammo as a second reader (or, in discordant cases, a third one) can eliminate much of the labour-intensive, time-consuming, repetitive work associated with double reading. This in turn creates opportunities to streamline detection processes and clinical workflows, substantially addressing Singapore’s healthcare manpower challenges.

The Singapore Medical Council reported that the number of registered radiologists grew by 20.5 per cent from 370 in 2018 to 446 in 2022. But that still isn’t enough, given our rapidly ageing population and the progressive expansion of health services. An effective and efficient AI tool would not only help hospitals overcome staff shortages, but also reduce the workload for radiologists, increasing their job satisfaction.

Other positive findings of FxMammo:

  • It has been proven to lower false positives by 20 per cent, and false negatives by more than 38 per cent.
  • It could potentially save about S$2.7 million each year, due to the time savings achieved in diagnosing non-malignant cases.

All this leads to enhanced patient outcomes, including quicker turnaround for results, reduced anxiety, and (if a positive diagnosis is made) early treatment and better prognosis.

Hopefully, this will build greater trust in mammography screening and improve the take-up rate among women. The National Population Health Survey 2023 revealed that only 34.7 per cent of Singaporean women aged 50 to 69 had gone for a mammogram in the past two years.

FxMammo is now being trialled in more than 30 hospitals and universities across Asia. What’s more, the team at FathomX has plans to expand further into Europe and North America.

“With half of the world’s population being women who need mammograms, plus a shortage of radiologists globally, the need for such a technology is huge,” said A/Prof Hartman. “I look forward to seeing FxMammo prove its worth in NUH’s clinical trials and eventually aid in the global fight against breast cancer.”

In consultation with A/Prof Mikael HartmanHead and Senior Consultant, Division of General Surgery (Breast Surgery), Department of Surgery, NUH; Associate Professor, Department of Surgery, NUS Medicine; Associate Professor, SSHSPH; Senior Consultant, Division of General Surgery (Trauma), Department of Surgery, NUH; Senior Consultant, Division of General Surgery (Breast Surgery), Department of Surgery, Alexandra Hospital; and Senior Consultant, Division of Surgical Oncology, National University Cancer Institute, Singapore; and A/Prof Feng Mengling, Associate Professor, SSHSPH.