An Israeli startup says it has taught an algorithm how to detect breast cancer
Founded in 2014 and backed by the likes of Salesforce billionaire Marc Benioff with $20 million (£16 million), the Tel Aviv-based company says it has taught an algorithm to identify early signs of breast cancer with the help of thousands of previous mammograms.
That constantly improving algorithm - trained using a technique known as machine learning, which is a type of AI that equips computers with the ability to learn without being explicitly programmed - is now better than radiologists using the best Computer Aided Detection (CAD) methods for mammography, the company claims.
Eldad Elnekave, Zebra's chief medical officer, told Business Insider in Tel Aviv that the algorithm can detect half of the breast cancer cases that are currently being missed by radiologists. Radiologists working for the NHS fail to spot breast cancer in thousands of mammograms every year, according to The Telegraph. The condition affects one in eight women in their lifetime, according to UK charity Breast Cancer Care.
"The challenge here is there's so much background noise in breasts," said Elnekave. "Breasts can be dense, they can be not dense, they can be have implants, and so on. There are so many possibilities so to find the actual breast cancer is a challenge. This is where we could utilise the fact that we had an enormous database of mammograms. We have 344,000 breast cancer studies [from hospitals]."
The mammography algorithm will be added to the company's growing list of clinical algorithms, which can automatically read and diagnose medical imaging data. Current algorithms are in the fields of bone health, cardiovascular analysis, liver and lung indications, and now mammography.
Dr Maya Cohen, director of the imaging Institute at Rabin Medical Center and director of the Breast Health Center at Herzeliya Medical Center in Israel, endorsed Zebra's technology in a press release published by the company.
"Some of the most challenging cancer diagnoses are ones where the visual cues are not distinct lesions but rather regional asymmetry or architectural distortion in the breast tissue," she said in a statement, adding that Zebra's algorithm could help mammographers detect even the "most subtle" cancers.