GE’s AI upgrade sharpens 3D mammogram images

GE HealthCare has secured US FDA authorisation for Pristina Recon DL, an AI tool that sharpens 3D mammogram images to support breast cancer detection.

Breast cancer is one of the most common cancers in women, with around one in eight expected to receive a diagnosis in their lifetime and more than a million deaths projected each year by 2050.

The newly authorised Pristina Recon DL sits on GE HealthCare’s Pristina Via system and is designed to improve image quality in digital breast tomosynthesis, a type of 3D mammography created from multiple X-ray images of the breast.

The technology uses deep learning, a form of artificial intelligence that learns patterns from large datasets, together with an approach known as iterative reconstruction, which repeatedly refines images to reduce noise and improve clarity.

According to GE HealthCare, Pristina Recon DL uses two deep learning models in sequence. One focuses on separating the useful signal in the image from background noise, while the second is trained to highlight clinically important details in a synthesised 2D view.

The company says it is the first mammography technology to combine deep learning with iterative reconstruction in this way, aiming to provide high-quality 3D images without increasing the radiation dose to patients.

Pristina Recon DL was born out of a deep commitment to our customers, listening closely to their feedback and working hand-in-hand with radiologists to enhance image quality and clarity,” said Jyoti Gupta, president and CEO, women’s health and X-ray at GE HealthCare. By applying advanced deep learning technologies, we’re shaping the future of breast imaging, one defined by uncompromised image quality, faster workflows and greater confidence in early cancer detection.”

In a recent study cited by the company, breast radiologists reportedly preferred the overall image quality of Pristina Recon DL in 99.1 per cent of image reviews when compared with a previous reconstruction method.

GE HealthCare also reports better performance for detecting microcalcifications, tiny deposits of calcium that can be an early sign of breast cancer, and breast masses in trials using modelled clinical data.

“Our collaboration with GE HealthCare has been instrumental in advancing breast imaging capabilities, and the new 3D image quality represents a meaningful upgrade that will benefit radiologists and patients alike,” said Dr Howard Berger, president and chief executive officer of RadNet. “This pioneering AI technology will help elevate breast care by delivering the clarity and consistency radiologists need to enable more confident diagnoses.”

The Pristina Via system with Recon DL is also marketed as offering workflow efficiencies, including automated image acquisition and personalised exam protocols designed to speed up appointments and reduce waiting times.

GE HealthCare highlights other features such as patient-assisted compression, which allows women to help adjust the pressure on the breast during imaging, with the aim of improving comfort and reducing anxiety.

Additional applications include a shortened biopsy workflow and contrast enhanced mammography, where a dye is injected to help highlight abnormal blood vessels. The company says diagnostic accuracy with its SenoBright HD contrast enhanced mammography is comparable to breast MRI in multiple studies.

“With Pristina Via with Recon DL, we’re setting a new benchmark in breast imaging, delivering sharper, clearer and more consistent images that empower radiologists with more confidence,” said Pooja Pathak, vice president and general manager, mammography at GE HealthCare. “As an upgradable feature on the Pristina Via platform, we are excited to now offer customers uncompromised image quality combined with fast, accurate workflows.”

GE HealthCare said it worked with academic centres and high-volume outpatient imaging sites to develop and validate the algorithms behind Pristina Recon DL, aiming to make it harder for early cancers to be missed on screening images.