by Lisa Chamoff
, Contributing Reporter | July 01, 2019
From the July 2019 issue of HealthCare Business News magazine
The offering, while common for other areas, such as MR imaging, is unique for mammography, Harvey said.
Hologic also has a new AI-powered solution for tomosynthesis currently pending FDA approval. The product is designed to decrease both read times and file size, while retaining high image quality, according to the company.
In December 2018, iCad announced FDA clearance of ProFound AI, a deep learning-based cancer detection software for digital breast tomosynthesis.
The software reads the tomosynthesis slices and provides a probability (certainty of finding) score based on deep learning hundreds of thousands of images, said Michael Klein, chairman and chief executive officer of iCad.
“It’s like having 100,000 pairs of eyes looking over your shoulder,” Klein said.
Based on an FDA reader study of 24 radiologists, the ProFound AI solution picked up an additional 8 percent of cancers than physicians alone, Klein said. There was also a 7 percent improvement in specificity, reducing the number of false positives and potentially unnecessary patient callbacks, and a 53 percent reduction in reading time. The most complex dense breast cases were read in 59 percent less time.
The company is currently increasing the number of images used for the deep learning process, with an eye toward greater sensitivity and fewer false positives.
iCad is also planning to release two updates to the software. One would update the patient’s probability and certainty of finding a score based on a comparison of their prior mammograms.
For example, if "a lesion or region of interest has a 60 percent score of algorithmically being cancerous," Klein said, the patient may or may not receive additional screening or a biopsy. "If, however, a physician looked at [the] prior year's images of the same slide and last year's image looked [like it] had a score of ... 28 percent certainty, and the year before that has only an 8 percent certainty score, the physician would have far more [information] and perhaps be able to make a more informed decision on further diagnostic procedures.”
The company is also partnering with researchers at The Karolinska Institutet in Stockholm, Sweden, to develop an AI-based solution that will identify a patient’s individual risk of developing breast cancer based on correlating personalized information.”
"When you add a woman's actual images and personal history and genetic profile, the ability to predict their breast cancer risk increases from 56 percent to a certainty of an interval cancer, for the next 12 to 24 months, of just shy of 80 percent," Klein said. "What this introduces is the movement to risk-adaptive screening, perhaps more frequent screening for high-risk and probability cases, and perhaps less frequent screening for women at more normal and lower risk. If we can find these cancers one stage earlier, it's obviously a win for the patients, it's a win for the hospitals and it's a huge win for managed care."