CHICAGO — A study presented last Wednesday at the RSNA meeting found that ScreenPoint’s Transpara deep learning computer detection system can perform as well as experienced radiologists.

“The decision support is where we come in — [the radiologist] knows where the abnormality is, but they want to know if it’s abnormal enough to recall the patient,” Jan-Jurre Mordang, technical sales and support specialist and medical engineer at ScreenPoint, told HCB News.

The researchers at Radbound University Medical Centre in Nijmegen, Netherlands compiled reader study data from multiple breast imaging centers in Europe. In four different studies, 24 radiologists reviewed over 1,400 mammograms from three vendors to measure their ability to detect breast cancer.

They found that there was no significant difference between Transpara’s automated reading and the radiologists’ reading. In two of the studies, the radiologists had a higher performance, and Transpara had higher performance in the other two.

Transpara, which was launched in February, is a computer-aided detection system, but with added decision support. It automatically identifies soft-tissue and calcification lesions and combines the findings of all available views to generate a cancer suspiciousness score on a scale from 0 to 100.

The system marks calcifications just like conventional CAD, but only a small number of soft-tissue lesion marks are shown and are proved to have extremely low false positive rates. The radiology can also leverage the decision support to probe any suspicious image region to determine if further investigation is required.

“It’s more about specificity — you want to have as few recalls as you can have,” said Mordang. “Considering that more than 95 percent of the screening population is normal, it’s better to eliminate those so the radiologist can look at the other five percent more closely.”

Rad Oncology Homepage