• ScreenPoint Medical | North - Hall B: 8046

Study presented at RSNA examined performance of artificial intelligence software that combines findings of all available views into single cancer suspiciousness score

January 12, 2018 — Deep learning and artificial intelligence improves the efficiency and accuracy of reading mammograms, according to research presented at the 103rd Annual Radiological Society of North America (RSNA) meeting, Nov. 26-Dec. 1, 2017 in Chicago. Three studies demonstrated the performance of Transpara deep learning system developed by ScreenPoint Medical BV is approaching that of experienced breast radiologists.

Utilizing state-of-the-art image analysis and deep learning technology, Transpara automatically identifies soft-tissue and calcification lesions and combines the findings of all available views into a single cancer suspiciousness score. While calcifications are marked as in traditional computer-aided detection (CAD) systems, only a small number of soft-tissue lesion marks are shown and are proven to have extremely low false positive rates. However, readers can probe any suspicious image region for decision support to help determine whether further investigation is needed.

The study “Detecting Breast Cancer in Mammography: How Close Are Computers to Radiologists?,” was presented by Alejandro Rodriguez-Ruiz . In the study, researchers from Radboud University Medical Centre in Nijmegen, Netherlands, compared the performance of experienced radiologists to that of the deep learning computer detection system Transpara in detecting breast cancer on mammograms.

Researchers collected reader study data from multiple breast imaging centers across Europe to assess performance. In four different studies, more than 1,400 mammograms from three different vendors were retrospectively reviewed by groups of radiologists to measure their ability to detect breast cancer. The data included 336 exams with cancer, 430 with benign abnormalities and 669 normal mammograms. In total, 24 radiologists participated in these studies. Results showed no significant difference between automated reading with the Transpara software and reading by the radiologists. In two studies the radiologists had a higher appropriate use criteria (AUC) performance, while Transpara had a higher AUC in the two other studies.

In the session, “Development of Deep Learning Systems for Improving Breast Cancer Screening,” Prof. Nico Karssemeijer, Ph.D., CEO of ScreenPoint Medical, presented on how recent developments in machine learning offer unprecedented opportunities for researchers to develop fully automated systems for the reading of mammograms and breast tomosynthesis.

“The scope of these systems will be much wider than that of existing CAD systems for mammography. They will provide decision support to improve recall decisions and pre-screening of exams by computers will become a reality. This will lead to more efficient screening procedures where human readers rely on automation to select normal exams that they don't need to read. This will allow them to focus on making optimal decisions for women with potentially abnormal exams in which cancer is most likely,” said Karssemeijer.

The scientific exhibit, “Automated Pre-Selection of Mammograms without Abnormalities Using Deep learning,” was presented by Jonas Teuwen, MSc, Ph.D., in poster discussions.

For more information: www.screenpoint-medical.com