Source: HealthImaging (link)

Dicom Systems has recently de-identified 5,3 million radiology exams and other medical imaging data for Hospital for Special Surgery in New York. The effort demonstrates Dicom’s support towards artificial intelligence (AI) research and innovation in healthcare, according to a recent Dicom release.

The enterprise health IT company de-identified exams with the Dicom Systems Unifier’s platform and delivered them to the HSS Global Innovation Institute, which has partnered with an AI company to develop algorithms aimed to address the issue of fracture misdiagnosis, according to the release.

“We found the Dicom Systems Unifier’s de-identification capabilities to be the enterprise-class platform we needed to tackle the complexity of our requirements.” said David King, executive director for HSS Global Innovation Institute. “This data pool is already in the hands of our AI partner, with the aim to significantly improve diagnostic accuracy of fractures and pathologies in radiology.”

Features of the Dicom Systems Unifier platform include:

  • Capacity to implement complete de-identification framework from data preparation and migration to building a data lake.
  • Bi-directional dynamic tag morphing makes changes on input and output.
  • Advanced pixel-level de-identification while avoiding accidental corruption or truncation of the image file.
  • Complex DICOM tag substitutions, removals or morphing are automated by designing transformations into LUA script.
  • Full customization of de-identification processes and output.

“Consumption of high-quality data by deep learning applications is an essential contribution to better machine learning algorithms, unleashing tremendous potential for AI solutions that benefit patient care,” Dicom stated.

Dicom’s de-identification feature follows the company’s recent initiative to focus on the supply side of imaging data for research, clinical trials and deep learning, explained Florent Saint-Clair, executive vice president of Dicom Systems.

“Machine learning in medical imaging is a voracious process that requires massive data consumption,” Saint-Clair said. “We’re excited to serve as the on-ramp to AI for pioneering clinical and research organizations who are pushing the limits of healthcare innovation.”