Tuesday, December 2 | 3:20 p.m.-3:30 p.m. | T7-SSIN04-3 | Room E450B
This imaging informatics reporting presentation discusses an integrated AI-PACS system designed to minimize workflow disruptions and reduce time to diagnosis for chest x-rays and CT exams.
Suvrankar Datta, MD, will share details around chest AI model management that allows radiologists to visualize AI-generated overlays and reports directly within their existing PACS interface. This type of AI-PACS integration may lower the barrier to widespread AI adoption, they said.
The integration architecture includes an AI orchestration engine for model management, a DICOM routing system for image transfer, and a lightweight PACS plugin, according to Datta and colleagues, who used it over two months on a total of 1,000 chest imaging cases (700 x-rays and 300 CTs).
Upon receipt of a new imaging study, the system automatically identifies the appropriate AI model and initiates inference, they explained. The output includes structured DICOM overlays highlighting findings and AI-generated reports, via large language models (LLMs), integrated into the study folder within PACS.
The group evaluated system performance using average inference time, integration success rate, and preliminary radiologist feedback on usability and workflow compatibility.
Among the results they highlighted prior to RSNA, average end-to-end processing time, including AI inference and delivery to PACS, was 1 ± 0.15 minutes per chest x-ray and 2 ± 0.3 minutes per chest CT. They also said preliminary feedback from radiologists indicated improved triage of critical findings such as pneumothorax and tuberculosis.
Earlier identification of critical findings may be possible. What about system downtime, data loss, and patient management? Attend this Tuesday talk to get the answers to your questions.
