Critical Care X-ray

Unmet need

GE Healthcare X-ray has developed a pneumothorax AI detection algorithm for frontal chest x-rays (AP & PA), which will be first deployed onto the Optima XR240amx portable x-ray system as a software update, and later deployed onto GE fixed x-ray systems.

This is planned to be the first in a suite of Critical Care AI algorithms focused on chest x-rays – Future algorithm development is currently planned to include, but is not limited to, detection algorithms for Pneumoperitoneum, Endotracheal Tubes, Nasogastric Tubes, and Chest Tubes. In addition, this may be the first imaging device to deploy a clinical detection AI algorithm within it.

Currently, the existing product features have been developed:

• Three quality control related AI algorithms for frontal chest exams:
  • An on-device auto upright rotation algorithm for frontal chest x-rays
  • An on-device alert when the protocol used does not match the image acquired (ie when an AP Abdomen protocol is accidentally used to image an AP Chest.
  • An on-device alert when lung fields are clipped out of the field of view
  • An on-device alert indicating the likelihood of the presence of a pneumothorax
• A colour overlay on the image that localizes where on the chest x-ray image, AI found suspicion for PTX
• A secondary DCM image, in the case of a positive PTX finding, to be sent to PACs, describing the PTX AI finding (i.e. PTX AI had a positive finding with an AI score of 90.1%, with the colour overlay)
• Private DCM field with all the AI results

Project aim

The purpose of this collaboration is to further the development of the GE X-ray Critical Care Suite. In order to develop state-of-the-art AI algorithms, large amounts of quality datasets are required for algorithm training, testing, and validation. The robustness and completeness of these datasets directly impact the performance of the algorithm, meaning additional metadata is required in addition to the X-ray image itself.
Additionally, the GE X-ray Critical Care suite is developed for deployment directly on X-ray imaging systems at the point-of-care that provide clinical decision support to the clinical care team. For this reason, direct feedback on the performance, workflow and usability from those who interact with the Ai output is critical.

Objectives

1. Retrospective data collection of de-identified datasets and accompanying Radiologist reports
2. Image Curation/Annotation