Researchers from Siemens Healthcare in Malvern, PA, thought it might, and in this Thursday scientific session presentation they will discuss their development of an image recognition algorithm that can automatically detect anatomical content and orientation of radiographs and launch a CAD application, if necessary.
The system is integrated into the real-time workflow of a PACS workstation, and it first detects multiple focal anatomical information within the medical image, according to the research team led by Xiang "Sean" Zhou, Ph.D., of Siemens. Next, an algorithm eliminates inconsistent findings. In the last step, a reasoning module assesses the evidence to determine the content and orientation of the image, according to the researchers.
Siemens trained and tested the system on recognizing general radiographs as posteroanterior/anteroposterior (PA-AP), lateral, or other (x-rays of the head, pelvis, etc.), as well as differentiating chest radiographs as PA-AP or lateral, according to the researchers.
The system achieved a 99.17% accuracy rate in recognizing 11,000 test images (3,708 PA-APs, 1,167 lateral studies; the balance were other studies). In the PA-AP and lateral separation task, the system had a 99.98% accuracy rate.
