Medical Imaging AI uses deep learning to assist radiologists in recognizing findings. Zebra Medical Vision is aiming to develop algorithms that "find the Zebras" (why we are called Zebra?) Testing our algorithms performance is subject to many challenges, specially when tagging significant amount of data requires enormous efforts.
I will present two use cases, in two algorithms where we developed a distributed dockerized reproducible pipeline system, enabling running multiple instances of algorithms parallelly.
We used large scale inference to evaluate emulated T-score, that measures lumbar spine density from CT scans, and Pneumothorax identification, a rare acute finding from Xray scans. Results of analyses done on these runs helped us evaluate performance during development process.