Reliable and Efficient Medical Image Segmentation

Medical images have become a critical diagnostics tool for clinicians. More recently noninvasive 3D images such as MRIs and OCT have taken their place as important replacements for invasive tests. Unfortunately these images have mostly been used for qualitative analysis. Our image processing technology provides a means to alleviate and eventually eliminate this unfortunate circumstance by placing powerful quantitative tools in hands of clinicians. The fast and accurate segmentation of medical scans is the pivotal missing piece impeding progress toward quantitative medical imaging or computer aided diagnosis (CAD). Simply decomposing a scan into semantic components (e.g. heart, vascular structure, retinal layers) affords a vast suite of new medical instruments. For example, a patient's progress can be tracked over time by measuring the size of tumors which can be useful for clinical studies as well as pharmaceutical trials, or a diagnosis can be corroborated by the statistical agreement with vast collections of previous cases. Prior attempts to segment out the constituent retinal layers in OCT data has met with only moderate success even for the most basic segmentation. The main issue has been robustness especially with pathologic cases.

We have recently developed a segmentation algorithm, Spectral Rounding, with broad applications in the medical image-processing domain. By efficiently computing and exploiting global functions of medical scans, such scans can be decomposed into a collection of meaningful components in nearly linear time. Unlocking the latent potential of medical imaging by providing quantitative analysis tools that satisfy the time and reliability constraints of a clinical workflow.

Team: