He devised an algorithm that took image segmentation (grouping pixels at different levels of detail) to a whole new level; he not only found regional objects, but also grouped spatially separate objects into region classes. In other words, applied to a satellite image, it could not only identify and separate lakes of varying depths, but could recognize lakes as a class of objects spatially distinguishable from, say, trees.
He calls this Recursive Hierarchical Segmentation, and it has been used to analyze Earth-imaging data from NASA's Landsat and
"My original concept was geared to Earth science," says Tilton, who was at first skeptical that his algorithm could enhance, say, mammography. "I never thought it would be used for medical imaging."
Then he processed cell images and saw details not visible in unprocessed displays of those images. "The cell features stood out real clearly, and this made me realize that Bartron was onto to something."
Bartron Medical Imaging, based out of Connecticut, has since developed the new MED-SEG system, which the FDA recently cleared for use by trained professionals to process images alongside other images, though stipulated that the system should not (at least yet) be used for primary image diagnosis.
Bartron, which first studied the software through Goddard's Innovative Partnerships Program Office, licensed the patented technology in 2003 to create a system that would differentiate hard-to-see medical image details. It then began to work with doctors to analyze CT scans, MRIs, ultrasounds, etc.
"Trained professionals can use the MED-SEG system to separate two-dimensional images into digitally related sections or regions that, after colorization, can be individually labeled by the user," says Bartron President and CEO Fitz Walker.
Next up: clinical trials to put MED-SEG to the test as a standalone imaging system. Dr. Molly Brewer, a professor with the Division of Gynecologic Oncology at the University of Connecticut Health Center, hopes to improve mammograms in particular:
Women who either have high breast density or a strong family history of breast cancer are often sent for MRIs, which are costly, very uncomfortable and have a high false positive rate resulting in many unnecessary biopsies. Neither imaging modality can detect cancers without a significant number of inaccuracies either missing cancer or overcalling cancer. In addition, reading these tests relies on detecting differences in density, which is highly subjective. The MED-SEG processes the image allowing a doctor to see a lot more detail in a more quantitative way. This new software could save patients a lot of money by reducing the number of costly and unnecessary tests.
Beyond medical applications, the software could be used for facial recognition, crop management, and more.