Technology
The various analyses done by IGROK could be readily done by a well-trained human with a ruler and lots of hours to spend flipping back and forth through images, with one exception: the independent tracking of millions of individual anatomic locations as they move and evolve from where and how they were imaged at one time through where and how they were imaged at multiple other times. This process is known as "deformable registration," and has long been a goal for medical imaging analysis. Key clinical facts, central to radiation therapy, that are unavailable without this analysis include the fundamental questions:
How much radiation dose has the tumor received so far?
How much dose have the healthy tissues been subjected to?
Accurate deformable registration promises to answer these questions more completely than ever before, and provide the tools to make mid-course adjustments in treatments, avoiding problems and seizing opportunities that previously went unnoticed until too late.
Patient Models
Registration between time points today is commonly done with the assumption that a patient is a rigid object in substantially the same position from day to day, requiring only a shift, and perhaps some rotation to line up the anatomy. This is a simplification used because alternatives have not been practical. Its shortcomings are well-understood, especially in cancer care, where treatments are spread out over months and weight and tumor changes are common, and anything but rigid. One alternative approach, "free-form deformation" has tried to model the patient as a mass of Jell-O, rather than a single rigid mass. This kind of registration is relatively easy to calculate, but makes anatomically unrealistic distortions based on its assumption that bones move and change the same way soft tissues do.
Recognizing this problem, an alternative scheme has been proposed which attempts to create a mechanically accurate model of a given patient’s tissues, and manipulate this model to best match changing patient images. Creating such a detailed model for a given patient from available data, with accurate representations of even pathologic anatomy is problematic however, and its underlying assumptions of normalcy make its errors insidious and hard to detect.
iGrok’s proprietary approach combines the best aspects of both of these approaches, efficiently developing an anatomically-informed registration without the need for human intervention, and providing the tools to quickly validate the results.
