Technology can help big data think outside the box in radiology.
There is much excitement about big data and its potential applications in medical imaging. The purpose of this blog is to consider related approaches that are providing results today and can therefore serve as a guide for big data driven applications.
Virtual colonoscopy will be used as a clinical example to clarify this subject. Big data in the form of machine learning has been applied to CADx and utilized for virtual colonoscopy.
The CADx approach of analyzing large numbers of 2D images, creating representative algorithms, and applying them to diagnosis is an established approach. Medical Sieve, from IBM, is one of the latest versions of diagnostic interpretation of 2D medical imaging datasets guided by clinical knowledge.
“Medical Sieve is an ambitious, long-term exploratory grand challenge project to build a next generation cognitive assistant with advanced multimodal analytics, clinical knowledge, and reasoning capabilities that is qualified to assist in clinical decision making in both radiology and cardiology.”
In all of the above, the fundamental source of data remains in the form of 2D views. When the 2D views are created as a projection of 3D volume data sets, we might refer to them as 2.5D views. The use of 2D/2.5D as the application of big data in medical imaging, is dated and therefore can be considered as thinking “inside the box.”
According to Shrestha, “Deriving value out of insightful analytics that could impact care processes and outcomes is an important, and often the most critical, goal of efforts related to big data.”
By working with “objects,” available with interactive virtual reality (IVR), as opposed to being limited by images alone, the goals noted by Shrestha are already being realized in virtual colonoscopy as follows:
“With IVR, I can lift the colon out of the display, open up the colon and rotate it in innumerable planes,” Judy Yee, MD, of UCSF, said. “Image manipulation is more intuitive and I can use the spatial resolution of lesions relative to folds and the background wall more easily, providing an ideal situation to identify polyps including flat lesions.”. This approach can be thought of as thinking, “outside the box,” with intuition driven knowledge, resulting in increasing clinical efficacy and workflow/patient outcomes.
The ultimate goal of big data is to improve patient outcomes. With IVR, the initial results are encouraging. They provide a ground truth and guide for raising the bar further with big data techniques applied to objects. At that point, big data will think outside the box.