Anyone who has ever had an MRI or CT scan knows how difficult it is to lie perfectly still during the procedure. Up the anty a bit and consider if the image is being used to perform surgery or administer radiation. A Ph.D. student at the Eindhoven University of Technology in the Netherlands is using machine learning software to address this issue.
When a cancer patient needs a radiation treatment or a doctor is about to excise a tumor, digital imaging is used in these types of situations to locate as precise a treatment area as possible. The problem is that simply breathing creates enough movement that the position is never exactly the same in the scans. Organs move as the patient breathes and even blood moving through the vessels changes the shape of the target area. Koen Eppenhof is developing a way to improve medical image analysis using AI to account for this during treatments. According to Eppenhof, current imaging software used during operations and radiation treatment takes too long to process the different images, and that is where machine learning can be useful. He is using the power of AI to create accurate images pixel by pixel.
The team used a set up developed for gaming, which is designed to handle large amounts of graphics, to create what they call a deep neural network processor that teaches itself to identify areas inside the human body by comparing thousands of images. One problem they ran into was a lack of examples for the program to study. There simply aren’t enough scanned images of some parts of the body to feed into the system. But in the areas where there was sufficient data, machine learning proved to be successful. It can pinpoint tumors very accurately and has been used experimentally for radiation treatments for prostate cancer patients.
Eppenhof was careful to point out that even if computers can be trained to perform radiation or surgery successfully, this kind of technology will probably never be used without human supervision. It does have promise for future medical treatment, however, and shows that machine learning is useful in many different environments.
Because it can take in and process enormous amounts of data at one time, machine learning can improve imaging software that doctors use during operations to precisely locate the location of tumors or other abnormalities. AI definitely has a bright future in the medical field.