Images produced by the researchers' algorithm, shown in the center column, are much closer to high-resolution scans shown in the right column. In the left column are images produced by a different technique for improving low-resolution scans. Photo: MIT researchers

Have you ever taken a great photo that you couldn’t wait to print out and frame? But when you tried to print it, an error message appeared saying it was “too low resolution.”

Take that feeling and amplify it—that’s how researchers studying the brain feel about every patient MRI taken at a hospital.

When patients suffer a stroke, medical doctors often take an MRI to determine location and extent of damage. But due to time and the critical nature of the patient population, the scans are sparse—sometimes producing an image that is only 1/7th that of a brain scan taken for research purposes. Bottom line: there is a wealth of untapped data at hospitals.

Now, researchers at MIT, working with Massachusetts General Hospital as well as other medical centers, have developed a way to boost the quality of hospital scans so clinical researchers can use them to study stroke survival and neurological diseases like Alzheimer’s.

“These images are quite unique because they are acquired in routine clinical practice when a patient comes in with a stroke,” says Polina Golland, an MIT professor of electrical engineering and computer science, and senior author of the recently published paper. “You couldn’t stage a study like that.”

Golland, Adrian Dalca, lead author of the paper, and their team established a new approach that, essentially, fills in the blanks.

Scanning the brain with MRI produces many 2-D “slices” that can be combined to form a 3-D representation of the brain. In a clinical setting, image slices are taken about 5 to 7 millimeters apart. In a research setting, image slices are only about 1 millimeter apart. Thus, the new MIT approach involves filling in the data that is missing from each patient scan. The researchers do so by taking information from an entire set of scans, and then using it to recreate anatomical features that are missing from other scans.

Filling in the blanks of the clinical image is almost like an educated guess—researchers have most of the picture, and they can use Parts A and B to rather confidently reconstruct the missing Part C.

“The key is to generate an image that is anatomically plausible, and to an algorithm looks like one of those research scans, and is completely consistent with clinical images that were acquired,” Golland says. “Once you have that, you can apply every state-of-the-art algorithm that was developed for the beautiful research images and run the same analysis, and get the results as if these were the research images.”

According to the MIT researchers, once the research-quality images are generated in this way, they can then run algorithms designed to help with analyzing anatomical features. These include the alignment of image slices and a process called “skull-stripping,” which eliminates everything but the brain from the final images.

One of the biggest perks of the new imaging approach is that the algorithm keeps track of which slices came from the original scans, and which were filled in afterward. Therefore, if any clinical analysis needs to be conducted later, the original patient scans are fully intact.

Going forward, Golland’s team intends to apply this new approach to 4,000 scans from stroke patients, obtained from 12 medical center partners involved with the original study.

“Understanding spatial patterns of the damage that is done to the white matter promises to help us understand in more detail how the disease interacts with cognitive abilities of the person, with their ability to recover from stroke, and so on,” Golland said.

In the future, the researchers also hope to apply the technique to scans of patients with other brain disorders as well.