A group of radiologists and engineers based at Stanford University, California, USA have made an unexpected observation about people with lower back pain. In the course of an AI-powered survey of thousands of existing CT scans, the radiologists discovered that people with chronic lower back pain often have smaller abdominal muscles relative to their abdominal fat than people who don’t.
It turns out that lower back pain increases your chances of having a higher ratio of abdominal fat to muscles. In practical terms this means weaker muscles in your core could be a factor in spinal degeneration of our lower backs as we age. It’s not the absolute size of our muscles that matters, but how big they are relative to the amount of fat we are carrying around.
The team of engineers and medical imaging experts, headed up by mechanical engineering PhD student Kathryn Marusich and Akshay Chaudhari, PhD, of Stanford’s Department of Radiology, are using the latest open source AI radiology tools to solve muscular skeletal medical mysteries.
Muscular Skeletal Mystery
One of those is, what is going on with our lower backs as we get older? Why do some people develop chronic lower back pain, while others stay limber into advanced age? Some radiologists have found that fatty deposits make their way into places they shouldn’t. Other experts say that losing muscle as we age or losing weight might destabilize our spines. There are all sorts of possibilities.
The problem is everybody’s body develops its own wear and tear. The small differences between individuals are so many that it’s hard to see patterns. If we want to discover clues in the commonalities, we must to take a bird’s-eye view, looking at a lot of people with lower back pain and a lot more who don’t. But where do we find these people and what are we searching for? How do we look? You’d need X-ray vision. Luckily for Marusich and her colleagues at Stanford, they had just that.
X-Ray Vision
The researchers had access to an enormous collection of CT scans performed at a single hospital between 2013 and 2018. Patients had agreed that their anonymized CT scans and records could be used for research. These were abdominal scans taken as part of routine care. If the researchers could find a way to analyse them, this data set could help the team figure out whether there are patterns in the distribution of fat and muscle that turn up in people with lower back pain. This was an ambitious plan. The team would have to review 33,548 abdominal CT scans and 23,508 medical records and look for similarities and differences in muscle composition, fat distribution and other variables.
Using an open source AI radiology tool, Comp2Comp, the researchers were able to crunch their way through tens of thousands of images. Comp2Comp uses deep learning to recognize and measure muscle mass, visceral fat tissue and adipose fat deposits in each frame of the CT scan. Importantly, the tool can also make sure that all the measurements are taken from the same part of the body. If we want to look for patterns, we have to compare like with like!
Analyse This!
Once the team had comparable measurements for every CT scan, they were able to use statistical tools to look for common differences between people with back pain and people without.
The team chose to specifically compare the fat and muscle composition of every patient at lumbar vertebra 3.
Before they looked at any scans or associations, the team decided to exclude participants who had a medical history that would interfere with the results or incomplete records. That left 10,835 patients. Next they sorted the patients by age, sex and BMI to make sure they were comparing people with bad backs to people similar enough to them to draw conclusions. Comparing the CT scan of an 18-year-old to a 90-year-old isn’t helpful.
In the end the examined scans from 6,121 female patients and 4,714 male patients with a median age of 51 years. The first thing they did was figure out how fat and muscle distribution changes in our lower back as we age. They found that in males and females, as we get older, the size of our muscles gets smaller. In women the size of our muscles seemed to peak around 30–39 years and in men at 40–49.
They also saw that aging led to an increase in the ratio of visceral fat to subcutaneous fat, with relative visceral fat levels getting larger. Overall it appears that the ratio of total abdominal fat to muscle increases slightly. So we lose muscle and gain some fat. They also saw that the size of our muscles increased with BMI. This is probably because they need to be stronger to hold everything in place when we are bigger.
Lower back pain vs abdominal muscles
So now we have an idea of what happens over all. Did people with chronic lower back pain show any special patterns? They compared 2,622 scans of patients with lower back pain to 2,577 matched controls who didn’t have lower back pain in their records. The only muscle or fat related difference between the groups was in the relative size of their abdominal muscles compared to the amount of fat they had. The absolute size of the muscles alone didn’t affect the chances of that person having lower back pain, it was only when their abdominal muscles were disproportionately small compared to the amount of fat.
So what this means is that people if you take two people of the same BMI the one with lower back pain, on average, would have less muscle mass in their abdomen compared to the amount of fat. What does this mean for treating lower back pain? It’s too soon to tell.
This is just the first clue- but it could be a very valuable one. The work now, for orthopaedists and physiotherapists, is to figure out whether lower back pain makes us change how we move in a way that makes our muscles weaker or if weak muscles are contributing to spinal degeneration and lower back pain.
AI opens New Doors
This is a story about how AI tools can help medical researchers to do projects that they would never otherwise be able to do.
No radiologists’ careers were harmed in the making of this project, simply because nobody could afford (in monetary terms or their sanity) to spend years doing nothing but inspecting frames from CT scans all day. As Marusich points out, ‘Even a few years ago, a similar imaging study would have been limited by the number of images a research team could manually analyse to identify muscle and fat extent. Modern AI tools now allow us to process entire databases of clinical images and link them with other health record data at unprecedented speed and scale.’
Source
Marusich KR, Gatti AA, Blankemeier L, et al. Deep learning reveals relationships between muscle and adipose tissue and their associations with age, body mass index, and low back pain: A cross-sectional study in 10,835 subjects. PLOS Digital Health. 2026;5(11):e0001788. doi:10.1371/journal.pdig.0001788