Building a Predictive Model of Patellar Dislocation

One of our most recent projects in patellofemoral instability has been applying machine learning to be able to predict the risk of patellar instability in patients based on their knee MRI. Basically, what we've been doing is taking hundreds of MRIs of knees of patients who have had this condition, and entering multiple data points into this machine learning algorithm, and then comparing it to hundreds of patients who didn't have this condition, and building a predictive model. Once this is complete, take any MRI of any knee patient and put it through this algorithm and it can give you a prediction of what percent likelihood the patient would be to have a patellar dislocation.

I think this is pretty powerful in terms of the ability to predict the direction of the patient's condition. We often base our surgical indications on whether the patient has what we call "recurrent stability." So, they may have a first-time patellar dislocation, and whether they are going to go on to be a recurrent dislocator is not really that clear. And so, being able to risk-stratify and to give that patient a number, in terms of the clinical decision-making process, would be really helpful.

I think things like predictive modeling, really are the future of orthopedic surgery and a lot of medicine. Being able to apply this in terms of an individual's risk of having a problem or having a recurrent problem, or needing surgery, is all in this progression towards personalized medicine and being able to tailor our care and tailor our treatments to exactly what that individual needs.

The application of machine learning to anything that we're doing in terms of surgical decision-making, surgical treatment, is a huge step in our ability to predict a patient's symptoms and maybe even predict their outcomes. In terms of the field of patellofemoral surgery, it is not as straightforward as a lot of areas within sports medicine. When someone has an ACL tear, we fix the ACL. If a ligament or tendon is torn, we fix that, but in patellar stabilization surgery there are a lot of factors that come into play, and having this sort of machine learning to look at large amounts of data from all perspectives is a way to advance what we know and really quantify somebody's risk level before having more symptoms.

What I really enjoy about doing research at Massachusetts General Hospital is that I am surrounded by people who are brilliant and who are experts in their respective fields. Being able to collaborate with people from other fields especially has shed a lot of light on problems that are orthopedic problems but being able to look at them from an engineering perspective or from a radiology perspective or from other disciplines has really helped us to problem solve in a different way and to look at things from a different perspective. For example, this machine learning project is in collaboration with the Harris Lab, and they are the ones who have really developed these sorts of computational models and I think that this is an incredible way to collaborate and I think it brings the entire field forward.