The Problem

Running is one of the most popular and effective forms of physical activity with numerous benefits to cardiovascular system health. However, about half of all recreational runners experience an injury each year. Patellofemoral pain (PFP), informally known as runner’s knee, is one of the most common running related injuries. This condition is characterized by a feeling of pain around the anterior side of the knee joint, especially during activities that involve knee flexion such as walking, running, cycling, or even sitting for some.

PFP is a very challenging condition for clinicians because of its complicated diagnostic process requiring mobility and strength tests, subjective patient reported outcomes, and even radiographic imaging. Once diagnosed, clinicians will instruct patients to perform a variety of exercises to strengthen and mobilize the joints of the lower extremities. However, these interventions are not always effective, which can lead to the next challenge - ensuring that patients do not redevelop the pain in the future. PFP is known to have a very high recurrence rate, so the chance that a patient experiences the pain after an intervention is high. Additionally, previous research has shown that runners with PFP present many biomechanical adaptations, and that these biomechanical adaptations should be restored in order to avoid longitudinal pain symptoms.

This leaves us with two major problems:

1. Initial diagnostic procedures are challenging
2. There is no way to understand if an intervention was effective at restoring healthy movement patterns and/or biomechanics

The Solution

To solve the first problem, my dissertation aims to develop a computer aided diagnostic system using wearable sensing systems and deep learning models. 3D optical motion capture, the gold-standard system for measuring biomechanics, are not typically available in clinical settings. However, wearable sensors offer a cost-effective and user-friendly solution for clinicians. Instead of requiring a full laboratory and personell to run the equipment, wearable sensors can be used right in a clinic to measure movement patterns. We hypothesize that a deep learning model will accurately differentiate healthy runners from those with PFP.

In addition to the initial diagnosis, this system will also be beneficial to clinicians for quantifying if an intervention was effective at restoring biomechanics. Research has shown that these biomechanical abnormalities need to be corrected in order for a person to stop experiencing PFP. Understanding this is challenging unless a clinician has access to some objective method of measuring if movement patterns have changed. Our system will do just this - classify movement patterns as healthy or unhealthy using signals recorded from wearable devices

The How

Through a unique collaboration of individuals with expertise in biomechanics, running, wearable sensors, and artificial intelligence, I will be guided through the process of designing, executing, and summarizing my dissertation project. Briefly, running gait data on healthy subjects and those with a diagnosis of PFP will be collected using inertial measurement units and pressure sensitive shoe insoles. Data from these wearable sensor will be fed into a deep learning model to predict if the running patterns resemble a healthy person or someone with PFP.

This project will be completed with three specific aims:

1. Draw correlations between wearable sensor data and data that is captured through traditional biomechanical analysis using 3D optical motion capture.
2. Characterize running gait of healthy runners and those with PFP using signals derived from wearable sensors, and relate to the findings of aim 1.
3. Develop a deep learning model that accurately classifies wearable sensor derived running patterns of those who are healthy and who have a diagnosis of PFP.

Progress Update

I am currently working on proposing this project and collecting some early stage pilot data.