The Mobilize/Restore center at Stanford invites you to join our next webinar, featuring Kevin Thomas from Stanford University.

DETAILS
Title: Accelerating Image-Based Knee Osteoarthritis Research Using Deep Learning: Research Examples and Best Practices
Speaker: Kevin Thomas, Stanford University
Time: Tuesday, Aug 17, 2021 at 10:00 AM Pacific Time
Registration: Click here to register

ABSTRACT
Deep learning is changing the way we study human mobility, driving many biomechanists and imaging researchers to begin adopting these methods in their research. Using deep learning to automatically assess individuals’ disease status from medical images is one active area of research that is poised to change how outcomes are measured in our field. In this webinar, we will demonstrate how to use deep learning models to automatically analyze knee X-rays and MRIs of individuals with osteoarthritis. We will also share tips and tricks for conducting similar analyses in your own research.

In the first half of the webinar, we will discuss our recent work to develop deep learning models to assess osteoarthritis severity and cartilage health. Assessing knee osteoarthritis severity from medical images is currently a time-consuming process that requires significant training and expertise. This limits the number of research groups able to utilize medical imaging in their work and makes it difficult to assess large cohorts. To address this, we trained deep learning models to automatically assess osteoarthritis severity from X-rays and to track cartilage health longitudinally from MRIs. These models are available for others to test with their own images. We will describe these models and show that they agree with experts as closely as experts agree with one another.

The second half of the webinar will cover strategies for training high-performing deep learning models using examples from our osteoarthritis research. We will discuss how to diagnose common causes of poor-performing models and provide strategies to address each common cause (e.g., data augmentation, regularization, model architecture changes). Examples will focus on imaging, but the discussion will provide context for applying these strategies to a wide variety of machine learning projects.