From Dr. Ramkumar’s original LinkedIn Post: 

Prem Ramkumar Leads Landmark AI and 3D CT Study on Arthritic Knee Anatomy

The study below represents four years of work and the starting point for much more. Knee surgeons agree that anatomy is the foundation of our work. We also agree there is broad anatomic variation that requires 3D thinking at the time of knee replacement. That’s pretty much all we agree upon. We all have our own preferences with respect to technique, alignment, implant, fixation, and adjunctive technology.

Moving Beyond 2D: How AI Unlocks Triplanar Knee Anatomy

The literature over the last hundred years has been written from the 2D perspective of x-rays. With the advent of preoperative CT-based imaging, we have the newfound ability to critically evaluate arthritic knee anatomy across coronal, sagittal, and axial planes to better appreciate anatomic variation. Although surgeons intuitively approximate this at the time of surgery, the evaluation of triplanar anatomy is mathematically complex.

Without a robust, validated artificial intelligence-based workflow, the simultaneous processing of triplanar data for 27 well-established anatomic landmarks across 1,352 CTs with 11,681,280 images and over 1.2 terabytes of data would be simply impossible. The emergence—and convergence—of preoperative CT imaging and unsupervised AI is nothing short of serendipity.

Dr. Ramkumar Establishes a New Foundation for Comparing TKA Outcomes

If we acknowledge that anatomy is our North Star and that arthritic knees are not all the same prior to surgery, how can we reasonably expect to compare our techniques and technologies, let alone our outcomes?

Thus, the purpose of the work was to three-dimensionally phenotype the broad variation of arthritic knee anatomy to establish a foundational language for longitudinal comparison along the episode of care for TKA.

Four Distinct Knee Phenotypes Identified in Groundbreaking Study

Quite frankly, we had no idea how many phenotypes would exist when you’re simultaneously evaluating 27 variables in three planes across 1,352 knees. We used spectral clustering because it embraces statistical “sensory overload” and acknowledges that nonlinear relationships exist and asymmetric proportions of populations exist.

We were surprised to find just a total of four Types exist, with two groups embodying obvious outliers. These outlier phenotypes, Type 1 and 3, are distinct from the pack and embody 26% of patients with extreme hip rotation, medial PTS, HKAA TFA, mPTA, and lDFA parameters. Type 2 and 4 knees have subtle differences that remain distinct but may converge as we add more patients across the globe; conversely, the gap may widen and even new clusters may emerge.

 

Toward Personalized Knee Replacement: Future Pathways for TKA

Future work must contextualize the preop Type to implant type, intraop technique, applied technology, and outcome. Perhaps this foundation may lead to pathways that allow for better balancing, more rapid recovery, and greater satisfaction after TKA with a more “personalized” approach that takes into account the uniqueness of each knee.

Thank you to Joshua Woo, Sayyida Hasan, Ben Zhang, Antonia Chen, Andrew Wassef, Danyal H. Nawabi, Cory Calendine, MD, Viktor Krebs, The Journal of Bone and Joint Surgery, Inc., and Commons Clinic for the support.

You can read more about the study here: 

Distinct 3-Dimensional Morphologies of Arthritic Knee Anatomy Exist