DG-PPU: Dynamical Graphs Based Post-Processing of Point Clouds Extracted from Knee Ultrasounds

Patellofemoral joint (PFJ) pain affects one in four people, with one in five experiencing chronic knee pain despite treatment. Incorrect patellar tracking after arthroplasty may contribute to poor outcomes and ongoing pain. Traditional imaging methods such as CT and MRI have limitations when it comes to visualising PFJ motion. Our goal is to improve the visualisation of patellar tracking and PFJ motion by utilising 3D registration of point clouds obtained from freehand ultrasound scans taken at various flexion angles. Soft tissues are often misidentified as bone during segmentation, leading to noisy 3D point clouds that hinder accurate registration of the bony joint anatomy. Utilising machine learning to analyse the intrinsic geometry of the knee may help eliminate these false positives, as the geometry of the knee remains consistent during PFJ motion. Our dynamical graphs-based post-processing of ultrasound (DG-PPU) algorithm effectively generates smoother point clouds that accurately represent the bony knee anatomy at various joint flexion angles. Point clouds were converted back to 2D and visually evaluated against the original ultrasound images. DG-PPU outperformed manual data cleaning performed by author CVC, achieving a precision of 98.2% in deleting false positives and noise across three different angles of joint flexion. DG-PPU is the first algorithm specifically developed to directly clean 3D point clouds generated from ultrasound scans, bypassing traditional 2D cleaning methods. Hence, it facilitates the development of a novel assessment system for patellar maltracking, which currently lacks a viable solution.