Predicting Hip-Knee-Ankle and Femorotibial Angles from Knee Radiographs with Deep Learning

Description

In this presentation, Jinhong Wang from Imperial College London explains a research project focusing on using deep learning techniques to predict knee alignment angles from knee radiographs. The study highlights the significance of knee alignment in diagnosing and treating osteoarthritis, emphasizing the difficulty and expense associated with traditional measurement methods like the Hip-Knee-Ankle angle (HKA) and the FemoroTibial Angle (FTA). The project employs convolutional neural networks to analyze plain radiographs, establishing separate models for FTA and HKA. Wang discusses the datasets used, the training process, and the accuracy of the predictions, showcasing the models' effectiveness via graphical representations and statistical analyses. The outcomes indicate a high level of accuracy in angle prediction, with 98.5% of FTA predictions and 83.9% of HKA predictions reflecting errors of less than 3 degrees. Wang also notes the potential for implementing this AI method in clinical settings due to its speed and cost-effectiveness, as well as the value of heat map analyses for evaluating prediction reliability. Questions from the audience address practical implementation, cost considerations, and the reasons for varying accuracy between the FTA and HKA measurements.

DOI: 10.1302/3114-221107

Specialties