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

Abstract

Knee alignment affects the development and surgical treatment of knee osteoarthritis. This research aimed to improve its automatic measurement from posteroanterior (PA) knee radiographs by using convolutional neural networks. Radiographs from the Osteoarthritis Initiative with corresponding femorotibial angle (FTA, N = 6149) and hip-knee-ankle angle (HKA, N = 2351) measurements, were split into training, validation, and test datasets in a 70:15:15 ratio. Separate models were developed for the prediction of FTA and HKA and their accuracy quantified. Heat maps were then used to identify the anatomical features within each image that most contributed to the predicted angles. High accuracy was achieved for FTA (mean absolute error 0.8°), and better than state-of-the-art accuracy was achieved for HKA (mean absolute error 1.7°). Heat maps for both models were concentrated on the knee anatomy and could prove a valuable tool for assessing prediction reliability in clinical application. Deep learning techniques enable fast, reliable and accurate predictions of both FTA and HKA from plain knee radiographs and could lead to cost savings for healthcare providers and reduced radiation exposure for patients.

Mixed Reality and Artificial Intelligence: A Holistic Approach to Multimodal Visualization and Extended Interaction in Knee Osteotomy

“…CNNs achieved prediction accuracies of 88.2% and 86.3% for femoral components and tibial components, respectively [19]. In another study CNNs, trained on 6,149 radiographs, reported a mean absolute error of 0.8 • on femoratibial angle prediction [20]. In another study on 1,842 knees X-ray images, the error obtained by CNNs on hinge point, surgical point, and Fujisawa point was, respectively 2.06±1.16 mm, 2.71±1.45 mm, and 2.01±1.30 mm [30].…”

“…Concurrently with the emergence of MR technologies, there have been substantial advances in artificial intelligence (AI) tools, such as convolutional neural networks (CNN), that showcased the potential of greatly automatizing image processing in many different clinical applications, for diagnostic and surgical planning purposes [13]. In orthopedics, CNN models were applied for the segmentation of knee bones and cartilage from magnetic resonance imaging providing invaluable support in the diagnosis of osteoarthritis [14], [15], for the automatic segmentation of bones in knee CT scans for the realization of personalized cutting guides [7], [16], [17], for total knee arthroplasty planning using X-ray radiographs [18]- [20]. Some initial studies have explored the potential of integrating AI tools with technologies for extended reality in hip and knee arthroplasty [21].…”