AI-Based Knee Implant Recommendation from Radiographic Images using Deep Learning

Abstract

In clinical decision-making for knee osteoarthritis management, determining whether a patient requires no implant, a partial knee replacement, or a total knee replacement is crucial. Manual interpretation of radiographs can be subjective and time-consuming. This study presents a deep learning-based framework that maps Kellgren–Lawrence grades to implant recommendations using four pretrained CNN architectures: ResNet50, MobileNetV2, EfficientNet-B0, and VGG16. Using a dataset of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$1,240 \text{AP}$</tex>-view knee radiographs, the models were trained with transfer learning and evaluated using accuracy, precision, recall, and Grad-CAM interpretability. EfficientNet-B0 achieved the highest accuracy of 91 %, demonstrating superior generalization and clinically relevant visual attention patterns. The proposed system shows strong potential as an automated decision-support tool for orthopedic practitioners.

Ähnliche Arbeiten