Artificial Intelligence in Orthopedic Surgery: Current Applications, Challenges, and Future Directions

Abstract

Artificial intelligence (AI) drives transformative changes in orthopedic surgery, steering it toward precision and personalization through intelligent applications in preoperative planning, intraoperative assistance, and postoperative rehabilitation/monitoring. Breakthroughs in deep learning, robotics, and multimodal data fusion have enabled AI to demonstrate significant advantages. Nonetheless, current applications face challenges such as limited real-time decision autonomy, fragmented medical data silos, standardization gaps restricting model generalization, and ethical/regulatory frameworks lagging behind technological advancements. Therefore, a critical analysis of the current status of AI and the acceleration of its clinical translation is urgently required. This study systematically reviews the core advancements, challenges, and future directions of AI in orthopedic surgery from technical, clinical, and ethical perspectives. It elaborates on the "perceptual-decisional-executional" intelligent closed loop formed by algorithmic innovation and hardware upgrades, summarizes AI applications across surgical continuum, analyzes ethical and regulatory challenges, and explores emerging trajectories. This review integrates the end-to-end applications of AI in orthopedics, illustrating its evolution. It introduces an "algorithm-hardware-ethics trinity" framework for technical translation, providing methodological guidance for interdisciplinary collaboration. Additionally, it evaluates the combined efficacy of diverse algorithms and devices through practical cases and details of future research frontiers, aiming to inform researchers of current landscapes and guide subsequent investigations.

Keywords: artificial intelligence; deep learning; multimodal data fusion; orthopedic surgery; robot‐assisted surgery.

FIGURE 1

Key applications of CNN‐based AI in preoperative imaging analysis for orthopedic surgery. (A) The U‐Net model utilizes an encoder–decoder symmetric structure with skip connections to achieve high‐precision segmentation of orthopedic imaging data. (B) The GAN model enhances the usability and analytical precision of preoperative orthopedic imaging through an adversarial training mechanism between the generator and discriminator. (C) The YOLO model, with its single‐stage object detection architecture and multiscale feature fusion capability, improves the accuracy and efficiency of orthopedic preoperative image classification. (D) The Faster R‐CNN model integrates a region proposal network and a multitask loss function for object detection, enabling efficient localization of bony anatomical structures, precise identification of pathological regions, and end‐to‐end training optimization for classification performance. (E) The Enhanced Attention Res‐UNet model combines ResNet, U‐Net architecture, and an attention mechanism to achieve high‐precision segmentation and 3D reconstruction of medical images. (F) The 3D U‐Net model integrates 3D convolution operations, an encoder–decoder structure, skip connections, and multiscale feature fusion to perform high‐precision voxel‐wise segmentation of 3D medical images.

FIGURE 2

The role of AI in personalized rehabilitation programs. (A) Gait analysis dataset generation through self‐selected speed treadmill walking to determine 3D kinematic and dynamic changes. Adapted with permission [177]. Copyright © 2025, The Author(s): Stephan Oehme et al. (B) Gait data collection via wearable six‐axis sensors attached to both feet. Adapted with permission [178]. Copyright © 2025, Yoshikawa et al. (C) Axial T2‐weighted MRI analysis of lumbar muscle cross‐sectional area (CSA) using Philips MRI. Key parameters include A (CSA of the multifidus muscle in mm²), P (perimeter of the scribed area in mm), M (average gray value of the scribed area), and SD (standard deviation of the gray value). Adapted with permission [175]. Copyright © 2021, The Author(s): Zhen Lyu et al. (D) A participant using the immersive virtual reality scenario robot that integrates five subsystems: odometry and control (OC), human–robot‐environment interaction (HREI), human–robot interaction (HRI), motion capture (MC), and virtual reality integration (VRI). Adapted with permission [186]. Copyright © 2024, by the authors: Matheus Loureiro et al.