The application of artificial intelligence in the design of highly compatible knee prostheses: a systematic review

With the rapid development of artificial intelligence (AI) technology, it has been widely used in the personalized design of orthopedic implants. Especially in total knee arthroplasty (TKA), AI has shown significant potential in prosthesis size prediction, implant identification and surgical planning. However, the research distribution in this field is scattered, the technical paths are diverse, and there is still a lack of systematic sorting and comprehensive evaluation. A systematic literature review was conducted to retrieve relevant literature from PubMed, Scopus, and Web of Science up to December 2025. Due to substantial heterogeneity across the included studies, a meta-analysis was not performed; instead, a descriptive synthesis approach was adopted. The inclusion criteria covered the application of AI in TKA prosthesis design, size prediction, type identification and surgical planning. Non-knee joint, non-quantitative research and review literature were excluded. A total of 16 studies were included. The descriptive comprehensive method was used to analyze the research characteristics, technical methods, application direction and performance. A total of 16 studies from 9 countries published between 2019 and 2025 were included. AI has been mainly applied in three directions in TKA prosthesis adaptation: prosthesis size prediction (12 articles), prosthesis type and manufacturer identification (4 articles), surgical planning and personalized design (4 articles). In size prediction, deep learning models, especially CNN, have excellent performance, with accuracy generally between 77 and 91%, and some models can reach more than 99% when allowing ± 1 size error. In the prosthesis recognition task, the model based on EfficientNet, YOLO and other architectures is close to perfect in terms of classification accuracy and AUC. In surgical planning and custom design, AI can automate the whole process from image segmentation to prosthesis generation, and the system processing time can be shortened to less than 15 mins. Despite the positive results, there are common limitations among studies, such as data heterogeneity, sample size differences, and insufficient clinical validation. Preliminary evidence suggests that AI holds promise in improving preoperative planning accuracy and enabling personalized prosthesis design. However, high-level clinical validation is urgently needed to confirm its impact on long-term outcomes such as prosthesis survival and functional recovery.