Artificial Intelligence-Assisted Preoperative Planning Combined with Patient-Specific Instrumentation Improves the Accuracy of Unicompartmental Knee Arthroplasty

Artificial Intelligence-Assisted Preoperative Planning Combined with Patient-Specific Instrumentation Improves the Accuracy of Unicompartmental Knee Arthroplasty | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Artificial Intelligence-Assisted Preoperative Planning Combined with Patient-Specific Instrumentation Improves the Accuracy of Unicompartmental Knee Arthroplasty Dehua Liu, Gang Ji, Zirun Gao, Ye Huang, Guanglei Cao, Yafang Zhang, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9340442/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 6 You are reading this latest preprint version Abstract Background Artificial intelligence (AI)-assisted preoperative planning based on computed tomography (CT) may improve anatomic characterization for unicompartmental knee arthroplasty (UKA), and patient-specific instrumentation (PSI) may improve the accuracy of component positioning. We developed and validated an AI-assisted preoperative planning workflow combined with PSI for medial UKA and evaluated its effect on implant positioning accuracy. Methods A hybrid architecture combining a convolutional neural network-based U-Net with a Transformer-based deep learning module (C-T Module) was developed to automate CT processing for AI-assisted preoperative planning and PSI design in UKA. Segmentation performance of the C-T Module was compared with that of a conventional 3D U-Net. PSI feasibility was validated using synthetic bone models. In a prospective randomized clinical study, 24 patients underwent AI-based planning plus PSI-assisted UKA (PSI group) and 24 underwent AI-based planning plus conventionally instrumented UKA (control group). Surgical accuracy, perioperative outcomes, and implant-size prediction accuracy were compared. Results The C-T Module demonstrated superior image segmentation accuracy compared to the classical 3D U-Net. Compared with the control group, the PSI group significantly improved surgical accuracy, including more accurate tibial component positioning, greater tibial coverage, and less deviation in proximal tibial resection (all P < 0.001). Except for the significantly longer skin incision in the PSI group ( P < 0.001), no other perioperative parameters differed significantly between groups. No evident learning curve was observed with PSI use. Moreover, the AI-based planning system demonstrated significantly higher accuracy in prosthesis size prediction than conventional templating ( P < 0.001). Conclusion The novel AI-assisted planning system accurately predicts prosthesis size and, when combined with PSI, significantly improved component positioning accuracy of UKA without a notable learning curve. patient-specific instrumentation artificial intelligence unicompartmental knee arthroplasty surgical precision learning curve Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Unicompartmental knee arthroplasty (UKA) is an effective treatment for end-stage unicompartmental osteoarthritis of the knee, offering advantages such as minimally invasive surgery, rapid rehabilitation, and favorable postoperative proprioception( 1 , 2 ). Numerous studies have shown that long-term success of UKA relies on refined surgical techniques. Optimal prosthetic component positioning, meticulous soft tissue balancing, and appropriate implant sizing are fundamental to achieving satisfactory clinical outcomes( 3 , 4 ). Suboptimal component positioning and sizing have been associated with accelerated wear, instability, bearing dislocation, subsidence, and persistent postoperative pain( 5 , 6 ). Therefore, precise preoperative planning and intraoperative execution are paramount to preempting surgical unpredictability, ensuring accurate component placement, and minimizing associated logistical and sterilization costs. Robotic and navigation-assisted techniques have been reported to improve surgical accuracy of UKA( 7 , 8 ). However, their widespread adoption has been limited by prolonged operative time, high equipment acquisition and maintenance costs, and steep learning curves—particularly in lower-resource healthcare settings( 9 , 10 ). In recent years, increasing attention has been directed toward the role of patient-specific instrumentation (PSI) in enhancing UKA precision( 11 , 12 ). PSI has attracted interest because of its conceptual simplicity, lower infrastructure demands, and potential workflow advantages However, traditional PSI designs are complex and require repeated validation of bone surface conformity across multiple anatomical planes, often resulting in production cycles lasting several days( 13 ). Moreover, the impact of PSI on surgical accuracy in UKA remains controversial( 14 ). While some studies have demonstrated that PSI significantly improves the precision of component placement, others have reported inconsistent outcomes or even detrimental effects on surgical precision( 15 ). In recent years, the integration of artificial intelligence (AI) with three-dimensional (3D) printing technologies has shown promising potential in clinical applications( 16 , 17 ). Leveraging AI’s efficiency in image segmentation and large-scale data processing may enhance both the accuracy and production efficiency of PSI systems. We therefore developed an AI-assisted preoperative planning workflow combined with PSI and evaluated whether this approach improved component positioning accuracy, resection accuracy, and implant-size prediction in medial UKA. Methods AI-Based Image Processing and Segmentation Workflow The AI plan system integrates a U-Net–based convolutional neural network (CNN) with a Transformer-based deep learning module (C-T Module) for image segmentation and landmark recognition. The segmentation pipeline includes a residual 3D CNN encoder for local feature extraction and a decoder combining CNN and Transformer branches for global context and fine detail enhancement. Outputs are fused via element-wise addition to optimize segmentation precision (Fig. 1 -A). Compared with a deep learning-based 3D U-Net model, segmentation performance was evaluated using Dice Similarity Coefficient (DSC), Intersection over Union (IoU), and 95th Percentile Hausdorff Distance (HD95)( 18 , 19 ). For landmark recognition, a low-resolution sub-network operates in parallel with the high-resolution backbone, enabling multi-scale feature fusion through staged interactions, with final outputs generated by the backbone network (Fig. 1 -B). Preoperative Planning The system generated automated 3-dimensional preoperative plans using full-length lower extremity CT scans and standardized radiographs. Stress radiographs were additionally reviewed to assess ligament function. Tibial component planning included resection level, posterior slope, coronal and rotational alignment, with adaptive adjustments for implant coverage. Femoral planning follows, and the system simulates postoperative implant–bone location, implant sizes, and resection volumes, with final confirmation by the surgeon (Fig. 2 -A). PSI was designed based on individual bony anatomy and osteophytes to maximize fit and minimize exposure. Tibial and femoral guide bone cuts via osteotomy slots, with depth control and distal marking aiding posterior femoral resections (Fig. 2 -B). The PSI was manufactured in nylon by 3-dimensional printing, and the complete design-to-fabrication workflow was completed within 12 hours (Figs. 2 -C, D). The Feasibility Verification of PSI Six synthetic bone models were utilized to validate the precision of the AI-assisted planning system. A 3D scanner (EinScan Pro 2X V2) was employed to digitize the bone models, and the resulting data were imported into the planning system to generate a surgical plan and fabricate PSI. After performing the surgery through PSI, the prosthesis was installed. Postoperative 3D scanning of the bone and prosthetic position was performed, and the data were archived. After model reconstruction, the discrepancy between the actual prosthetic position and the preoperative plan was measured. Clinical Research Design This prospective randomized study was approved by the institutional review board (approval number: 2025-YanShen00212). The study was registered in the China Medical Research Registration and Filing Information System. From January 2nd 2025 to August 15th 2025, 53 patients undergoing UKA were enrolled. Inclusion required standard UKA indications, completion of preoperative planning, and informed consent. Exclusion criteria included contraindications, enrollment in other studies, or refusal to participate. Patients were randomized in a 1:1 ratio using a computer-generated randomization sequence. Allocation concealment was achieved using sequentially numbered, opaque, sealed envelopes, which were opened only after patient enrollment. Patients were assigned to either the PSI group (AI-based planning plus PSI-assisted UKA) or the control group (AI-based planning plus conventionally instrumented UKA). The primary outcome was postoperative tibial coronal alignment deviation (varus-valgus angle) from the preoperative plan. Secondary outcomes included femoral and tibial component alignment in the remaining planes, tibial prosthesis coverage, osteotomy deviation, perioperative outcomes, and implant-size prediction accuracy. Because of the nature of the intervention, the operating surgeon could not be blinded to group allocation. Participants were blinded to group allocation. Radiographic measurements and implant-size evaluations were performed by independent assessors blinded to patient information and group allocation. The study flow diagram is shown in Fig. 3 . Demographic Characteristics There were no statistically significant differences between the PSI group and the control group in demographic variables, preoperative hemoglobin level, hematocrit, or ASA classification (Table Sup I). Surgical Technique 3-dimensional preoperative planning was performed for all patients in in both groups. All patients underwent medial UKA performed by a single surgeon using a fixed-bearing prosthesis. All procedures were performed by a single high-volume UKA surgeon who had no prior clinical experience with this specific PSI workflow. Procedures were performed with a tourniquet via an anteromedial parapatellar approach. After soft tissue release, tibial and femoral bone surfaces were exposed and cartilage in the region of the guide attachment was removed. Guides were fixed with pins, followed by vertical/horizontal tibial osteotomies and femoral distal resection. Using an electrocautery device, a line was marked along the pre-designed femoral notch on the guide, serving as a reference for posterior femoral condyle resection using the guide. Subsequent steps followed the standard conventional UKA procedure. Resected bone blocks were collected to compare with preoperative plans. No drainage was used postoperatively, and routine antibiotic and anticoagulation therapy were administered. Uniform perioperative care and surgical education were provided to all patients. Radiological Outcomes Postoperative CT and radiographs were obtained for each patient. CT datasets were processed using Mimics and UG software for model reconstruction. Preoperative and postoperative skeletal models were aligned to assess deviations in prosthesis positions. Measurements were conducted by two independent observers. Inter-observer reliability coefficients (ICC) were calculated, and the mean of the two measurements was used for statistical analysis. Parameters measured included varus/valgus, internal/external rotation, and flexion deviation angles of the femoral component; varus/valgus, internal/external rotation, and posterior tilt deviation angles of the tibial component; and tibial prosthesis coverage rate (Figure Sup1-A to 1-F). Perioperative Parameters Surgical duration, perioperative adverse events, estimated blood loss, length of hospital stay and length of incision were recorded for each group. Surgical duration was defined as the time interval from incision to wound closure. Estimated blood loss was calculated according to published methods, based on the change in hematocrit from before to after surgery and the patient’s body weight( 20 ). Learning Curve To assess the learning curve associated with PSI-assisted UKA, PSI patients were stratified into three cohorts based on case sequence: Group A (cases 1–8), Group B (cases 9–16), and Group C (cases 17–24). Comparative analyses were conducted across these groups focusing on operative duration, intraoperative blood loss, prosthetic alignment deviations, and tibial prosthesis coverage rate. Comparison of Implant Size Planning Accuracy Two independent observers, blinded to patient information, estimated the implant sizes based on preoperative radiographs using standardized sizing templates provided by the implant manufacturer. Prior to evaluation, radiographic magnification was calibrated uniformly using the templates. All estimations were reviewed by an additional senior orthopedic surgeon. The predicted implant sizes and those generated by the AI–based planning system were each compared with the actual implant sizes. Sample Size Calculation The sample size was estimated based on the primary outcome of postoperative tibial coronal alignment (varus-valgus angle) after UKA. According to the study by Alvand et al., which compared UKA performed with patient-specific instrumentation and conventional instrumentation using similar radiographic parameters, the assumed standard deviation of tibial coronal alignment in the control group was 3.6°. A between-group difference of 3.0° was considered clinically relevant( 21 ). Using a 2-sided alpha of 0.05 and a power of 0.80, the minimum required sample size was 44 patients (22 per group). To account for potential loss to follow-up and to improve study robustness, the planned enrollment was increased to 48 patients (24 per group). Data Analyses Absolute angular deviations were defined as the absolute difference between the planned and postoperative measurements. Statistical analyses were performed using SPSS version 25.0. Continuous variables were expressed as mean(ranges) and mean ± standard deviation, as appropriate, whereas categorical variables were presented as counts and percentages. For continuous data with normal distribution, depending on the situation, either the independent-samples t-test or one-way analysis of variance (ANOVA) was used to compare the differences between different groups; for non-normally distributed data, nonparametric tests were applied. Categorical data were analyzed using the chi-square test or Fisher’s exact test, as appropriate. The significance level was set at α = 0.05, and a P < 0.05 was considered statistically significant. The root mean square error (RMSE) was calculated to quantify angular deviations and was compared with previously published data on robotic-assisted UKA( 8 ). Results Image Segmentation Accuracy The C-T Module demonstrated superior automatic segmentation accuracy of CT images compared to the 3D U-Net (Figure 4-A). The DCS for the C-T Module was 0.9409, higher than the 3D U-Net's 0.8914. The IoU for the C-T Module was 0.8884, surpassing the 3D U-Net's 0.8041. The 95th percentile HD95 for the C-T Module was 14.98, lower than the 3D U-Net's 21.23 (Figure 4-B). Figure. 4. Comparison of Image Segmentation Accuracy. (A) Direct visual comparison of image segmentation performance. (B) Comparison of segmentation accuracy metrics. The Feasibility Verification of PSI The absolute angular deviations for the femoral component in the coronal, sagittal, and axial planes were 1.2 ± 0.7°, 1.0 ± 0.4°, and 0.9 ± 0.3°. The absolute angular deviations for the tibial component in the coronal, sagittal, and axial planes were 1.0 ± 0.5°, 1.2 ± 0.6°, and 0.8 ± 0.5°. The mean tibial prosthesis coverage rate was 97.0 ± 0.7%. Except for the femoral coronal deviation, the absolute angular deviations in all other planes were within 2° (Table I). Table I Deviation of prosthesis placement of bone model. PSI Femoral coronal deviation (°) 1.2(0.3-2.2) ≤2° 5(83.3%) Femoral sagittal deviation (°) 1.0(0.6-1.6) ≤2° 6(100%) Femoral axial deviation (°) 0.9(0.6-1.3) ≤2° 6(100%) Tibial coronal deviation (°) 1.0(0.5-1.8) ≤2° 6(100%) Tibial sagittal deviation (°) 1.2(0.5-1.9) ≤2° 6(100%) Tibial axial deviation (°) 0.8(0.1-1.5) ≤2° 6(100%) Tibial prosthesis coverage rate (%) 97.0(95.8-97.7) ≥90% 6(100%) Presented as the mean (ranges) and number (percentage). Abbreviations: PSI, patient-specific instrumentation. Surgical Precision All ICCs for angular measurements exceeded 0.9. The PSI group demonstrated significantly lower absolute errors than the control group for femoral coronal (2.0 ± 1.1° vs. 4.4 ± 2.6°, P =0.003), sagittal (3.4 ± 1.8° vs. 5.5± 2.8°, P =0.005), and axial deviations (3.2 ± 1.6° vs. 4.6 ± 2.1°, P =0.008). No significant difference was observed in the proportion of femoral components within 2° between groups. Tibial component positioning was also more accurate in the PSI group across coronal (1.5 ± 0.8° vs. 2.9 ± 1.1°), sagittal (1.5 ± 0.6° vs. 3.1 ± 1.5°), and axial planes (2.4 ±1.2° vs. 6.7 ± 2.6°), all P <0.001. A greater proportion of tibial components in the PSI group were positioned within 2° of the plans in all 3 planes ( P < 0.05) Tibial prosthesis coverage was higher in the PSI group (93.8% ± 2.3% vs. 90.3% ± 3.7%, P <0.001). The PSI group showed significantly lower proximal tibial osteotomy deviation (0.6 ± 0.2 mm vs. 1.1 ± 0.4 mm, P <0.001), while distal femoral deviation showed no significant difference. A greater proportion of osteotomies with a deviation of ≤ 1 mm was observed in the PSI group ( P =0.01). The detailed comparison is presented in Table II. And Figure 5A-D. Furthermore, the outcomes achieved with the PSI in this study were comparable to those reported for robotic-assisted UKA (Figure 5-E). Table II Deviation of prosthesis placement and osteotomy amount. PSI Control P Femoral coronal deviation (°) 2.0(0.5-4.0) 4.4(0.3-8.3) 0.003 ≤2° 13(54%) 9(38%) 0.773 Femoral sagittal deviation (°) 3.4(0.7-6.4) 5.5(0.6-9.9) 0.005 ≤2° 10(42%) 5(21%) 0.212 Femoral axial deviation (°) 3.2(0.8-5.7) 4.6(0.5-7.0) 0.008 ≤2° 10(42%) 6(25%) 0.359 Tibial coronal deviation (°) 1.5(0.3-3.0) 2.9(1.1-5.2) <0.001 ≤2° 18(75%) 8(33%) 0.008 Tibial sagittal deviation (°) 1.5(0.3-2.5) 3.1(0.6-5.2) <0.001 ≤2° 20(83%) 9(38%) 0.003 Tibial axial deviation (°) 2.4(0.6-4.0) 6.7(0.9-10.5) <0.001 ≤2° 11(46%) 3(13%) 0.024 Tibial prosthesis coverage rate (%) 93.8(89.3-98.9) 90.3(82.7-95.4) <0.001 ≥90% 22(92%) 16(67%) 0.072 Femoral distal osteotomy deviation(mm) 0.7(0.4-1.2) 1.0(0.2-2.1) 0.093 ≤1 mm 23(96%) 15(63%) 0.010 Tibial osteotomy deviation (mm) 0.6(0.3-1.3) 1.1(0.7-2.4) <0.001 ≤1 mm 23(96%) 15(63%) 0.010 Presented as the mean (ranges) and number (percentage). Abbreviations: PSI, patient-specific instrumentation. Figure. 5. Comparison of surgical precision. (A) Deviation of femoral prosthesis. (B) Deviation of tibial prosthesis. (C) Deviation of osteotomy. (D) Tibial prosthesis coverage rate. (E) Comparation of RMSE values of component positioning angles for the current study, as well as the robot-assisted UKA by Bell et al. Note: RMSE(root mean square). Data are shown as mean ± SD. * P < 0.05, ** P < 0.01, *** P < 0.001, ns = not significant. Perioperative Parameters Operative time, blood loss, and length of stay did not differ between groups. Incision length of the PSI group (9.9 ± 0.6 cm) was significantly longer than the control group (8.2± 0.5 cm) ( P < 0.001). Detailed data are presented in Table III. No serious surgery-related complications occurred during hospitalization in either group. Table III Perioperative related indicators. PSI Control P Operation time (min) 81.4(65.0-110.0) 84.2(65.0-112.0) 0.590 Blood loss (ml) 210.6(42.8-364.9) 214.7 (82.2-444.5) 0.887 Length of stay (day) 6.6(5.0-10.0) 6.8(5.0-11.0) 0.653 Length of incision (cm) 9.9(9.1-11.3) 8.2(7.6-9.5) <0.001 Presented as the mean (ranges). Abbreviations: PSI, patient-specific instrumentation. min, minute. Learning Curve Tibial coronal deviation differed among the 3 PSI case-order subgroups ( P =0.034). No significant intergroup differences were observed for the remaining parameters (Table IV). Table IV Learning Curve in PSI-Assisted UKA. Case 1-8 Case 9-16 Case 17-24 P Operation time (min) 81.3(65.0-102.0) 82.0(71.0-110.0) 81.0(70.0-90.0) 0.949 Blood loss (ml) 267.0(131.0-364.9) 167.9(42.7-282.7) 197.1(48.3-337.7) 0.108 Femoral coronal deviation (°) 2.1(0.8-2.9) 2.0(0.5-3.8) 2.1(0.9-4.0) 0.797 Femoral sagittal deviation (°) 2.8(0.7-5.5) 2.8(0.8-6.1) 4.7(2.0-6.4) 0.055 Femoral axial deviation (°) 3.4(1.3-4.8) 2.7(0.8-5.4) 3.7(0.9-5.7) 0.438 Tibial coronal deviation (°) 1.3(0.4-2.3) a 1.1(0.3-2.2) a 2.1(0.6-3.0) b 0.034 * Tibial sagittal deviation (°) 1.7(0.6-2.5) 1.5(0.3-2.3) 1.5(0.5-2.1) 0.708 Tibial axial deviation (°) 1.8(0.8-3.6) 3.1(1.3-4.0) 2.2(0.6-3.3) 0.084 Tibial prosthesis coverage rate (%) 93.4(89.4-96.2) 93.8(92.0-96.4) 94.2(89.3-98.9) 0.785 Femoral distal osteotomy deviation(mm) 0.7(0.4-1.0) 0.8(0.5-1.2) 0.5(0.4-0.7) 0.068 Femoral condyle osteotomy deviation (mm) 1.0(0.8-1.3) 1.1(0.7-1.7) 1.2(0.5-1.7) 0.489 Tibial osteotomy deviation (mm) 0.5(0.4-0.9) 0.7(0.5-1.3) 0.5(0.3-0.9) 0.112 Length of incision (cm) 9.8(9.1-10.9) 10.1(9.2-11.3) 9.9(9.6-10.7) 0.449 Presented as the mean (ranges). Abbreviations: min, minute. * p <0.05, Post-hoc analysis was conducted, the difference between case 1-8 and case 17-24 was significant ( p= 0.045). Post-hoc analysis was conducted, the difference between case 9-16 and case 17-24 was significant ( p= 0.014). Prosthesis Size Planning Accuracy No malposition, overhang, or obvious protrusion beyond the bony surface was observed in either group. For femoral components, exact-match accuracy was 93.8% for the AI planning system and 58.3% for conventional templating ( P < 0.001). Accuracy within 1 size was 95.8% and 85.4%, respectively, and accuracy within 2 sizes was 100% and 95.8%, respectively. For tibial components, exact-match accuracy was 95.8% for the AI planning system and 47.9% for conventional templating ( P < 0.001). Accuracy within 1 size was 100% and 75.0%, respectively, and accuracy within 2 sizes was 100% and 93.8%, respectively (Table V). Table V Comparison of predicted prosthesis size and implanted prosthesis size. AI (n=48) Conventional templating (n=48) P Femoral prosthesis size (n, %) Same 45 (93.8) 28 (58.3) <0.001 ±1 46 (95.8) 41(85.4) 0.159 ±2 48 (100) 46 (95.8) 0.495 Tibial prosthesis size (n, %) Same 46 (95.8) 23 (47.9) <0.001 ±1 48 (100) 36 (75.0) <0.001 ±2 48 (100) 45 (93.8) 0.242 Abbreviations: AI, Artificial intelligence. Discussion The most important finding of this study is that the newly developed AI-assisted planning and PSI system significantly improved component positioning accuracy in UKA, particularly for the tibial component, with performance that appeared comparable to the range previously reported for robot-assisted UKA. It is possible that AI-based planning contributed to surgical execution in both groups; however, this study was not designed to isolate the independent effect of AI-based planning. In addition, the C–T module demonstrated a clear advantage in image segmentation performance, and the AI planning system accurately predicted prosthesis size. Image accuracy, the engineer's design concept, and the surgeon's experience are all factors influencing the efficacy of PSI, leading to variability in its clinical outcomes( 11 , 22 ). First and foremost, precise image segmentation and recognition algorithms are crucial for accurate preoperative bone model reconstruction. Both CT- and MRI-based imaging workflows can be used to generate PSI, although each modality has distinct trade-offs in preoperative planning and guide production( 23 , 24 ). By using full-length lower limb CT scans combined with X-ray data, we can accurately capture the complex structural morphology of the femorotibial joint and effectively avoid compensatory errors arising from localized imaging data. The CNN-based image architecture U-Net, integrated with the deep learning-based C-T Module, utilizes a hybrid structure that performs global complementary feature perception and enhances detail retention through parallel mixed feature extraction and fusion modules. By continuously interacting with the original input features through element-wise addition and information exchange, the system achieves collaborative optimization of local accuracy and global perception, ensuring high precision and efficiency in information output. Moreover, the innovative design concept improved the conformity between the cutting guide and the bony surface. In this PSI system, three prominent osteophytes on the femoral and tibial sides—anteromedial, anterolateral, and posterolateral—were selected as anchoring zones. These volumetric landmarks formed a stable triangular structure to enhance guide fixation. Notably, to address the commonly reported issue of prolonged operative time associated with PSI use, we minimized the utilization of femoral-side PSI( 12 , 21 ). Following distal femoral resection, electrocautery was used to mark the bone through a reserved central slot in the guide. Posterior femoral condyle resection was then performed using conventional instruments, centered on the marked line. This approach avoided the need for reinstallation of the femoral PSI before posterior condylar cutting, thereby reducing operative time. Furthermore, the AI-based planning system accurately predicted the implant size, facilitating intraoperative trialing and effectively mitigating the issue of prolonged operative time commonly associated with conventional PSI systems. Interestingly, we found that preoperative AI-assisted planning alone improved surgical accuracy. This may be attributed to the increased availability of predictable preoperative information provided to the surgeon through the planning report. The surgeon was able to anticipate critical intraoperative parameters, including expected bone resection thickness, implant size, as well as details such as osteophyte distribution and bone loss. This enhanced preoperative insight facilitated a more efficient and precise translation of the surgical plan into intraoperative execution. The ideal tibial component coverage aims to evenly distribute physiological loads onto the underlying cancellous bone, thereby minimizing stress concentration and preventing complications such as implant subsidence, loosening, and periprosthetic fractures( 25 , 26 ). Compared to total knee arthroplasty (TKA), UKA involves smaller component sizes and coverage, which may lead to higher stress concentration due to the reduced surface area of the implant. Therefore, the goal for orthopedic surgeons is to perform intraoperative bone resection to select the most anatomically appropriate implant size that maximizes tibial plateau bone coverage, while avoiding overhang and soft tissue irritation caused by oversized components. During medial UKA, tibial tray implantation requires a balance between maximizing bone coverage and minimizing component overhang( 27 ). Morphometric or patient-specific tibial designs may improve cortical coverage while reducing overhang and undercoverage( 28 , 29 ). Currently, there are still relatively few studies on the ideal coverage rate of the tibial prosthesis for UKA. The use of our designed PSI tool significantly improved tibial component coverage, with 92% of cases achieving over 90% coverage. Nevertheless, whether this increased tibial component coverage reduces the risk of implant subsidence and loosening remains to be explored further through long-term follow-up. Robotic-assisted UKA require substantial financial investment, are associated with a notable learning curve, and often result in longer operative times( 30 ). We did not observe a broad deterioration in operative efficiency during PSI adoption, although tibial coronal alignment differed across PSI case-order subgroups. This suggests that AI-PSI-assisted UKA may serve as a more efficient and accessible alternative, particularly in settings with limited medical resources. Although our PSI system significantly enhanced surgical accuracy in UKA, the intraoperative accuracy achieved in clinical cases did not match that observed in synthetic bone model validation. This discrepancy underscores several challenges in the practical implementation of PSI, including interindividual anatomical variations, interference from intraoperative soft tissues, and the limitations of imaging-based planning in accurately replicating cartilage damage and actual bone quality. Moreover, the use of PSI often necessitates a longer surgical incision to ensure adequate exposure of the bony surface and accurate placement of the instrumentation, which may represent an inherent limitation of this technique. This study has several limitations. The focus was primarily on the development of a novel AI-assisted planning system and its effectiveness in improving surgical accuracy. Whether this approach translates into improved long-term clinical outcomes or implant survivorship in UKA remains unclear and will be the subject of our future investigations. Additionally, all procedures were performed by a single experienced UKA surgeon, which may limit generalizability to other surgeons and practice settings. As such, the applicability and generalizability of the findings to less experienced surgeons remain uncertain and should be evaluated in future targeted investigations. Conclusion The AI-assisted planning system accurately predicted implant size and, when combined with PSI, improved technical accuracy in medial UKA. Larger studies with longer follow-up are needed to determine whether these technical gains translate into superior clinical outcomes or implant survivorship. Abbreviations AI artificial intelligence CT computed tomography UKA unicompartmental knee arthroplasty PSI patient-specific instrumentation 3D three-dimensional CNN convolutional neural network C-T Module CNN-Transformer module DSC Dice similarity coefficient IoU intersection over union HD95 95th percentile Hausdorff distance ASA American Society of Anesthesiologists ICC intraclass correlation coefficient RMSE root mean square error MRI magnetic resonance imaging TKA total knee arthroplasty. Declarations Ethics approval statement This study was conducted in accordance with the Declaration of Helsinki. Informed consent was obtained from the volunteer. Ethical approval was obtained from the institutional review board of the First Hospital of Hebei Medical University (approval number: 2025-YanShen00212). The study was registered in the China Medical Research Registration and Filing Information System. Consent for publication Not applicable. Availability of data and materials The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. Competing interests Xingyu Liu and Yafang Zhang are affiliated with Longwood Valley Medical Technology company. The company collaborated in this study and was involved in the artificial intelligence-related technical development and manufacture of the 3-dimensional printed patient-specific instrumentation used in this work. The other authors declare no competing interests. Funding This research was funded by the National Key Research and Development Program (grant numbers 2023YFC3604905) and the Science Fund for Distinguished Young Scholars of Hebei Province (Grants number: ZF2024132 and ZF2024143). Authors' contributions Dehua Liu: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Validation, Writing – original draft. Gang Ji: Conceptualization, Data curation, Formal analysis, Methodology, Validation, Writing – original draft. Zirun Gao: Investigation, Writing – original draft . Ye Huang: Validation, Writing – original draft. Guanglei Cao: Methodology, Writing – original draft. Yafang Zhang: Data curation, Writing – original draft. Xingyu Liu: Project administration, Supervision, Investigation, Validation, Data curation, Writing – review & editing. Guobin Liu: Project administration, Funding acquisition, Supervision, Resources, Writing – review & editing. Acknowledgements We would like to express our gratitude to Tianyu Zhang, Zhao Gu, Rui Yu, Jinlong Niu, Xue Wang, Zhengyu Xu and others for their contributions to this study. References Kugelman DN, Wu KA, Goel RK, Dilbone ES, Ryan SP, Bolognesi MP, et al. Comparing Functional Recovery Between Total and Unicompartmental Knee Arthroplasty: A Prospective Health Kit Study. J Arthroplasty. 2025;40(7S1):S84–7. Arthur LW, Jenkins C, Dodd CAF, Price AJ, Jackson WFM, Bottomley N, et al. Mid-term outcomes of the fixed-bearing lateral Oxford unicompartmental knee arthroplasty. Bone Joint J. 2025;107–B(4):432–9. Tay ML, Matthews BG, Monk AP, Young SW. Disease progression, aseptic loosening and bearing dislocations are the main revision indications after lateral unicompartmental knee arthroplasty: a systematic review. J ISAKOS. 2022;7(5):132–41. Thoreau L, Morcillo Marfil D, Thienpont E. Periprosthetic fractures after medial unicompartmental knee arthroplasty: a narrative review. Arch Orthop Trauma Surg. 2021;142(8):2039–48. Wang F, Zhao J, He W, He H, Wang Q. Biomechanical analysis of femoral component malalignment in medial unicompartmental knee arthroplasty. Sci Rep. 2025;15(1):44071. de Ten Noever GV, Vossen RJM, Bayoumi T, Sierevelt IN, Burger JA, Pearle AD, et al. Distinct age-related modes of failure in cemented and cementless Oxford medial unicompartmental knee arthroplasty: results from 25,762 patients in the Dutch Arthroplasty Register. Bone Joint J. 2025;107–B(3):329–36. Zhang J, Ng N, Scott CEH, Blyth MJG, Haddad FS, Macpherson GJ, et al. Robotic arm-assisted versus manual unicompartmental knee arthroplasty: a systematic review and meta-analysis of the MAKO robotic system. Bone Joint J. 2022;104–B(5):541–8. Bell SW, Anthony I, Jones B, MacLean A, Rowe P, Blyth M. Improved Accuracy of Component Positioning with Robotic-Assisted Unicompartmental Knee Arthroplasty: Data from a Prospective, Randomized Controlled Study. J Bone Joint Surg Am. 2016;98(8):627–35. Goh GS, Haffar A, Tarabichi S, Courtney PM, Krueger CA, Lonner JH. Robotic-Assisted Versus Manual Unicompartmental Knee Arthroplasty: A Time-Driven Activity-Based Cost Analysis. J Arthroplasty. 2022;37(6):1023–8. Tay ML, Carter M, Bolam SM, Zeng N, Young SW. Robotic-arm assisted unicompartmental knee arthroplasty system has a learning curve of 11 cases and increased operating time. Knee Surg Sports Traumatol Arthrosc. 2022;31(3):793–802. Cao G, Du M, Li Z, An S, Huang J, Liu S, et al. Novel patient-specific instrument with comparable accuracy to robotic assistance in medial unicompartmental knee arthroplasty: a prospective study. Int J Surg. 2025;111(7):4487–94. Leenders AM, Kort NP, Koenraadt KLM, van Geenen RCI, Most J, Kerens B, et al. Patient-specific instruments do not show advantage over conventional instruments in unicompartmental knee arthroplasty at 2 year follow-up: a prospective, two-centre, randomised, double-blind, controlled trial. Knee Surg Sports Traumatol Arthrosc. 2021;30(3):918–27. Hafez MA, Moholkar K. Patient-specific instruments: advantages and pitfalls. SICOT J. 2017;3:66. Deng Y, Bai X, Zhao Z, Cao L, Liu Y, Jiang Q. Patient-specific vs. Oxford microplasty instrumentation in unicompartmental knee arthroplasty: a randomized controlled trial. Eur J Med Res. 2025;30(1):1238. Kerens B, Leenders AM, Schotanus MGM, Boonen B, Tuinebreijer WE, Emans PJ, et al. Patient-specific instrumentation in Oxford unicompartmental knee arthroplasty is reliable and accurate except for the tibial rotation. Knee Surg Sports Traumatol Arthrosc. 2017;26(6):1823–30. Rodriguez HC, Rust BD, Roche MW, Gupta A. Artificial intelligence and machine learning in knee arthroplasty. Knee. 2025;54:28–49. Schwarz GM, Simon S, Mitterer JA, Frank BJH, Aichmair A, Dominkus M, et al. Artificial intelligence enables reliable and standardized measurements of implant alignment in long leg radiographs with total knee arthroplasties. Knee Surg Sports Traumatol Arthrosc. 2022;30(8):2538–47. Han K, Wang Y, Chen H, Chen X, Guo J, Liu Z et al. A Survey on Vision Transformer. IEEE Trans Pattern Anal Mach Intell. 2022;45(1). Ourselin S, Sabuncu MR, Wells W, Joskowicz L, Unal G, Maier A. The 19th International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI 2016). Med Image Anal. 2017;41:1. Park JH, Rasouli MR, Mortazavi SMJ, Tokarski AT, Maltenfort MG, Parvizi J. Predictors of perioperative blood loss in total joint arthroplasty. J Bone Joint Surg Am. 2013;95(19):1777–83. Alvand A, Khan T, Jenkins C, Rees JL, Jackson WF, Dodd CAF, et al. The impact of patient-specific instrumentation on unicompartmental knee arthroplasty: a prospective randomised controlled study. Knee Surg Sports Traumatol Arthrosc. 2017;26(6):1662–70. van Leeuwen JAMJ, Röhrl SM. Patient-specific positioning guides do not consistently achieve the planned implant position in UKA. Knee Surg Sports Traumatol Arthrosc. 2016;25(3):752–8. Volpi P, Prospero E, Bait C, Cervellin M, Quaglia A, Redaelli A, et al. High accuracy in knee alignment and implant placement in unicompartmental medial knee replacement when using patient-specific instrumentation. Knee Surg Sports Traumatol Arthrosc. 2013;23(5):1292–8. Thijs E, Theeuwen D, Boonen B, van Haaren E, Hendrickx R, Vos R, et al. Comparable clinical outcome and implant longevity after CT- or MRI-based patient-specific instruments for total knee arthroplasty: a 2-year follow-up of a RCT. Knee Surg Sports Traumatol Arthrosc. 2019;28(6):1821–6. Makaram NS, Yapp LZ, Bowley ALW, Garner A, Scott CEH. Polyethylene wear in metal-backed tibial components in unicompartmental knee prostheses. J ISAKOS. 2024;9(6):100324. Kazarian GS, Barrack TN, Okafor L, Barrack RL, Nunley RM, Lawrie CM. High Prevalence of Radiographic Outliers and Revisions with Unicompartmental Knee Arthroplasty. J Bone Joint Surg Am. 2020;102(13):1151–9. Escudier J-C, Jacquet C, Flecher X, Parratte S, Ollivier M, Argenson J-N. Better Implant Positioning and Clinical Outcomes With a Morphometric Unicompartmental Knee Arthroplasty. Results of a Retrospective, Matched-Controlled Study. J Arthroplasty. 2019;34(12):2903–8. Miyake Y, Namba Y, Mitani S, Umehara N, Kawamoto T, Furuichi S. Comparison of tibial implant positioning between symmetrical and anatomical design implants in unicompartmental knee arthroplasty for Japanese patients. J Orthop Surg (Hong Kong). 2023;31(1):10225536221149485. Carpenter DP, Holmberg RR, Quartulli MJ, Barnes CL. Tibial plateau coverage in UKA: a comparison of patient specific and off-the-shelf implants. J Arthroplasty. 2014;29(9):1694–8. Goh GS, Haffar A, Tarabichi S, Courtney PM, Krueger CA, Lonner JH. Robotic-Assisted Versus Manual Unicompartmental Knee Arthroplasty: A Time-Driven Activity-Based Cost Analysis. J Arthroplasty. 2022;37(6):1023–8. Additional Declarations No competing interests reported. Supplementary Files Supplementalmaterial.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 05 May, 2026 Reviewers agreed at journal 01 May, 2026 Reviewers agreed at journal 30 Apr, 2026 Reviewers invited by journal 29 Apr, 2026 Submission checks completed at journal 23 Apr, 2026 First submitted to journal 20 Apr, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9340442","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":635099959,"identity":"56f9dacc-5f90-4be0-aa93-6d69a727ea3a","order_by":0,"name":"Dehua Liu","email":"","orcid":"","institution":"First Affiliated Hospital of Hebei Medical University","correspondingAuthor":false,"prefix":"","firstName":"Dehua","middleName":"","lastName":"Liu","suffix":""},{"id":635099962,"identity":"79cbcbb5-5837-4b54-a3a0-d4a2d9ce2dd9","order_by":1,"name":"Gang Ji","email":"","orcid":"","institution":"First Affiliated Hospital of Hebei Medical University","correspondingAuthor":false,"prefix":"","firstName":"Gang","middleName":"","lastName":"Ji","suffix":""},{"id":635099963,"identity":"8c97d528-09f3-4d2f-8060-8aa61efea9d5","order_by":2,"name":"Zirun Gao","email":"","orcid":"","institution":"First Affiliated Hospital of Hebei Medical University","correspondingAuthor":false,"prefix":"","firstName":"Zirun","middleName":"","lastName":"Gao","suffix":""},{"id":635099964,"identity":"bafff45d-0665-408f-a341-a67cca8fa1e7","order_by":3,"name":"Ye Huang","email":"","orcid":"","institution":"Beijing Jishuitan Hospital","correspondingAuthor":false,"prefix":"","firstName":"Ye","middleName":"","lastName":"Huang","suffix":""},{"id":635099965,"identity":"313f72e0-f153-4694-bafc-af6eba50564b","order_by":4,"name":"Guanglei Cao","email":"","orcid":"","institution":"Beijing Jishuitan Hospital","correspondingAuthor":false,"prefix":"","firstName":"Guanglei","middleName":"","lastName":"Cao","suffix":""},{"id":635099966,"identity":"52d3fa11-9e7c-4ce0-8965-a7f97bad205d","order_by":5,"name":"Yafang Zhang","email":"","orcid":"","institution":"Longwood Valley Medical Technology Co Ltd","correspondingAuthor":false,"prefix":"","firstName":"Yafang","middleName":"","lastName":"Zhang","suffix":""},{"id":635099967,"identity":"dc5d0184-e4bd-4acf-b542-c4a8287dc7b9","order_by":6,"name":"Xingyu Liu","email":"","orcid":"","institution":"Tsinghua University","correspondingAuthor":false,"prefix":"","firstName":"Xingyu","middleName":"","lastName":"Liu","suffix":""},{"id":635099968,"identity":"283b9ce6-4768-4eb5-872c-6d0d2a2fd009","order_by":7,"name":"Guobin Liu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4klEQVRIiWNgGAWjYBACPmYwdYCHgb2HAcJmSMCvhQ2uhecMsVog1AEGBokcYrWw8xi/5qm4I2Mu+fbg48Kcwwz87DkGDD934HMYj5k1z5lnPJaz85KNZ247zCDZ88aAsfcMfi3GvG2HeQxu55hJ8wK1GNzIMWBmbCNGy80z5r9BWuyJ0GL8GKzlBo8ZM9gWCYJa2MoY5wD9YnAmx1h65rZ0HokzzwoO9uLRws9/ePOHNxV37A2OnzH8XLjNWo6/PXnjg594tIAskkDm8YCIA3g1MDAwfyCgYBSMglEwCkY6AAAjB0kHeRqfhAAAAABJRU5ErkJggg==","orcid":"","institution":"First Affiliated Hospital of Hebei Medical University","correspondingAuthor":true,"prefix":"","firstName":"Guobin","middleName":"","lastName":"Liu","suffix":""}],"badges":[],"createdAt":"2026-04-07 06:26:48","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9340442/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9340442/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":108946310,"identity":"1ec9b760-42bc-42c5-b589-cac1e47e4ace","added_by":"auto","created_at":"2026-05-11 06:22:08","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":2157537,"visible":true,"origin":"","legend":"\u003cp\u003eSchematic Diagrams of Image Segmentation and Recognition Algorithms. (A) Schematic diagram of the image segmentation algorithm. (B) Schematic diagram of the landmark detection algorithm. Note: Conv 3D(3D Convolutional Neural Network Layer), BN(Batch Normalization), GeLU(Gaussian Error Linear Unit), C-T Module(U-Net–based convolutional neural network with a Transformer-based deep learning module), MSA(Masked Separable Attention), LN(Layer Normalization), MLP(Multilayer Perceptron), Conv 2D(2D Convolutional Neural Network Layer), ReLU（Rectified Linear Unit).\u003c/p\u003e","description":"","filename":"Fig.1.png","url":"https://assets-eu.researchsquare.com/files/rs-9340442/v1/848bab62b06b7e57283a3102.png"},{"id":108977871,"identity":"3dee3fdf-052e-426b-929a-8735157c6cd9","added_by":"auto","created_at":"2026-05-11 11:33:17","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":4319930,"visible":true,"origin":"","legend":"\u003cp\u003eDesigning process of patient-specific instrument (PSI) of UKA. (A) Preoperative planning of CT images based on artificial intelligence. (B) Printing of PSI. (C) The synthetic bone validation of PSI. (D) Intraoperative application scenario of PSI.\u003c/p\u003e","description":"","filename":"Fig.2.png","url":"https://assets-eu.researchsquare.com/files/rs-9340442/v1/f3683be2c79337a69dbc6c90.png"},{"id":108978170,"identity":"8cafab75-847e-4653-b463-b107206c48cd","added_by":"auto","created_at":"2026-05-11 11:34:32","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":627461,"visible":true,"origin":"","legend":"\u003cp\u003eSchematic diagram of the prospective clinical research\u003c/p\u003e","description":"","filename":"Fig.3.png","url":"https://assets-eu.researchsquare.com/files/rs-9340442/v1/98df603d8d7ee67d3626f413.png"},{"id":109081341,"identity":"c8a1e8be-ea32-43d8-a8d5-99717b77d702","added_by":"auto","created_at":"2026-05-12 12:17:10","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":500109,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of Image Segmentation Accuracy. (A) Direct visual comparison of image segmentation performance. (B) Comparison of segmentation accuracy metrics.\u003c/p\u003e","description":"","filename":"Fig.4.png","url":"https://assets-eu.researchsquare.com/files/rs-9340442/v1/0624aef98f9f87f4c2e9915f.png"},{"id":108977806,"identity":"db528da4-080e-46d3-ab31-83ab4cacb0e6","added_by":"auto","created_at":"2026-05-11 11:33:00","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":513610,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of surgical precision. (A) Deviation of femoral prosthesis. (B) Deviation of tibial prosthesis. (C) Deviation of osteotomy. (D) Tibial prosthesis coverage rate. (E) Comparation of RMSE values of component positioning angles for the current study, as well as the robot-assisted UKA by Bell et al. Note: RMSE(root mean square). Data are shown as mean ± SD. *\u003cem\u003eP\u003c/em\u003e\u0026lt; 0.05, **\u003cem\u003e P\u003c/em\u003e \u0026lt; 0.01, ***\u003cem\u003e P\u003c/em\u003e \u0026lt; 0.001, ns = not significant.\u003c/p\u003e","description":"","filename":"Fig.5.png","url":"https://assets-eu.researchsquare.com/files/rs-9340442/v1/72b9571cc5c6e061493512cf.png"},{"id":109252373,"identity":"492b6c41-f214-441c-a710-734aebb1055b","added_by":"auto","created_at":"2026-05-14 09:25:31","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":8053914,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9340442/v1/5c69bbf9-db34-4a08-bbad-980d929d25bf.pdf"},{"id":108977737,"identity":"6f67fa04-2fd6-4e1c-9f3c-c07acc72e34c","added_by":"auto","created_at":"2026-05-11 11:32:45","extension":"docx","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":360371,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementalmaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-9340442/v1/630a4ece388864d266b33818.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Artificial Intelligence-Assisted Preoperative Planning Combined with Patient-Specific Instrumentation Improves the Accuracy of Unicompartmental Knee Arthroplasty","fulltext":[{"header":"Introduction","content":"\u003cp\u003eUnicompartmental knee arthroplasty (UKA) is an effective treatment for end-stage unicompartmental osteoarthritis of the knee, offering advantages such as minimally invasive surgery, rapid rehabilitation, and favorable postoperative proprioception(\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e). Numerous studies have shown that long-term success of UKA relies on refined surgical techniques. Optimal prosthetic component positioning, meticulous soft tissue balancing, and appropriate implant sizing are fundamental to achieving satisfactory clinical outcomes(\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e). Suboptimal component positioning and sizing have been associated with accelerated wear, instability, bearing dislocation, subsidence, and persistent postoperative pain(\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e). Therefore, precise preoperative planning and intraoperative execution are paramount to preempting surgical unpredictability, ensuring accurate component placement, and minimizing associated logistical and sterilization costs.\u003c/p\u003e \u003cp\u003eRobotic and navigation-assisted techniques have been reported to improve surgical accuracy of UKA(\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e). However, their widespread adoption has been limited by prolonged operative time, high equipment acquisition and maintenance costs, and steep learning curves\u0026mdash;particularly in lower-resource healthcare settings(\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e). In recent years, increasing attention has been directed toward the role of patient-specific instrumentation (PSI) in enhancing UKA precision(\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). PSI has attracted interest because of its conceptual simplicity, lower infrastructure demands, and potential workflow advantages\u003c/p\u003e \u003cp\u003eHowever, traditional PSI designs are complex and require repeated validation of bone surface conformity across multiple anatomical planes, often resulting in production cycles lasting several days(\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e). Moreover, the impact of PSI on surgical accuracy in UKA remains controversial(\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e). While some studies have demonstrated that PSI significantly improves the precision of component placement, others have reported inconsistent outcomes or even detrimental effects on surgical precision(\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e). In recent years, the integration of artificial intelligence (AI) with three-dimensional (3D) printing technologies has shown promising potential in clinical applications(\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e). Leveraging AI\u0026rsquo;s efficiency in image segmentation and large-scale data processing may enhance both the accuracy and production efficiency of PSI systems.\u003c/p\u003e \u003cp\u003eWe therefore developed an AI-assisted preoperative planning workflow combined with PSI and evaluated whether this approach improved component positioning accuracy, resection accuracy, and implant-size prediction in medial UKA.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eAI-Based Image Processing and Segmentation Workflow\u003c/h2\u003e \u003cp\u003eThe AI plan system integrates a U-Net\u0026ndash;based convolutional neural network (CNN) with a Transformer-based deep learning module (C-T Module) for image segmentation and landmark recognition. The segmentation pipeline includes a residual 3D CNN encoder for local feature extraction and a decoder combining CNN and Transformer branches for global context and fine detail enhancement. Outputs are fused via element-wise addition to optimize segmentation precision (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e-A). Compared with a deep learning-based 3D U-Net model, segmentation performance was evaluated using Dice Similarity Coefficient (DSC), Intersection over Union (IoU), and 95th Percentile Hausdorff Distance (HD95)(\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e). For landmark recognition, a low-resolution sub-network operates in parallel with the high-resolution backbone, enabling multi-scale feature fusion through staged interactions, with final outputs generated by the backbone network (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e-B).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003ePreoperative Planning\u003c/h3\u003e\n\u003cp\u003eThe system generated automated 3-dimensional preoperative plans using full-length lower extremity CT scans and standardized radiographs. Stress radiographs were additionally reviewed to assess ligament function. Tibial component planning included resection level, posterior slope, coronal and rotational alignment, with adaptive adjustments for implant coverage. Femoral planning follows, and the system simulates postoperative implant\u0026ndash;bone location, implant sizes, and resection volumes, with final confirmation by the surgeon (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e-A). PSI was designed based on individual bony anatomy and osteophytes to maximize fit and minimize exposure. Tibial and femoral guide bone cuts via osteotomy slots, with depth control and distal marking aiding posterior femoral resections (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e-B). The PSI was manufactured in nylon by 3-dimensional printing, and the complete design-to-fabrication workflow was completed within 12 hours (Figs.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e-C, D).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eThe Feasibility Verification of PSI\u003c/h3\u003e\n\u003cp\u003eSix synthetic bone models were utilized to validate the precision of the AI-assisted planning system. A 3D scanner (EinScan Pro 2X V2) was employed to digitize the bone models, and the resulting data were imported into the planning system to generate a surgical plan and fabricate PSI. After performing the surgery through PSI, the prosthesis was installed. Postoperative 3D scanning of the bone and prosthetic position was performed, and the data were archived. After model reconstruction, the discrepancy between the actual prosthetic position and the preoperative plan was measured.\u003c/p\u003e\n\u003ch3\u003eClinical Research Design\u003c/h3\u003e\n\u003cp\u003e This prospective randomized study was approved by the institutional review board (approval number: 2025-YanShen00212). The study was registered in the China Medical Research Registration and Filing Information System. From January 2nd 2025 to August 15th 2025, 53 patients undergoing UKA were enrolled. Inclusion required standard UKA indications, completion of preoperative planning, and informed consent. Exclusion criteria included contraindications, enrollment in other studies, or refusal to participate. Patients were randomized in a 1:1 ratio using a computer-generated randomization sequence. Allocation concealment was achieved using sequentially numbered, opaque, sealed envelopes, which were opened only after patient enrollment. Patients were assigned to either the PSI group (AI-based planning plus PSI-assisted UKA) or the control group (AI-based planning plus conventionally instrumented UKA). The primary outcome was postoperative tibial coronal alignment deviation (varus-valgus angle) from the preoperative plan. Secondary outcomes included femoral and tibial component alignment in the remaining planes, tibial prosthesis coverage, osteotomy deviation, perioperative outcomes, and implant-size prediction accuracy. Because of the nature of the intervention, the operating surgeon could not be blinded to group allocation. Participants were blinded to group allocation. Radiographic measurements and implant-size evaluations were performed by independent assessors blinded to patient information and group allocation. The study flow diagram is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eDemographic Characteristics\u003c/h3\u003e\n\u003cp\u003eThere were no statistically significant differences between the PSI group and the control group in demographic variables, preoperative hemoglobin level, hematocrit, or ASA classification (Table Sup I).\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eSurgical Technique\u003c/h2\u003e \u003cp\u003e3-dimensional preoperative planning was performed for all patients in in both groups. All patients underwent medial UKA performed by a single surgeon using a fixed-bearing prosthesis. All procedures were performed by a single high-volume UKA surgeon who had no prior clinical experience with this specific PSI workflow. Procedures were performed with a tourniquet via an anteromedial parapatellar approach. After soft tissue release, tibial and femoral bone surfaces were exposed and cartilage in the region of the guide attachment was removed. Guides were fixed with pins, followed by vertical/horizontal tibial osteotomies and femoral distal resection. Using an electrocautery device, a line was marked along the pre-designed femoral notch on the guide, serving as a reference for posterior femoral condyle resection using the guide. Subsequent steps followed the standard conventional UKA procedure. Resected bone blocks were collected to compare with preoperative plans. No drainage was used postoperatively, and routine antibiotic and anticoagulation therapy were administered. Uniform perioperative care and surgical education were provided to all patients.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eRadiological Outcomes\u003c/h3\u003e\n\u003cp\u003ePostoperative CT and radiographs were obtained for each patient. CT datasets were processed using Mimics and UG software for model reconstruction. Preoperative and postoperative skeletal models were aligned to assess deviations in prosthesis positions. Measurements were conducted by two independent observers. Inter-observer reliability coefficients (ICC) were calculated, and the mean of the two measurements was used for statistical analysis. Parameters measured included varus/valgus, internal/external rotation, and flexion deviation angles of the femoral component; varus/valgus, internal/external rotation, and posterior tilt deviation angles of the tibial component; and tibial prosthesis coverage rate (Figure Sup1-A to 1-F).\u003c/p\u003e\n\u003ch3\u003ePerioperative Parameters\u003c/h3\u003e\n\u003cp\u003eSurgical duration, perioperative adverse events, estimated blood loss, length of hospital stay and length of incision were recorded for each group. Surgical duration was defined as the time interval from incision to wound closure. Estimated blood loss was calculated according to published methods, based on the change in hematocrit from before to after surgery and the patient\u0026rsquo;s body weight(\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e).\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eLearning Curve\u003c/h2\u003e \u003cp\u003eTo assess the learning curve associated with PSI-assisted UKA, PSI patients were stratified into three cohorts based on case sequence: Group A (cases 1\u0026ndash;8), Group B (cases 9\u0026ndash;16), and Group C (cases 17\u0026ndash;24). Comparative analyses were conducted across these groups focusing on operative duration, intraoperative blood loss, prosthetic alignment deviations, and tibial prosthesis coverage rate.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eComparison of Implant Size Planning Accuracy\u003c/h2\u003e \u003cp\u003eTwo independent observers, blinded to patient information, estimated the implant sizes based on preoperative radiographs using standardized sizing templates provided by the implant manufacturer. Prior to evaluation, radiographic magnification was calibrated uniformly using the templates. All estimations were reviewed by an additional senior orthopedic surgeon. The predicted implant sizes and those generated by the AI\u0026ndash;based planning system were each compared with the actual implant sizes.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eSample Size Calculation\u003c/h2\u003e \u003cp\u003eThe sample size was estimated based on the primary outcome of postoperative tibial coronal alignment (varus-valgus angle) after UKA. According to the study by Alvand et al., which compared UKA performed with patient-specific instrumentation and conventional instrumentation using similar radiographic parameters, the assumed standard deviation of tibial coronal alignment in the control group was 3.6\u0026deg;. A between-group difference of 3.0\u0026deg; was considered clinically relevant(\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e). Using a 2-sided alpha of 0.05 and a power of 0.80, the minimum required sample size was 44 patients (22 per group). To account for potential loss to follow-up and to improve study robustness, the planned enrollment was increased to 48 patients (24 per group).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eData Analyses\u003c/h2\u003e \u003cp\u003eAbsolute angular deviations were defined as the absolute difference between the planned and postoperative measurements. Statistical analyses were performed using SPSS version 25.0. Continuous variables were expressed as mean(ranges) and mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation, as appropriate, whereas categorical variables were presented as counts and percentages. For continuous data with normal distribution, depending on the situation, either the independent-samples t-test or one-way analysis of variance (ANOVA) was used to compare the differences between different groups; for non-normally distributed data, nonparametric tests were applied. Categorical data were analyzed using the chi-square test or Fisher\u0026rsquo;s exact test, as appropriate. The significance level was set at α\u0026thinsp;=\u0026thinsp;0.05, and a \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered statistically significant. The root mean square error (RMSE) was calculated to quantify angular deviations and was compared with previously published data on robotic-assisted UKA(\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cem\u003eImage Segmentation Accuracy\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe C-T Module demonstrated superior automatic segmentation accuracy of CT images compared to the 3D U-Net (Figure 4-A). The DCS for the C-T Module was 0.9409, higher than the 3D U-Net\u0026apos;s 0.8914. The IoU for the C-T Module was 0.8884, surpassing the 3D U-Net\u0026apos;s 0.8041. The 95th percentile HD95 for the C-T Module was 14.98, lower than the 3D U-Net\u0026apos;s 21.23 (Figure 4-B).\u003c/p\u003e\n\u003cp\u003eFigure. 4. Comparison of Image Segmentation Accuracy. (A) Direct visual comparison of image segmentation performance. (B) Comparison of segmentation accuracy metrics.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eThe Feasibility Verification of PSI\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe absolute angular deviations for the femoral component in the coronal, sagittal, and axial planes were 1.2 \u0026plusmn; 0.7\u0026deg;, 1.0 \u0026plusmn; 0.4\u0026deg;, and 0.9 \u0026plusmn; 0.3\u0026deg;. The absolute angular deviations for the tibial component in the coronal, sagittal, and axial planes were 1.0 \u0026plusmn; 0.5\u0026deg;, 1.2 \u0026plusmn; 0.6\u0026deg;, and 0.8 \u0026plusmn; 0.5\u0026deg;. The mean tibial prosthesis coverage rate was 97.0 \u0026plusmn; 0.7%. Except for the femoral coronal deviation, the absolute angular deviations in all other planes were within 2\u0026deg; (Table I).\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" style=\"width: 100px;\"\u003e\n \u003cp\u003eTable I Deviation of prosthesis placement of bone model.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 67px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32px;\"\u003e\n \u003cp\u003ePSI\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 67px;\"\u003e\n \u003cp\u003eFemoral coronal deviation (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32px;\"\u003e\n \u003cp\u003e1.2(0.3-2.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 67px;\"\u003e\n \u003cp\u003e\u0026le;2\u0026deg;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32px;\"\u003e\n \u003cp\u003e5(83.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 67px;\"\u003e\n \u003cp\u003eFemoral sagittal deviation (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32px;\"\u003e\n \u003cp\u003e1.0(0.6-1.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 67px;\"\u003e\n \u003cp\u003e\u0026le;2\u0026deg;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32px;\"\u003e\n \u003cp\u003e6(100%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 67px;\"\u003e\n \u003cp\u003eFemoral axial deviation (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32px;\"\u003e\n \u003cp\u003e0.9(0.6-1.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 67px;\"\u003e\n \u003cp\u003e\u0026le;2\u0026deg;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32px;\"\u003e\n \u003cp\u003e6(100%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 67px;\"\u003e\n \u003cp\u003eTibial coronal deviation (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32px;\"\u003e\n \u003cp\u003e1.0(0.5-1.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 67px;\"\u003e\n \u003cp\u003e\u0026le;2\u0026deg;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32px;\"\u003e\n \u003cp\u003e6(100%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 67px;\"\u003e\n \u003cp\u003eTibial sagittal deviation (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32px;\"\u003e\n \u003cp\u003e1.2(0.5-1.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 67px;\"\u003e\n \u003cp\u003e\u0026le;2\u0026deg;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32px;\"\u003e\n \u003cp\u003e6(100%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 67px;\"\u003e\n \u003cp\u003eTibial axial deviation (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32px;\"\u003e\n \u003cp\u003e0.8(0.1-1.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 67px;\"\u003e\n \u003cp\u003e\u0026le;2\u0026deg;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32px;\"\u003e\n \u003cp\u003e6(100%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 67px;\"\u003e\n \u003cp\u003eTibial prosthesis coverage rate (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32px;\"\u003e\n \u003cp\u003e97.0(95.8-97.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 67px;\"\u003e\n \u003cp\u003e\u0026ge;90%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 32px;\"\u003e\n \u003cp\u003e6(100%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" style=\"width: 100px;\"\u003e\n \u003cp\u003ePresented as the mean (ranges) and number (percentage).\u0026nbsp;Abbreviations: PSI,\u0026nbsp;patient-specific instrumentation.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cem\u003eSurgical Precision\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eAll ICCs for angular measurements exceeded 0.9. The PSI group demonstrated significantly lower absolute errors than the control group for femoral coronal (2.0 \u0026plusmn; 1.1\u0026deg; vs. 4.4 \u0026plusmn; 2.6\u0026deg;,\u003cem\u003e\u0026nbsp;P\u0026nbsp;\u003c/em\u003e=0.003), sagittal (3.4 \u0026plusmn; 1.8\u0026deg; vs. 5.5\u0026plusmn; 2.8\u0026deg;, \u003cem\u003eP\u0026nbsp;\u003c/em\u003e=0.005), and axial deviations (3.2 \u0026plusmn; 1.6\u0026deg; vs. 4.6 \u0026plusmn; 2.1\u0026deg;, \u003cem\u003eP\u0026nbsp;\u003c/em\u003e=0.008). No significant difference was observed in the proportion of femoral components within 2\u0026deg; between groups. Tibial component positioning was also more accurate in the PSI group across coronal (1.5 \u0026plusmn; 0.8\u0026deg; vs. 2.9 \u0026plusmn; 1.1\u0026deg;), sagittal (1.5 \u0026plusmn; 0.6\u0026deg; vs. 3.1 \u0026plusmn; 1.5\u0026deg;), and axial planes (2.4 \u0026plusmn;1.2\u0026deg; vs. 6.7 \u0026plusmn; 2.6\u0026deg;), all \u003cem\u003eP\u0026nbsp;\u003c/em\u003e\u0026lt;0.001. A greater proportion of tibial components in the PSI group were positioned within 2\u0026deg; of the plans in all 3 planes (\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05) Tibial prosthesis coverage was higher in the PSI group (93.8% \u0026plusmn; 2.3% vs. 90.3% \u0026plusmn; 3.7%, \u003cem\u003eP\u0026nbsp;\u003c/em\u003e\u0026lt;0.001). The PSI group showed significantly lower proximal tibial osteotomy deviation (0.6 \u0026plusmn; 0.2 mm vs. 1.1 \u0026plusmn; 0.4 mm, \u003cem\u003eP\u0026nbsp;\u003c/em\u003e\u0026lt;0.001), while distal femoral deviation showed no significant difference. A greater proportion of osteotomies with a deviation of \u0026le; 1 mm was observed in the PSI group (\u003cem\u003eP\u0026nbsp;\u003c/em\u003e=0.01). The detailed comparison is presented in Table II. And Figure 5A-D. Furthermore, the outcomes achieved with the PSI in this study were comparable to those reported for robotic-assisted UKA (Figure 5-E).\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" valign=\"top\"\u003e\n \u003cp\u003eTable II Deviation of prosthesis placement and osteotomy amount.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePSI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eControl\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eFemoral coronal deviation (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e2.0(0.5-4.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e4.4(0.3-8.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.003\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026le;2\u0026deg;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e13(54%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e9(38%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.773\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eFemoral sagittal deviation (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e3.4(0.7-6.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e5.5(0.6-9.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.005\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026le;2\u0026deg;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e10(42%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e5(21%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.212\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eFemoral axial deviation (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e3.2(0.8-5.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e4.6(0.5-7.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.008\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026le;2\u0026deg;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e10(42%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e6(25%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.359\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eTibial coronal deviation (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.5(0.3-3.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e2.9(1.1-5.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026le;2\u0026deg;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e18(75%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e8(33%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.008\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eTibial sagittal deviation (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.5(0.3-2.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e3.1(0.6-5.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026le;2\u0026deg;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e20(83%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e9(38%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.003\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eTibial axial deviation (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e2.4(0.6-4.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e6.7(0.9-10.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026le;2\u0026deg;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e11(46%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e3(13%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.024\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eTibial prosthesis coverage rate (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e93.8(89.3-98.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e90.3(82.7-95.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ge;90%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e22(92%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e16(67%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.072\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eFemoral distal\u003c/p\u003e\n \u003cp\u003eosteotomy deviation(mm)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.7(0.4-1.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.0(0.2-2.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.093\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026le;1 mm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e23(96%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e15(63%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.010\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eTibial osteotomy deviation (mm)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.6(0.3-1.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.1(0.7-2.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026le;1 mm\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e23(96%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e15(63%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.010\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" valign=\"top\"\u003e\n \u003cp\u003ePresented as the mean (ranges) and number (percentage).\u0026nbsp;Abbreviations: PSI,\u0026nbsp;patient-specific instrumentation.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eFigure. 5. Comparison of surgical precision. (A) Deviation of femoral prosthesis. (B) Deviation of tibial prosthesis. (C) Deviation of osteotomy. (D) Tibial prosthesis coverage rate. (E) Comparation of RMSE values of component positioning angles for the current study, as well as the robot-assisted UKA by Bell et al. Note: RMSE(root mean square). Data are shown as mean \u0026plusmn; SD. *\u003cem\u003eP\u003c/em\u003e\u0026lt; 0.05, **\u003cem\u003e\u0026nbsp;P\u003c/em\u003e \u0026lt; 0.01, ***\u003cem\u003e\u0026nbsp;P\u003c/em\u003e \u0026lt; 0.001, ns = not significant.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003ePerioperative Parameters\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eOperative time, blood loss, and length of stay did not differ between groups. Incision length of the PSI group (9.9 \u0026plusmn; 0.6 cm) was significantly longer than the control group (8.2\u0026plusmn; 0.5 cm) (\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.001). Detailed data are presented in Table III. No serious surgery-related complications occurred during hospitalization in either group.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" valign=\"top\"\u003e\n \u003cp\u003eTable III Perioperative related indicators.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePSI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eControl\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eOperation time (min)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e81.4(65.0-110.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e84.2(65.0-112.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.590\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBlood loss (ml)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e210.6(42.8-364.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e214.7 (82.2-444.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.887\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLength of stay (day)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e6.6(5.0-10.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e6.8(5.0-11.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.653\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLength of incision (cm)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e9.9(9.1-11.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e8.2(7.6-9.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" valign=\"top\"\u003e\n \u003cp\u003ePresented as the mean (ranges).\u0026nbsp;Abbreviations: PSI,\u0026nbsp;patient-specific instrumentation.\u0026nbsp;min, minute.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cem\u003eLearning Curve\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eTibial coronal deviation differed among the 3 PSI case-order subgroups (\u003cem\u003eP\u0026nbsp;\u003c/em\u003e=0.034). No significant intergroup differences were observed for the remaining parameters (Table IV).\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"5\" valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003eTable IV Learning Curve in PSI-Assisted UKA.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19px;\"\u003e\n \u003cp\u003eCase 1-8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eCase 9-16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eCase 17-24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eOperation time (min)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19px;\"\u003e\n \u003cp\u003e81.3(65.0-102.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e82.0(71.0-110.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e81.0(70.0-90.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.949\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eBlood loss\u0026nbsp;(ml)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19px;\"\u003e\n \u003cp\u003e267.0(131.0-364.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e167.9(42.7-282.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e197.1(48.3-337.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.108\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eFemoral coronal deviation (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19px;\"\u003e\n \u003cp\u003e2.1(0.8-2.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e2.0(0.5-3.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e2.1(0.9-4.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.797\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eFemoral sagittal deviation (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19px;\"\u003e\n \u003cp\u003e2.8(0.7-5.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e2.8(0.8-6.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e4.7(2.0-6.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.055\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eFemoral axial deviation (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19px;\"\u003e\n \u003cp\u003e3.4(1.3-4.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e2.7(0.8-5.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e3.7(0.9-5.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.438\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eTibial coronal deviation (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19px;\"\u003e\n \u003cp\u003e1.3(0.4-2.3) \u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e1.1(0.3-2.2) \u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e2.1(0.6-3.0) \u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.034\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eTibial sagittal deviation (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19px;\"\u003e\n \u003cp\u003e1.7(0.6-2.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e1.5(0.3-2.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e1.5(0.5-2.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.708\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eTibial axial deviation (\u0026deg;)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19px;\"\u003e\n \u003cp\u003e1.8(0.8-3.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e3.1(1.3-4.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e2.2(0.6-3.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.084\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eTibial prosthesis coverage rate (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19px;\"\u003e\n \u003cp\u003e93.4(89.4-96.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e93.8(92.0-96.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e94.2(89.3-98.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.785\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eFemoral distal osteotomy deviation(mm)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19px;\"\u003e\n \u003cp\u003e0.7(0.4-1.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e0.8(0.5-1.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e0.5(0.4-0.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.068\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eFemoral condyle osteotomy deviation (mm)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19px;\"\u003e\n \u003cp\u003e1.0(0.8-1.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e1.1(0.7-1.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e1.2(0.5-1.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.489\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eTibial osteotomy deviation (mm)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19px;\"\u003e\n \u003cp\u003e0.5(0.4-0.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e0.7(0.5-1.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e0.5(0.3-0.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.112\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eLength of incision (cm)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19px;\"\u003e\n \u003cp\u003e9.8(9.1-10.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e10.1(9.2-11.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e9.9(9.6-10.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e0.449\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"5\" valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003ePresented as the mean (ranges). Abbreviations: min, minute. \u003csup\u003e*\u0026nbsp;\u003c/sup\u003e\u003cem\u003ep\u0026nbsp;\u003c/em\u003e\u0026lt;0.05, Post-hoc analysis was conducted, the difference between case 1-8 and case 17-24 was significant (\u003cem\u003ep=\u003c/em\u003e0.045). Post-hoc analysis was conducted, the difference between case 9-16 and case 17-24 was significant (\u003cem\u003ep=\u003c/em\u003e0.014).\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cem\u003eProsthesis Size Planning Accuracy\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eNo malposition, overhang, or obvious protrusion beyond the bony surface was observed in either group. For femoral components, exact-match accuracy was 93.8% for the AI planning system and 58.3% for conventional templating (\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.001). Accuracy within 1 size was 95.8% and 85.4%, respectively, and accuracy within 2 sizes was 100% and 95.8%, respectively. For tibial components, exact-match accuracy was 95.8% for the AI planning system and 47.9% for conventional templating (\u003cem\u003eP\u003c/em\u003e \u0026lt; 0.001). Accuracy within 1 size was 100% and 75.0%, respectively, and accuracy within 2 sizes was 100% and 93.8%, respectively (Table V).\u003c/p\u003e\n\u003cdiv align=\"center\"\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" style=\"width: 100px;\"\u003e\n \u003cp\u003eTable V Comparison of predicted prosthesis size and implanted prosthesis size.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003eAI (n=48)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003eConventional templating (n=48)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e\u003cem\u003eP\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"3\" style=\"width: 86px;\"\u003e\n \u003cp\u003eFemoral prosthesis size (n, %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003eSame\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e45 (93.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e28 (58.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026plusmn;1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e46 (95.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e41(85.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e0.159\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026plusmn;2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e48 (100)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e46 (95.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e0.495\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"3\" style=\"width: 86px;\"\u003e\n \u003cp\u003eTibial prosthesis size (n, %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003eSame\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e46 (95.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e23 (47.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026plusmn;1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e48 (100)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e36 (75.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11px;\"\u003e\n \u003cp\u003e\u0026plusmn;2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17px;\"\u003e\n \u003cp\u003e48 (100)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e45 (93.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e0.242\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" style=\"width: 100px;\"\u003e\n \u003cp\u003eAbbreviations: AI, Artificial intelligence.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe most important finding of this study is that the newly developed AI-assisted planning and PSI system significantly improved component positioning accuracy in UKA, particularly for the tibial component, with performance that appeared comparable to the range previously reported for robot-assisted UKA. It is possible that AI-based planning contributed to surgical execution in both groups; however, this study was not designed to isolate the independent effect of AI-based planning. In addition, the C\u0026ndash;T module demonstrated a clear advantage in image segmentation performance, and the AI planning system accurately predicted prosthesis size.\u003c/p\u003e \u003cp\u003eImage accuracy, the engineer's design concept, and the surgeon's experience are all factors influencing the efficacy of PSI, leading to variability in its clinical outcomes(\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e). First and foremost, precise image segmentation and recognition algorithms are crucial for accurate preoperative bone model reconstruction. Both CT- and MRI-based imaging workflows can be used to generate PSI, although each modality has distinct trade-offs in preoperative planning and guide production(\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e). By using full-length lower limb CT scans combined with X-ray data, we can accurately capture the complex structural morphology of the femorotibial joint and effectively avoid compensatory errors arising from localized imaging data. The CNN-based image architecture U-Net, integrated with the deep learning-based C-T Module, utilizes a hybrid structure that performs global complementary feature perception and enhances detail retention through parallel mixed feature extraction and fusion modules. By continuously interacting with the original input features through element-wise addition and information exchange, the system achieves collaborative optimization of local accuracy and global perception, ensuring high precision and efficiency in information output.\u003c/p\u003e \u003cp\u003eMoreover, the innovative design concept improved the conformity between the cutting guide and the bony surface. In this PSI system, three prominent osteophytes on the femoral and tibial sides\u0026mdash;anteromedial, anterolateral, and posterolateral\u0026mdash;were selected as anchoring zones. These volumetric landmarks formed a stable triangular structure to enhance guide fixation. Notably, to address the commonly reported issue of prolonged operative time associated with PSI use, we minimized the utilization of femoral-side PSI(\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e). Following distal femoral resection, electrocautery was used to mark the bone through a reserved central slot in the guide. Posterior femoral condyle resection was then performed using conventional instruments, centered on the marked line. This approach avoided the need for reinstallation of the femoral PSI before posterior condylar cutting, thereby reducing operative time. Furthermore, the AI-based planning system accurately predicted the implant size, facilitating intraoperative trialing and effectively mitigating the issue of prolonged operative time commonly associated with conventional PSI systems. Interestingly, we found that preoperative AI-assisted planning alone improved surgical accuracy. This may be attributed to the increased availability of predictable preoperative information provided to the surgeon through the planning report. The surgeon was able to anticipate critical intraoperative parameters, including expected bone resection thickness, implant size, as well as details such as osteophyte distribution and bone loss. This enhanced preoperative insight facilitated a more efficient and precise translation of the surgical plan into intraoperative execution.\u003c/p\u003e \u003cp\u003eThe ideal tibial component coverage aims to evenly distribute physiological loads onto the underlying cancellous bone, thereby minimizing stress concentration and preventing complications such as implant subsidence, loosening, and periprosthetic fractures(\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e). Compared to total knee arthroplasty (TKA), UKA involves smaller component sizes and coverage, which may lead to higher stress concentration due to the reduced surface area of the implant. Therefore, the goal for orthopedic surgeons is to perform intraoperative bone resection to select the most anatomically appropriate implant size that maximizes tibial plateau bone coverage, while avoiding overhang and soft tissue irritation caused by oversized components. During medial UKA, tibial tray implantation requires a balance between maximizing bone coverage and minimizing component overhang(\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e). Morphometric or patient-specific tibial designs may improve cortical coverage while reducing overhang and undercoverage(\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e). Currently, there are still relatively few studies on the ideal coverage rate of the tibial prosthesis for UKA. The use of our designed PSI tool significantly improved tibial component coverage, with 92% of cases achieving over 90% coverage. Nevertheless, whether this increased tibial component coverage reduces the risk of implant subsidence and loosening remains to be explored further through long-term follow-up.\u003c/p\u003e \u003cp\u003eRobotic-assisted UKA require substantial financial investment, are associated with a notable learning curve, and often result in longer operative times(\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e). We did not observe a broad deterioration in operative efficiency during PSI adoption, although tibial coronal alignment differed across PSI case-order subgroups. This suggests that AI-PSI-assisted UKA may serve as a more efficient and accessible alternative, particularly in settings with limited medical resources. Although our PSI system significantly enhanced surgical accuracy in UKA, the intraoperative accuracy achieved in clinical cases did not match that observed in synthetic bone model validation. This discrepancy underscores several challenges in the practical implementation of PSI, including interindividual anatomical variations, interference from intraoperative soft tissues, and the limitations of imaging-based planning in accurately replicating cartilage damage and actual bone quality. Moreover, the use of PSI often necessitates a longer surgical incision to ensure adequate exposure of the bony surface and accurate placement of the instrumentation, which may represent an inherent limitation of this technique.\u003c/p\u003e \u003cp\u003eThis study has several limitations. The focus was primarily on the development of a novel AI-assisted planning system and its effectiveness in improving surgical accuracy. Whether this approach translates into improved long-term clinical outcomes or implant survivorship in UKA remains unclear and will be the subject of our future investigations. Additionally, all procedures were performed by a single experienced UKA surgeon, which may limit generalizability to other surgeons and practice settings. As such, the applicability and generalizability of the findings to less experienced surgeons remain uncertain and should be evaluated in future targeted investigations.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThe AI-assisted planning system accurately predicted implant size and, when combined with PSI, improved technical accuracy in medial UKA. Larger studies with longer follow-up are needed to determine whether these technical gains translate into superior clinical outcomes or implant survivorship.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eartificial intelligence\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCT\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ecomputed tomography\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eUKA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eunicompartmental knee arthroplasty\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePSI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003epatient-specific instrumentation\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e3D\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ethree-dimensional\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCNN\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003econvolutional neural network\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eC-T Module\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eCNN-Transformer module\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eDSC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eDice similarity coefficient\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eIoU\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eintersection over union\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eHD95\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003e95th percentile Hausdorff distance\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eASA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAmerican Society of Anesthesiologists\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eICC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eintraclass correlation coefficient\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRMSE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eroot mean square error\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMRI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003emagnetic resonance imaging\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTKA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003etotal knee arthroplasty.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was conducted in accordance with the Declaration of Helsinki. Informed consent was obtained from the volunteer. Ethical approval was obtained from the institutional review board of the First Hospital of Hebei Medical University\u0026nbsp;(approval number: 2025-YanShen00212).\u0026nbsp;The study was registered in the China Medical Research Registration and Filing Information System.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eXingyu Liu and Yafang Zhang are affiliated with\u0026nbsp;Longwood Valley Medical Technology company. The company collaborated in this study and was involved in the artificial intelligence-related technical development and manufacture of the 3-dimensional printed patient-specific instrumentation used in this work. The other authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research was funded by the National Key Research and Development Program (grant numbers 2023YFC3604905) and the Science Fund for Distinguished Young Scholars of Hebei Province (Grants number: ZF2024132 and ZF2024143).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026apos; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDehua Liu:\u003c/strong\u003e Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Validation, Writing \u0026ndash; original draft. \u003cstrong\u003eGang Ji:\u003c/strong\u003e Conceptualization, Data curation, Formal analysis, Methodology, Validation, Writing \u0026ndash; original draft. \u003cstrong\u003eZirun Gao:\u003c/strong\u003e Investigation, Writing \u0026ndash; original draft\u003cstrong\u003e.\u003c/strong\u003e Ye Huang: Validation, Writing \u0026ndash; original draft. \u003cstrong\u003eGuanglei Cao:\u003c/strong\u003e Methodology, Writing \u0026ndash; original draft. \u003cstrong\u003eYafang Zhang:\u003c/strong\u003e Data curation, Writing \u0026ndash; original draft. \u003cstrong\u003eXingyu Liu:\u003c/strong\u003e Project administration, Supervision, Investigation, Validation, Data curation, Writing \u0026ndash; review \u0026amp; editing. \u003cstrong\u003eGuobin Liu:\u003c/strong\u003e Project administration, Funding acquisition, Supervision, Resources, Writing \u0026ndash; review \u0026amp; editing.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe would like to express our gratitude to Tianyu Zhang, Zhao Gu, Rui Yu, Jinlong Niu, Xue Wang, Zhengyu Xu and others for their contributions to this study.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eKugelman DN, Wu KA, Goel RK, Dilbone ES, Ryan SP, Bolognesi MP, et al. Comparing Functional Recovery Between Total and Unicompartmental Knee Arthroplasty: A Prospective Health Kit Study. J Arthroplasty. 2025;40(7S1):S84\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArthur LW, Jenkins C, Dodd CAF, Price AJ, Jackson WFM, Bottomley N, et al. Mid-term outcomes of the fixed-bearing lateral Oxford unicompartmental knee arthroplasty. Bone Joint J. 2025;107\u0026ndash;B(4):432\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTay ML, Matthews BG, Monk AP, Young SW. Disease progression, aseptic loosening and bearing dislocations are the main revision indications after lateral unicompartmental knee arthroplasty: a systematic review. J ISAKOS. 2022;7(5):132\u0026ndash;41.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThoreau L, Morcillo Marfil D, Thienpont E. Periprosthetic fractures after medial unicompartmental knee arthroplasty: a narrative review. Arch Orthop Trauma Surg. 2021;142(8):2039\u0026ndash;48.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang F, Zhao J, He W, He H, Wang Q. Biomechanical analysis of femoral component malalignment in medial unicompartmental knee arthroplasty. Sci Rep. 2025;15(1):44071.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ede Ten Noever GV, Vossen RJM, Bayoumi T, Sierevelt IN, Burger JA, Pearle AD, et al. Distinct age-related modes of failure in cemented and cementless Oxford medial unicompartmental knee arthroplasty: results from 25,762 patients in the Dutch Arthroplasty Register. Bone Joint J. 2025;107\u0026ndash;B(3):329\u0026ndash;36.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang J, Ng N, Scott CEH, Blyth MJG, Haddad FS, Macpherson GJ, et al. Robotic arm-assisted versus manual unicompartmental knee arthroplasty: a systematic review and meta-analysis of the MAKO robotic system. Bone Joint J. 2022;104\u0026ndash;B(5):541\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBell SW, Anthony I, Jones B, MacLean A, Rowe P, Blyth M. Improved Accuracy of Component Positioning with Robotic-Assisted Unicompartmental Knee Arthroplasty: Data from a Prospective, Randomized Controlled Study. J Bone Joint Surg Am. 2016;98(8):627\u0026ndash;35.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGoh GS, Haffar A, Tarabichi S, Courtney PM, Krueger CA, Lonner JH. Robotic-Assisted Versus Manual Unicompartmental Knee Arthroplasty: A Time-Driven Activity-Based Cost Analysis. J Arthroplasty. 2022;37(6):1023\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTay ML, Carter M, Bolam SM, Zeng N, Young SW. Robotic-arm assisted unicompartmental knee arthroplasty system has a learning curve of 11 cases and increased operating time. Knee Surg Sports Traumatol Arthrosc. 2022;31(3):793\u0026ndash;802.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCao G, Du M, Li Z, An S, Huang J, Liu S, et al. Novel patient-specific instrument with comparable accuracy to robotic assistance in medial unicompartmental knee arthroplasty: a prospective study. Int J Surg. 2025;111(7):4487\u0026ndash;94.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLeenders AM, Kort NP, Koenraadt KLM, van Geenen RCI, Most J, Kerens B, et al. Patient-specific instruments do not show advantage over conventional instruments in unicompartmental knee arthroplasty at 2 year follow-up: a prospective, two-centre, randomised, double-blind, controlled trial. Knee Surg Sports Traumatol Arthrosc. 2021;30(3):918\u0026ndash;27.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHafez MA, Moholkar K. Patient-specific instruments: advantages and pitfalls. SICOT J. 2017;3:66.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeng Y, Bai X, Zhao Z, Cao L, Liu Y, Jiang Q. Patient-specific vs. Oxford microplasty instrumentation in unicompartmental knee arthroplasty: a randomized controlled trial. Eur J Med Res. 2025;30(1):1238.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKerens B, Leenders AM, Schotanus MGM, Boonen B, Tuinebreijer WE, Emans PJ, et al. Patient-specific instrumentation in Oxford unicompartmental knee arthroplasty is reliable and accurate except for the tibial rotation. Knee Surg Sports Traumatol Arthrosc. 2017;26(6):1823\u0026ndash;30.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRodriguez HC, Rust BD, Roche MW, Gupta A. Artificial intelligence and machine learning in knee arthroplasty. Knee. 2025;54:28\u0026ndash;49.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchwarz GM, Simon S, Mitterer JA, Frank BJH, Aichmair A, Dominkus M, et al. Artificial intelligence enables reliable and standardized measurements of implant alignment in long leg radiographs with total knee arthroplasties. Knee Surg Sports Traumatol Arthrosc. 2022;30(8):2538\u0026ndash;47.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHan K, Wang Y, Chen H, Chen X, Guo J, Liu Z et al. A Survey on Vision Transformer. IEEE Trans Pattern Anal Mach Intell. 2022;45(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOurselin S, Sabuncu MR, Wells W, Joskowicz L, Unal G, Maier A. The 19th International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI 2016). Med Image Anal. 2017;41:1.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePark JH, Rasouli MR, Mortazavi SMJ, Tokarski AT, Maltenfort MG, Parvizi J. Predictors of perioperative blood loss in total joint arthroplasty. J Bone Joint Surg Am. 2013;95(19):1777\u0026ndash;83.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlvand A, Khan T, Jenkins C, Rees JL, Jackson WF, Dodd CAF, et al. The impact of patient-specific instrumentation on unicompartmental knee arthroplasty: a prospective randomised controlled study. Knee Surg Sports Traumatol Arthrosc. 2017;26(6):1662\u0026ndash;70.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan Leeuwen JAMJ, R\u0026ouml;hrl SM. Patient-specific positioning guides do not consistently achieve the planned implant position in UKA. Knee Surg Sports Traumatol Arthrosc. 2016;25(3):752\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVolpi P, Prospero E, Bait C, Cervellin M, Quaglia A, Redaelli A, et al. High accuracy in knee alignment and implant placement in unicompartmental medial knee replacement when using patient-specific instrumentation. Knee Surg Sports Traumatol Arthrosc. 2013;23(5):1292\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThijs E, Theeuwen D, Boonen B, van Haaren E, Hendrickx R, Vos R, et al. Comparable clinical outcome and implant longevity after CT- or MRI-based patient-specific instruments for total knee arthroplasty: a 2-year follow-up of a RCT. Knee Surg Sports Traumatol Arthrosc. 2019;28(6):1821\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMakaram NS, Yapp LZ, Bowley ALW, Garner A, Scott CEH. Polyethylene wear in metal-backed tibial components in unicompartmental knee prostheses. J ISAKOS. 2024;9(6):100324.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKazarian GS, Barrack TN, Okafor L, Barrack RL, Nunley RM, Lawrie CM. High Prevalence of Radiographic Outliers and Revisions with Unicompartmental Knee Arthroplasty. J Bone Joint Surg Am. 2020;102(13):1151\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEscudier J-C, Jacquet C, Flecher X, Parratte S, Ollivier M, Argenson J-N. Better Implant Positioning and Clinical Outcomes With a Morphometric Unicompartmental Knee Arthroplasty. Results of a Retrospective, Matched-Controlled Study. J Arthroplasty. 2019;34(12):2903\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMiyake Y, Namba Y, Mitani S, Umehara N, Kawamoto T, Furuichi S. Comparison of tibial implant positioning between symmetrical and anatomical design implants in unicompartmental knee arthroplasty for Japanese patients. J Orthop Surg (Hong Kong). 2023;31(1):10225536221149485.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCarpenter DP, Holmberg RR, Quartulli MJ, Barnes CL. Tibial plateau coverage in UKA: a comparison of patient specific and off-the-shelf implants. J Arthroplasty. 2014;29(9):1694\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGoh GS, Haffar A, Tarabichi S, Courtney PM, Krueger CA, Lonner JH. Robotic-Assisted Versus Manual Unicompartmental Knee Arthroplasty: A Time-Driven Activity-Based Cost Analysis. J Arthroplasty. 2022;37(6):1023\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"arthroplasty","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Arthroplasty](https://link.springer.com/journal/42836)","snPcode":"42836","submissionUrl":"https://submission.springernature.com/new-submission/42836/3","title":"Arthroplasty","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Open","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"patient-specific instrumentation, artificial intelligence, unicompartmental knee arthroplasty, surgical precision, learning curve","lastPublishedDoi":"10.21203/rs.3.rs-9340442/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9340442/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eArtificial intelligence (AI)-assisted preoperative planning based on computed tomography (CT) may improve anatomic characterization for unicompartmental knee arthroplasty (UKA), and patient-specific instrumentation (PSI) may improve the accuracy of component positioning. We developed and validated an AI-assisted preoperative planning workflow combined with PSI for medial UKA and evaluated its effect on implant positioning accuracy.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eA hybrid architecture combining a convolutional neural network-based U-Net with a Transformer-based deep learning module (C-T Module) was developed to automate CT processing for AI-assisted preoperative planning and PSI design in UKA. Segmentation performance of the C-T Module was compared with that of a conventional 3D U-Net. PSI feasibility was validated using synthetic bone models. In a prospective randomized clinical study, 24 patients underwent AI-based planning plus PSI-assisted UKA (PSI group) and 24 underwent AI-based planning plus conventionally instrumented UKA (control group). Surgical accuracy, perioperative outcomes, and implant-size prediction accuracy were compared.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eThe C-T Module demonstrated superior image segmentation accuracy compared to the classical 3D U-Net. Compared with the control group, the PSI group significantly improved surgical accuracy, including more accurate tibial component positioning, greater tibial coverage, and less deviation in proximal tibial resection (all \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Except for the significantly longer skin incision in the PSI group (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001), no other perioperative parameters differed significantly between groups. No evident learning curve was observed with PSI use. Moreover, the AI-based planning system demonstrated significantly higher accuracy in prosthesis size prediction than conventional templating (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eThe novel AI-assisted planning system accurately predicts prosthesis size and, when combined with PSI, significantly improved component positioning accuracy of UKA without a notable learning curve.\u003c/p\u003e","manuscriptTitle":"Artificial Intelligence-Assisted Preoperative Planning Combined with Patient-Specific Instrumentation Improves the Accuracy of Unicompartmental Knee Arthroplasty","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-11 06:22:03","doi":"10.21203/rs.3.rs-9340442/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-05-06T01:00:22+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"219020308560245797704249219417131431370","date":"2026-05-01T10:54:38+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"213609243150986086364423640643355522530","date":"2026-05-01T00:23:53+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-29T09:54:47+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-04-23T09:23:03+00:00","index":"","fulltext":""},{"type":"submitted","content":"Arthroplasty","date":"2026-04-20T09:54:33+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"arthroplasty","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Arthroplasty](https://link.springer.com/journal/42836)","snPcode":"42836","submissionUrl":"https://submission.springernature.com/new-submission/42836/3","title":"Arthroplasty","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Open","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"aca3984f-e505-48d7-9b33-4143095541fb","owner":[],"postedDate":"May 11th, 2026","published":true,"recentEditorialEvents":[{"type":"editorInvitedReview","content":"","date":"2026-05-06T01:00:22+00:00","index":14,"fulltext":""},{"type":"reviewerAgreed","content":"219020308560245797704249219417131431370","date":"2026-05-01T10:54:38+00:00","index":12,"fulltext":""},{"type":"reviewerAgreed","content":"213609243150986086364423640643355522530","date":"2026-05-01T00:23:53+00:00","index":11,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-11T06:22:03+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-11 06:22:03","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9340442","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9340442","identity":"rs-9340442","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}