Full Motion Arc, Laxity-Based Planning for Total Knee Arthroplasty: The Next Milestone for Enabling Technology

Abstract

Background

Although numerous technologies show promise to improve total knee arthroplasty outcomes, a need to further advance these solutions remains. An area for potential improvement is intraoperative efficiency by reducing the number of required surgeon and software interactions. Therefore, the purpose of the study was to compare the number of surgeon and software interactions for 9 surgical steps between a laxity-based algorithm and a historical algorithm.

Methods

Technical logs of 313 total knee arthroplasties performed by 6 surgeons were reviewed. Nine surgical parameters were assessed, including bone cuts and implant positioning. These parameters were compared between the laxity-based algorithm and the historical algorithm cohorts. The number of adjustments from the initial predictive plan to the final plan was calculated for each of the 9 parameters to assess potential improvements in surgical efficiency with the laxity-based algorithm.

Results

Overall, the laxity-based algorithm substantially improved the initial planning definition. The mean number of active surgeon click adjustments to the original plan decreased from 13 to 8 clicks. When assessing based on the surgeon, a mean 39% reduction in clicks was noted, ranging from a 19% reduction to a 70% reduction.

Conclusions

This study exemplified the potential of an unsupervised learning algorithm in conjunction with a laxity-based device. The laxity-based algorithm substantially reduced the number of active surgeon interactions with the software, which can potentially help reduce cognitive burden and operative time.

Introduction

Soft tissue and bone balancing play a critical role in successful total knee arthroplasty (TKA) outcomes [1]. Over the past decade, intraoperative technologies have targeted balancing theories to help optimize bone cuts and implant positioning [2]. These technologies can integrate patient-specific anatomy, shifting techniques from a general to a more personalized approach [3]. In addition, these technologies allow for real-time, dynamic, 3-dimensional intraoperative measurements rather than static, 2-dimensional measurements. Given these advancements, intraoperative technologies have proven advantageous in the settings of severe deformity, precise and accurate implant positioning, and all-cause 5-year revision rates [4,5].

A large reason why intraoperative technologies are advantageous is that they can provide an immense depth and breadth of information, far more than with manual instrumentation. However, the incorporation of this volume of data then falls onto the surgeon to decipher in real time in the operating room. Currently, robotic, navigation, and computer-assisted devices require the surgeon to interact with the software either directly or indirectly by guiding a member of the operating room staff. The next phase in the technological revolution of TKA is to optimize these established devices by reducing surgeon-technology interaction burden.

Unsupervised learning algorithms have the potential to autonomously learn surgeon preferences, thereby reducing the required number of surgeon-technology interactions. However, little on this topic has been investigated in the scope of TKA. Therefore, the purpose of the study was to assess the validity of an unsupervised learning algorithm. Specifically, we compared if the algorithm could reduce the number of surgeon and software active interactions for 9 surgical parameters.

Material and methods

Patients

A retrospective review was conducted on a proprietary, cloud-based database that archives technical logs of cases performed with an instrumented computer-assisted orthopaedic surgery (CAOS) system (Fig. 1) [6]. The study included 313 cases associated with a tibia-first technique performed by 6 surgeons without any exclusions, such that 1 cohort would not be skewed one way or the other. For each case, deidentified surgical reports were stored, containing only technical details, such as surgical time, workflow, surgical parameters, implant information, etc. All surgeons had prior experience with this system and were beyond their initial learning curve, as evidenced by the number of cases reported in Tables 1 and 2 [7]. In addition, no preoperative and postoperative clinical data were stored; therefore, this study was exempt from institutional review board approval.

Figure 1.

A platform combining a CAOS system and an intra-articular device (instrumented), allowing the acquisition of the compartment-specific knee joint gaps throughout the full range of motion under force-controlled distraction maintained by 2 antagonist mechanical actuators [6].

Table 1.

Distribution of 313 cases associated with 6 surgeons in terms of historic planning algorithm strategies.

Surgeon	Number of cases	Mechanical alignment cases	Kinematic alignment cases
1	71	71	0
2	109	5	104
3	35	0	35
4	21	21	0
5	31	31	0
6	46	46	0

Table 2.

Distribution of 313 cases associated with 6 surgeons in terms of deformity classes.

Surgeon	Number of cases	Valgus cases	Neutral cases	Low varus cases	Moderate varus cases	High varus cases
1	71	6	11	21	28	5
2	109	13	27	34	29	6
3	35	0	8	19	8	0
4	21	0	3	3	4	11
5	31	0	0	5	8	18
6	46	3	15	12	12	4

Surgical technique

All cases followed a similar surgical workflow using a gap-balancing, tibia-first cutting technique guided by the CAOS system (Advita GPS; Blue Ortho, Meylan, France). After active tracking arrays were attached to the femur and tibia, anatomic landmarks were acquired by the imageless CAOS (Fig. 2a). To assess patient deformity and kinematic patterns, the knee joint was manipulated through its flexion arc, whereas the CAOS captured varus and valgus (V&V) angles at distinct flexion angles through the flexion arc (Fig. 2b). The proximal tibia resection was then performed based on what the surgeon deemed necessary for each unique patient, regarding cut thickness, V&V, and posterior slope (Fig. 2c). At this stage, the intra-articular device (Newton; Exactech, Gainesville, Florida) was placed between the proximal tibial cut surface and the native femur and then was mechanically released to apply a distraction force with the extensor mechanism in place. This patented intra-articular device [8] features 2 antagonist mechanical actuators intended to maintain a quasi-constant distraction force of 90 N per compartment (ie, 180N across the joint) regardless of the joint gap (Fig. 1) [6]. Next, the knee was taken through the arc of motion, and both medial and lateral gap sizes were captured by the CAOS system (Fig. 2d). After the acquisition of the joint gaps, the planning algorithm proposes a preliminary femoral surgical planning (initial planning), which was subsequently personalized by the surgeon (final planning), leveraging these inputs. Therefore, the planning of femoral resection parameters was determined based on soft-tissue balance, alignment, and sizing considerations (Fig. 2e) based on the surgeon’s preference.

Figure 2.

Overview of the tibia-first surgical workflow: (a) anatomic landmark acquisition, (b) preoperative kinematics acquisition, (c) planning and resection of proximal tibia, (d) medial and lateral gap acquisition for full flexion arc with Newton, and (e) femoral cuts planning in terms of alignment, size, and laxity [6].

Planning algorithms

The planning algorithm aims to generate an initial femoral surgical plan, which can be personalized by the surgeon according to their preference in terms of alignment, size, and laxity. The femoral surgical plan should be considered from a holistic point of view by detailing not only information relative to the femur but also to the entire knee joint with the definition of the laxities and the overall alignment (eg, hip-knee-ankle [HKA] angle). Each planning algorithm takes as input a subset of anatomic parameters intended to lock the degrees of freedom associated with the femoral component.

Historical methods

The historical methods focus on alignment, cut thickness, and size considerations through 2 different philosophies: the mechanical alignment and the kinematic alignment. The mechanical algorithm uses V&V angle, component flexion, axial rotation, distal cut height from one of the condyles, and anterior notching as targets to optimize femoral implant position (Table 2). The kinematic algorithm utilizes distal cut height from both condyles, posterior cut height from both condyles, and anterior notching to optimize the femoral implant position (Table 2). In both algorithms, the femoral implant position is centered on the mediolateral axis, and the size is optimized to match anteroposterior anatomic landmarks of the femur. All the target values are defined by surgeons’ preferences before surgery and are not case dependent. The distribution of cases in terms of historical algorithm strategies for 6 surgeons is presented in Table 1.

Laxity-based method

In contrast, the patent-pending laxity-based method uses targeted medial and lateral laxity values in both extension and flexion, as well as the component flexion, as inputs to optimize the femoral implant size, orientation, and position (Table 3). In this case, the femoral component is also centered on the mediolateral axis, and the algorithm aims for a neutral anterior notching. As an attempt to further personalize the proposed surgical plan, for each surgeon, the cases were categorized based on patient deformity cluster. The patient deformity clustering (PDC) technique and results are described in the next subsection. The average final laxity planned values were computed for each cluster separately as the target input for the laxity-based planning algorithm.

Table 3.

Usage of different anatomic values as target for the historic planning algorithm and laxity-based algorithm.

Parameter	Historical mechanical alignment	Historical kinematic alignment	Laxity-based optimization
Proximodistal position	Distal CH (from 1 condyle)	Distal CH medial and lateral	Laxities in extension medial and lateral
Anteroposterior position	Notching	Posterior CH medial and lateral + notching	Laxities in flexion medial and lateral
Mediolateral position	Centered on anatomic landmark	Centered on anatomic landmark	Centered on anatomic landmark
Orientation in frontal plane	Varus	Distal CH medial and lateral	Laxities in extension medial and lateral
Orientation in sagittal plane	Flexion	Posterior CH medial and lateral + notching	Flexion
Orientation in axial plane	Rotation	Posterior CH medial and lateral	Laxities in flexion medial and lateral
Implant size	Native knee anteroposterior dimension	Native knee anteroposterior dimension	Laxities in flexion medial and lateral + notching

CH, cut height.

Patient deformity clustering

The objective of the PDC approach is to categorize kinematics data recorded during surgery (Fig. 2b). Unlike methods that focus on static HKA angle acquired at specific flexion angles [[9], [10], [11]], the PDC method leverages the dynamic HKA angle continuously acquired from extension to flexion to classify patient deformities. The PDC method was trained on a dataset of 1221 cases, capturing full flexion-arc coronal plane angles from operations performed by 11 surgeons and is the foundational data on which the learning algorithm utilized in this study was built. This method applies an unsupervised learning algorithm, specifically K-means clustering [12], in an N-dimensional V&V feature space to categorize deformities into $K$ clusters.

K-means clustering was used to group cases with similar V&V kinematic patterns across knee flexion. For each case, V&V angles were measured at 9 flexion angles (10°, 20°, 30°, 45°, 60°, 75°, 90°, 105°, and 120°), allowing the full coronal-plane kinematic profile of the knee to be represented rather than a single flexion position. Although the clustering model operates in a multidimensional feature space, cases exhibiting similar V&V kinematic trajectories across flexion were grouped into the same cluster (Fig. 3).

Figure 3.

K-means fitting results. (a) Data visualization before clustering. (b) Data visualization after clustering. (c) Centroid 2D trajectory visualization (mean V&V at flexion angles for each cluster). The figure presents a 3-dimensional projection of representative flexion angles (10°, 45°, and 90°) to visualize cluster separation, along with centroid trajectories used to label clusters as neutral, valgus, low varus, moderate varus, and high varus.

Evaluation metrics

For both algorithms (historical and laxity based), the total number of adjustment clicks across 9 cut parameters was calculated to measure the transition from the initial plan proposed by the algorithm to the final plan validated by the surgeon. Adjustment clicks represent the number of screen interactions made by the surgeon, with each click indicating a 1 mm or 1° change. The adjustment clicks for a parameter are computed by rounding off to the nearest whole number as follows:

$$\text{Adjustment}\text{clicks}=\text{round}\left|\text{final}\text{planning}\text{value}-\text{initial}\text{planning}\text{value}\right|$$

The cut parameters include distal medial cut height, distal lateral cut height, posterior medial cut height, posterior lateral cut height, notching, coronal plane angle, axial rotation, flexion, and size. Rather than directly calculating the mean adjustment clicks for each surgeon across all cases, a weighted mean was computed by grouping with respect to deformity classes. For each deformity class, mean adjustment clicks ($X_i$) are computed separately for both the historical method and the laxity-based method across all cases related to the specific deformity class. Weights $w_i$ s were determined based on the proportion of cases in each deformity class relative to the total cases. The weighted mean $W$ for each parameter corresponding to each surgeon, adjusted for deformity class distribution, was computed as follows:

$$W=\frac{\sum _{i=1}^{i=5}{w}_i{X}_i}{\sum _{i=1}^5{w}_i}$$

Finally, the sum of the weighted mean adjustment clicks across all 9 cut parameters for each surgeon represents the mean total clicks required to transition from the initial plan to the final plan. This metric will help us to compare both methods: the historical method and the laxity-based method.

Data analysis

Comparisons between the historical and laxity-based algorithms were performed at the surgeon level using the sum of weighted mean adjustment clicks for each surgeon under both algorithms, yielding 6 paired observations. A paired t-test was used to compare algorithms, and a Wilcoxon signed-rank test was performed as a distribution-free confirmation. Statistical significance was defined as P < .05.

Results

Overall, the laxity-based algorithm substantially reduced the mean number of active intraoperative adjustment clicks performed by the surgeon compared with the historical algorithm (8 vs 13 clicks; Tables 4 and 5). This reduction was statistically significant (paired t-test: t(5) = 2.80, P = .038; Wilcoxon signed-rank test: W = 21, P = .031). In addition, the number of surgeon interactive clicks was reduced or the same for 8 of the 9 parameters assessed. Specifically, clicks were reduced for distal medial (1 vs 2), distal lateral (1 vs 2), posterior medial (1 vs 2), and posterior lateral (1 vs 2) cut heights, as well as varus adjustments (1 vs 2). The number of interactive clicks was the same for notching avoidance, rotation, flexion, and size (all 1 click).

Table 4.

Historical method results: each value in the table represents weighted mean adjustment clicks computed using weights and mean adjustment clicks corresponding to deformity classes.

Surgeon/parameter	Distal medial CH	Distal lateral CH	Posterior medial CH	Posterior lateral CH	Notching	Varus	Rotation	Flexion	Size	Sum clicks
Surgeon 1 (N = 71)	1.44	1.63	3.03	3.03	0.79	1.82	0.04	2.76	0.04	14.58
Surgeon 2 (N = 109)	1.46	1.92	1.25	1.49	0.08	2.72	1.74	1.68	0.06	12.40
Surgeon 3 (N = 35)	1.91	2.86	2.11	3.16	0.60	2.63	2.34	2.77	0.23	18.62
Surgeon 4 (N = 21)	1.57	1.81	3.05	2.43	0.52	1.33	0.86	1.33	0.14	13.04
Surgeon 5 (N = 31)	1.35	0.90	2.13	1.81	0.23	1.52	0.87	1.58	0	10.39
Surgeon 6 (N = 46)	1.33	2.24	1.15	1.13	0.07	1.91	0	0.70	0.04	8.57
Mean clicks	1.51	1.89	2.12	2.18	0.38	1.99	0.98	1.8	0.09	12.93

CH, cut height; sum clicks, sum of weighted mean adjustment clicks for 9 parameters; mean clicks, mean of weighted mean adjustment clicks for 6 surgeons.

Table 5.

Laxity-based method results: each value in the table represents weighted mean adjustment clicks computed using weights and mean adjustment clicks corresponding to deformity classes.

Surgeon/parameter	Sum clicks (historical)	Sum clicks (laxity)	Sum clicks (reduction %)
Surgeon 1 (N = 71)	14.58	8.12	44.3
Surgeon 2 (N = 109)	12.40	10	19.4
Surgeon 3 (N = 35)	18.62	5.54	70.2
Surgeon 4 (N = 21)	13.04	8.76	32.8
Surgeon 5 (N = 31)	10.39	8.45	18.7
Surgeon 6 (N = 46)	8.57	6.84	20.2
Mean clicks	12.93	7.95	38.5

CH, cut height; sum clicks, sum of weighted mean adjustment clicks for 9 parameters; mean clicks, mean of weighted mean adjustment clicks for 6 surgeons.

When assessing based on the surgeon, all surgeons had a reduction in the number of clicks using the laxity-based algorithm in the CAOS system with an intra-articular device (Table 6). Surgeon 3 had the greatest reduction in interactive clicks from 19 to 6 clicks (70% reduction), whereas surgeon 5 had the least reduction in the number of clicks from 10 to 8 (19%). Overall, compared with the historical algorithm, there was a nearly 39% reduction in the number of clicks with the laxity-based algorithm.

Table 6.

Reduction in number of weighted mean adjustment clicks for laxity-based method when compared with historical method.

Surgeon/parameter	Distal medial CH	Distal lateral CH	Posterior medial CH	Posterior lateral CH	Notching	Varus	Rotation	Flexion	Size	Sum clicks
Surgeon 1 (N = 71)	1.04	1.13	0.90	0.93	0.79	1.03	0.17	1.96	0.17	8.12
Surgeon 2 (N = 109)	1.24	1.27	0.99	1.40	0.08	1.50	1.44	1.92	0.16	10
Surgeon 3 (N = 35)	0.69	0.60	0.74	0.66	0.31	0.51	0.66	1.37	0	5.54
Surgeon 4 (N = 21)	1.10	1.57	1.19	1.29	0.52	0.90	0.86	1.33	0	8.76
Surgeon 5 (N = 31)	0.71	1.35	1.29	1.16	0.23	1.45	0.87	1.29	0.10	8.45
Surgeon 6 (N = 46)	1	1.02	1	1.13	0.07	0.87	0.04	1.54	0.17	6.84
Mean clicks	0.96	1.16	1.02	1.10	0.33	1.04	0.67	1.57	0.10	7.95

Sum clicks, sum of weighted mean adjustment clicks for 9 parameters; mean clicks, mean of weighted mean adjustment clicks for 6 surgeons.

Discussion

As new technologies become more widely utilized in the operating room, progress from the early phases of establishing the utility of this technology to making this technology more efficient is needed. One area for potential optimization is developing technology with robust predictive algorithms that can incorporate real-time bony and ligamentous data throughout the entire knee arc of motion. Prior studies have already highlighted the benefits of the gap-balancing device utilized in this study [6,13]. The importance of this study was to investigate the adaptive learning capabilities of the integrated automated planning software for the gap-balancing device. The data from this study found a reduction in surgeon- and software-required interactions, thus supporting the utility of the predictive algorithm.

Other studies have also identified the potential advantages of a laxity-based device. Mooney et al. [14] compared computer-assisted navigation vs manual instrumentation to assess TKA cut outliers (final cut >2° or 2 mm from the plan). The group found the navigation cohort to have significantly fewer outliers from 35% to 4% for experienced users (P < .001) and from 34% to 10% for trainees (P < .001). Fan et al. [6] evaluated the first 273 cases using an intraoperative navigation device and found that the device was able to achieve consistent medial and lateral compartment gaps throughout the full arc of motion from predictive to final executed plans. Another study by Valtanen et al. [13] assessed patient-reported outcomes in 100 consecutive computer-navigated TKAs where a balancer was used to 100 consecutive computer-navigated TKAs where a balancer was not used and found statistically and clinically significant postoperative improvements in the balancer cohort. The authors hypothesized that these outcomes are likely because of the improved full arc of motion stability data provided by the balancer.

While personalized workflows offer significant benefits, they also introduce complexity. The transition from traditional 2D static assessments to 3D dynamic acquisitions throughout the arc of motion increases the data to be processed, lengthens the learning curve, and extends surgical durations. Also, the inclusion of soft-tissue considerations along with alignment and sizing considerations further complexify surgical planning [15]. Similarly, the personalized workflows are expected to increase both physical and cognitive burden. This increased burden highlights the need for automated, intelligent algorithms to streamline surgical planning and mitigate the cognitive and physical demands placed on surgeons. Such algorithms can dynamically integrate surgeon preferences, patient-specific anatomic data, and intraoperative inputs to generate optimized surgical plans. By reducing the number of manual adjustments required during surgery, these systems help minimize decision fatigue and allow surgeons to focus on the precise execution of the procedure.

In this study, the implementation of a laxity-based algorithm demonstrated a significant reduction in surgeon interaction clicks, nearly 39% compared with historical algorithms. This reduction reflects the efficiency of the algorithm in addressing alignment and soft-tissue considerations, decreasing the need for intraoperative recalibration and fine-tuning. Importantly, the variations in click reductions (Table 6) observed among different surgeons suggest that the effectiveness of the algorithm is influenced by individual preferences, particularly the prioritization of laxity vs alignment in surgical planning. This highlights the need for customizable planning tools that can adapt to varying surgical philosophies and techniques. Another contribution to the significant reduction in surgeon to software interaction is the familiarity with the software and device with a quick learning curve. Future work will aim to refine these algorithms further by assigning dynamic weightages to these attributes, enabling even greater personalization and efficiency.

This study is not without limitations. There could potentially be some functional alignment incorporated in the laxity-based cohort that might not have been part of the historical cohort. However, while this might be a potential concern, for the purposes of this study, surgeons focused on being within the described parameters for the laxity-based technique and should not skew the cohort comparison applicability substantially. In addition, surgeons were able to utilize multiple types of implant constraints, per surgeon's choice. Furthermore, clinical correlations and surgeon's past experiences with this device were not evaluated. Actual operative time savings from the reduction in the number of clicks were also not calculated; however, operative time was not compared as it can be impacted by a large number of factors, and the particular scope of this study was to focus on surgeon-technology interaction. Another limitation is the small sample size of 6 surgeons, which reduces statistical power and may limit generalizability. To address this, both paired t-test and Wilcoxon signed-rank test were performed, and results were consistent across methods. Nevertheless, despite these limitations, this study represents positive findings toward technology, which has reproducible and excellent results that can be applied to various TKA techniques and implant designs. Future studies can include randomization instead of comparison to historical controls to help reduce any potential confounding or inherent biases as well as comparisons of operative times and clinical outcomes.

Conclusions

The integrated laxity-based algorithm evaluated in this study substantially reduced the mean number of surgeon interactive clicks needed to adjust the autogenerated operative plan to achieve optimal final implant positioning. Although clinical implications of this application remain to be tested, the results from the study show strong promise for unsupervised, predictive algorithms in the surgical setting.

CRediT authorship contribution statement

Nipun Sodhi: Writing – review & editing, Writing – original draft, Validation, Investigation, Formal analysis, Data curation, Conceptualization. Laurent D. Angibaud: Writing – review & editing, Writing – original draft, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Prudhvi Tej Chinimilli: Writing – review & editing, Writing – original draft, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Florian Kerveillant: Writing – review & editing, Writing – original draft, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. James I. Huddleston: Writing – review & editing, Writing – original draft, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization.

Conflicts of interest

L.D. Angibaud is a paid employee for Advita Ortho, has stock or stock options in Advita Ortho, is a board member/committee appointments for a society Personalized Arthroplasty Society (PAS); P.T. Chinimilli is a paid employee for Advita Ortho, has stock or stock options in Advita Ortho; F. Kerveillant is a paid employee for Blue-Ortho. J.I. Huddleston receives royalties from Advita Ortho and DePuy 2, is a speaker bureau/paid presentations for Advita Ortho, is a paid consultant for Advita Ortho and DePuy, has stock or stock options in Corin and Porosteon, receives research support as a principal investigator from ZimmerBiomet and Apple, receives royalties, financial, or material support from publishers WoltersKluwer Medical/Orthopaedic publications editorial/governing board at Journal of Arthroplasty, and is a board member/committee appointments for AAHKS, AJRR, Knee Society, and AAOS; all other author(s) declare(s) no potential conflicts of interest. For full disclosure statements, refer to https://doi.org/10.1016/j.artd.2026.101959.