Development and validation of machine learning models for classifying cancer-related sarcopenia using Kinect-based mixed-reality exercises in breast cancer survivors

Abstract

Background

Sarcopenia in cancer survivors is often underdiagnosed due to limited access to imaging-based diagnostic tools such as computed tomography (CT) or dual-energy X-ray absorptiometry (DXA). Indirect classification using movement data may offer a practical, scalable alternative. This study aimed to develop and validate machine learning (ML)-based classification models for cancer-related sarcopenia using joint angle data obtained from Kinect-based mixed-reality (KMR) devices, aiming to improve classification accuracy and identify key movement-related predictors.

Methods

Overall, 77 breast cancer survivors (mean age, 48.9±5.4 years) were included based on stage I–III diagnosis, treatment completion ≥6 months prior, no metastasis, low physical activity, and no major comorbidities. Sarcopenia was diagnosed using skeletal muscle index (SMI) (<5.7 kg/m²) and handgrip strength (HGS) (<18 kg). KMR device data were collected during 8 weeks of exercise. After preprocessing, the dataset was randomly split (8:2) for training and testing. Four ML models—support vector machine (SVM), K-nearest neighbor (KNN), random forest (RF), and XGBoost (XGB)—were trained. Five-fold cross-validation was used for tuning, and feature importance was analyzed.

Results

Of the 38 participants in the exercise group included in the final analysis, 12 (31.5%) were initially diagnosed with sarcopenia. After the 8-week KMR device exercise intervention, 3 participants showed recovery from sarcopenia, resulting in 9 (23.6%) remaining classified with the condition. In the test set, the XGB model demonstrated the highest performance, achieving 94.7% accuracy, 91.2% recall, 95.8% precision, 93.4% F1 score, and 96.2% area under the curve (AUC). Feature importance analysis using RF and XGB consistently identified right “knee flexion (right)” as the most influential predictor.

Conclusions

Among ML classification models trained on KMR device joint data, XGB demonstrated the best performance. Right knee flexion emerged as the most influential feature in sarcopenia classification. These findings suggest that KMR device movement analysis may serve as a practical, non-invasive screening tool for sarcopenia, enabling early detection and personalized intervention strategies for breast cancer survivors in both clinical and remote settings.

Highlight box.

Key findings

• Our study demonstrated that among machine learning models, XGBoost (XGB) outperformed others in sarcopenia classification for breast cancer survivors, achieving 94.7% accuracy, 91.2% recall, 95.8% precision, 93.4% F1 score, and 96.2% area under the receiver operating characteristic curve (AUC). Feature-importance analysis revealed that “right knee flexion” as the most impactful variable in sarcopenia classification.

What is known and what is new?

• Sarcopenia is a significant health challenge for cancer survivors, and existing methods for classification rely heavily on clinical assessments, such as the skeletal muscle index and hand grip strength.

• This manuscript demonstrates that Kinect-based mixed-reality (KMR) device data, combined with machine learning (ML) models, can accurately classify sarcopenia. The XGB model showed the best classification performance, and right knee flexion was identified as a key predictor of sarcopenia.

What is the implication, and what should change now?

• The study highlights the potential of leveraging KMR device technology and ML algorithms for personalized sarcopenia management in breast cancer survivors. These findings suggest a shift toward more precise and data-driven interventions targeting key movement patterns like knee flexion. Future rehabilitation programs should consider integrating KMR exercise monitoring and ML-guided evaluation to enhance outcomes for survivors.

Introduction

Background

Cancer-related sarcopenia, a progressive loss of muscle mass and function, is a common complication among cancer survivors (1). It affects physical performance and increases the risk of adverse health outcomes, such as decreased quality of life, increased risk of falls and fractures, and compromised ability to undergo cancer treatment. Recognizing and addressing sarcopenia in cancer survivors is crucial for improving their overall well-being and treatment outcomes (2,3).

Rationale and knowledge gap

Exercise interventions for sarcopenia in patients with cancer have shown promise in preserving muscle mass and function (4,5). They have been widely explored as non-pharmacological approaches to mitigate sarcopenia (6). However, the existing studies on exercise interventions for sarcopenia in breast cancer survivors have limitations.

First, the volume and intensity of exercise interventions vary across studies, making it challenging to compare and generalize the findings. Second, some studies have focused on aerobic exercise, whereas others have examined resistance training or a combination of both (7,8). However, the optimal exercise modality for the prevention or management of sarcopenia in breast cancer survivors remains unclear.

Adherence to exercise interventions is a significant challenge. Breast cancer survivors face various barriers to exercise participation, including fatigue, treatment-related side effects, and time constraints (9). Considering these factors and developing strategies to improve adherence to and long-term engagement in exercise programs is essential.

In addition to the challenges in exercise interventions, there are also practical limitations in the standard diagnostic methods for sarcopenia, such as computed tomography (CT), dual-energy X-ray absorptiometry (DXA), or bioelectrical impedance analysis (BIA). Standard diagnostic tools for sarcopenia are limited in outpatient and community settings due to high cost, radiation exposure, equipment demands, and reduced accuracy in non-controlled environments (10,11). Several studies have explored indirect approaches to classify sarcopenia using gait speed or grip strength, which are easier to implement than imaging-based methods (12,13). However, these assessments are often limited to clinical settings due to the lack of widespread infrastructure, and may not be feasible in community or remote environments. To address these issues, our study utilizes Kinect-based mixed-reality (KMR) devices to collect high-resolution joint angle data and applies machine learning (ML) models to classify sarcopenia. This approach offers a practical, scalable, and potentially remote-friendly solution.

In recent years, Kinect devices have emerged as a novel approach for exercise intervention (14). Kinect devices, originally developed for gaming purposes, offer unique advantages for monitoring and analyzing movement patterns during exercise sessions (15). These devices use depth-sensing cameras to precisely capture joint angle data, thereby providing valuable insights into the quality and efficiency of movement.

Virtual reality (VR) has been widely explored in exercise intervention studies. By leveraging VR technology, users can exercise in realistic virtual environments, which can enhance their enjoyment and participation in physical activities (16,17). Using VR devices and sensors, the real-time tracking of user movements can be achieved, allowing for accurate feedback and improved effectiveness (18).

Furthermore, exergaming, which is a combination of exercise and gaming, is used in exercise interventions (19). It promotes user participation and motivation by incorporating game elements into physical activities (20). However, it is worth noting that many of these studies using Kinect devices, VR, and exergaming have limitations in terms of data collection and analysis of user data. The lack of extensive data collection and analysis has been a challenge for evaluating the effectiveness of these interventions.

Our study addresses this limitation by collecting joint data from users through KMR device. These devices allow for the collection of precise joint data, providing a unique advantage in analyzing movement patterns and evaluating the effectiveness of exercise interventions. This differentiation allowed for a more comprehensive understanding of the impact of these interventions on user movement and exercise outcomes.

ML techniques are promising in the field of sarcopenia research, particularly in the development of classification models and prediction algorithms (21), which can help identify individuals who may be at risk of sarcopenia and help monitor the disease progression. However, most existing studies using ML techniques for sarcopenia have focused on imaging or ordinary health-related data (22,23). These studies typically employed various types of data such as CT images and measurements related to muscle mass or strength to develop models that can accurately classify individuals as having sarcopenia.

Additionally, ML techniques have been employed to analyze joint data and recognize specific movements or actions (24). These models can be used in various applications, such as gait analysis, sports performance monitoring, and rehabilitation exercises (25,26). However, these studies focused on only movement assessment and recognition.

Objectives

Therefore, the present study’s objective was to use ML techniques to develop sarcopenia classification models specifically for breast cancer survivors. Through the use of a KMR device and an in-depth analysis of joint angle data, we aimed to enhance the accuracy of sarcopenia classification and gain valuable insights into particular joint angles that have a substantial impact on sarcopenia. This approach seeks to uncover specific joint angles that considerably influence sarcopenia and provides valuable insights into effective intervention strategies. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2024-2337/rc).

Methods

Participants and data collection

This study conducted a secondary analysis of motion data collected from a previously published randomized controlled trial (27). The previous trial evaluated the effects of an 8-week KMR exercise intervention on physical function and quality of life in breast cancer survivors. Details of the original intervention design can be found in the previously published article (27). Participants in the trial were randomly assigned to either an exercise group or a control group. In the current study, we exclusively utilized data from the 38 breast cancer survivors in the exercise group of the original trial (mean age: 49.4±7.5 years). Although the number of participants was relatively small (n=38), the sample size of the original trial was determined using G*Power 3.1 (effect size =0.2, power =0.95, α=0.05), ensuring statistical adequacy. Furthermore, each participant contributed a large number of joint angle data points across the 8-week intervention period, totaling over 10,000 samples per movement type (see Table S1). This allowed for robust training of the ML models despite the limited number of participants. These participants underwent the same 8-week exercise program. Recruitment for the original trial occurred between April 2023 and June 2023, and the intervention period was conducted from July 2023 to September 2023.

Participants were recruited through local cancer-related organizations and advertisements on community radio and newspapers. Additionally, social media and referrals from the Korean Cancer Society (Seoul, Korea) were used. The inclusion criteria included stage I–III breast cancer diagnosis; completion of all radiotherapy and/or chemotherapy sessions at least 6 months prior; absence of metastatic diseases and other types of cancer; 30 min or less of physical activity per week; and absence of any neurological, musculoskeletal, metabolic, respiratory, or cardiovascular diseases or contraindications.

Before participating in the exercise program, the participants underwent body composition assessment, specifically focusing on muscle mass, using a multifrequency BIA platform (Inbody BWA 2.0, Biospace, Seoul, Korea), which has been validated as a suitable substitute for dual-energy X-ray absorptiometry (28). Participants underwent measurements in a controlled environment with stable laboratory temperature (20–25 ℃) and moderate humidity levels (40–60%). Measurements were conducted between 8 AM and noon, prior to the start of the experiment. Handgrip strength (HGS) was assessed using an electronic HGS measuring device (TKK-5401, Takei Scientific Instruments, Niigata, Japan). Three attempts were made with each hand, and the average was used. Sarcopenia stages were determined based on skeletal muscle index (SMI) of <5.7 kg/m² and HGS of <18 kg (29).

All participants completed an exercise program that consisted of three unsupervised sessions (30 min) on non-consecutive days per week for 8 weeks, for a total of 24 sessions using KMR device. During the participants’ engagement in the exercise program, the KMR device recognized 25 joints and stored the raw data in the cloud (Figure S1). We obtained data during the 8-week exercise intervention and analyzed the exercise data from weeks 3 to 8, excluding the adaptation period of weeks 1 and 2. The data included information on the duration of the exercise and the following joint movements: knee flexion (right and left), trunk flexion and extension, ankle flexion (right and left), shoulder range of motion (ROM), and shoulder flexion maxima and minima.

Then, the vectors of the thigh and shin, shin and foot, and arm and trunk (Figure 1) were computed on the basis of two different joint coordinates in three-dimensional space. The distance between a pair of joints was computed using the Euclidean distance (30), as stated in Eqs. [1-3]:

Figure 1.

Location of interest points and angles measured during exercise.

$$d_{\left(k,l\right)}=\sqrt{{\left(x_k-x_l\right)}^2+{\left(y_k-y_l\right)}^2+{\left(z_k-z_l\right)}^2}$$ [1]

$$d_{\left(l,m\right)}=\sqrt{{\left(x_l-x_m\right)}^2+{\left(y_l-y_m\right)}^2+{\left(z_l-z_m\right)}^2}$$ [2]

$$d_{\left(k,m\right)}=\sqrt{{\left(x_k-x_m\right)}^2+{\left(y_k-y_m\right)}^2+{\left(z_k-z_m\right)}^2}$$ [3]

From the computed vectors, the knee angle (q1) was calculated using the Law of Cosines, as shown in Eq. [4].

$${\theta }_1={\cos }^{-1}\left(\frac{d_{\left(i,j\right)}^2+d_{\left(j,k\right)}^2-d_{\left(i,k\right)}^2}{2d_{\left(i,j\right)}{d}_{\left(j,k\right)}}\right),0\le {\theta }_1\le \pi$$ [4]

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Seoul National University Institutional Review Board (No. 2310/002-012), and informed consent was obtained from all individual participants.

Data preprocessing

We conducted a data-processing workflow to extract meaningful insights from the exercise-related dataset. Each stage played a crucial role in ensuring the reliability and accuracy of our classification models and the subsequent analysis of sarcopenia. Figure S2 illustrates the data-processing workflow, which encompasses raw processing, data preprocessing, model learning, evaluation, and derivation of the final data-driven products.

The pre-processing steps were employed to enhance the quality of the data and ensure accurate data analysis. One or a few cycles were identified as the foundation for subsequent processing to determine the complete cycle for each exercise (Figure S3). After smoothing and cycle detection, we obtained the final features for each movement along with their respective counts per single movement repetition (Table S1). Further details on data preprocessing are provided in the Supplementary file (Appendix 1).

Model training

To identify the most suitable ML classification algorithm for predicting sarcopenia, the performances of four classification models [support vector machine (SVM), K-nearest neighbor (KNN), random forest (RF), and XGBoost (XGB)] that were primarily used in the analysis of Kinect data and older individuals were compared. The dataset was randomly split into training and testing sets in an 8:2 ratio to ensure robust model evaluation. Additionally, 5-fold cross-validation was performed on the training data to optimize the hyperparameters and evaluate the consistency of each model’s performance. Further details are provided in the Supplementary file (Appendix 1).

Model evaluation

The classification models were evaluated using a confusion matrix (CM), which includes accuracy, precision, recall (sensitivity), specificity, and F1 score. Sensitivity was calculated as the proportion of true positives over the sum of true positives and false negatives, while specificity was calculated as the proportion of true negatives over the sum of true negatives and false positives. The overall model performance was visualized using the area under the receiver operating characteristic curve (AUC). To estimate the precision of these performance metrics, 95% confidence intervals (CIs) were calculated using bootstrap resampling (1,000 iterations). Further details are provided in the Supplementary file (Appendix 1).

Statistical analyses

In the present study, data were analyzed using Python (Ver 3.10.0) programming language to develop the models.

Results

General characteristics

Overall, 38 breast cancer survivors were included in this study. The participants’ general characteristics are presented in Table S2. The participants were aged 48.9±5.4 years, with an average body weight of 58±7.3 kg. Additionally, 13 individuals had an SMI <5.7 kg/m², and 15 individuals had an HGS <18 kg. On the basis of these criteria, 12 of 38 participants (31.5%) were diagnosed with sarcopenia. Furthermore, after the 8-week exercise intervention, three participants exhibited a reversal of sarcopenia, whereas a total of nine out of 38 individuals were classified as having sarcopenia.

Model evaluation

The results of the four models, as well as the CM and AUC scores are presented in Table 1.

Table 1. Evaluation of classification models for test data.

Model	Accuracy (95% CI) (%)	Recall (95% CI) (%)	Precision (95% CI) (%)	F1 score (%)	AUC (%)
SVM	91.8 (79.2–97.3)	87.6 (72.7–94.2)	93.9 (82.7–98.5)	90.6	92.8
KNN	90.5 (75.9–95.8)	85.3 (69.6–92.6)	92.1 (79.2–97.3)	88.5	93.2
RF	92.4 (79.2–97.4)	88.9 (75.9–95.7)	94.3 (82.6–98.4)	91.5	94.7
XGB	94.7 (87.7–98.5)	91.2 (79.3–97.2)	95.8 (82.7–98.6)	93.4	96.2

95% confidence intervals were estimated using bootstrap with 1,000 iterations. AUC, area under the receiver operating characteristic curve; CI, confidence interval; KNN, K-nearest neighbor; RF, random forest; SVM, support vector machine; XGB, XGBoost.

The SVM model achieved impressive results in classifying sarcopenia with an accuracy of 91.8% (95% CI: 79.2–97.3%). This demonstrated a strong ability to accurately identify positive sarcopenia cases, as indicated by a recall of 87.6% (95% CI: 72.7–94.2%). Moreover, the model exhibited a high precision of 93.9% (95% CI: 82.7–98.5%), demonstrating its ability to minimize false positives. An F1 score of 90.6% further confirms the overall performance of the model, balancing both precision and recall. Additionally, an AUC of 92.8% highlighted the ability of the model to effectively distinguish between the sarcopenia and non-sarcopenia cases.

The KNN model demonstrated its capability to classify sarcopenia, achieving an accuracy of 90.5% (95% CI: 75.9–95.8%). With a recall of 85.3% (95% CI: 69.6–92.6%), the model demonstrated good ability to identify positive sarcopenia cases. Its precision of 92.1% (95% CI: 79.2–97.3%) signified its capability to minimize false positives. An F1 score of 88.5% further validated the balanced performance of the model. Notably, an AUC of 93.2% emphasized the effectiveness of the model in distinguishing between sarcopenia and non-sarcopenia cases.

The RF model delivered strong results with an accuracy of 92.4% (95% CI: 79.2–97.4%). It excelled in correctly identifying positive sarcopenia cases, as evidenced by a recall of 88.9% (95% CI: 75.9–95.7%). The model exhibited a high precision of 94.3% (95% CI: 82.6–98.4%), indicating its proficiency in minimizing false positives. An F1 score of 91.5% further confirmed the overall performance of the model. Notably, an AUC of 94.7% highlighted the exceptional ability of the model to differentiate between sarcopenia and non-sarcopenia cases.

These results suggest that all models (SVM, KNN, RF, and XGB) performed well in sarcopenia classification, with each model demonstrating strengths in different evaluation metrics.

However, the XGB exhibited the highest performance: accuracy, 94.7% (95% CI: 87.7–98.5%); recall, 91.2% (95% CI: 79.3–97.2%); precision, 95.8% (95% CI: 82.7–98.6%); F1 score, 93.4%; AUC, 96.2% (Figure 2). These findings confirmed that XGB achieved the best overall performance in accuracy, recall, precision, F1 score, and AUC. It exhibited the highest level of performance in accurately classifying sarcopenia.

Figure 2.

AUCs among the four classification models. AUC, area under the receiver operating characteristic curve; KNN, K-nearest neighbor; RF, random forest; SVM, support vector machine; XGB, XGBoost.

Feature importance

Table 2 shows a summary of the ranked feature-importance results from RF and XGB. Among the 12 features, the top five with the highest importance values are shown. These findings highlight the significance of various features in sarcopenia classification.

Table 2. Feature importance ranking based on different modeling approaches.

Feature name	Rank	Random forest	XGBoost	Mean
Knee flexion (right)	1	0.465	0.381	0.423
Squat time	2	0.324	0.366	0.345
Shoulder ROM	3	0.175	0.155	0.165
Knee flexion (left)	4	0.009	0.037	0.023
Squat trunk flexion	5	0.012	0.016	0.014

ROM, range of motion.

In the two models, knee flexion (right) had the highest importance value, followed by squat time and shoulder ROM. Specifically, in RF, “knee flexion (right)” showed the highest importance value at 0.465.

The second-highest ranked feature was “squat time”, which had a value of 0.324 in RF and 0.366 in XGB. The third highest importance was observed for “shoulder ROM”, with an average of 0.165 for both RF and XGB.

These results indicate a similar trend in both models; therefore, “knee flexion (right)” had the most remarkable impact on the classification model for sarcopenia.

Discussion

Key findings

This study developed a sarcopenia classification model by comprehensively analyzing Kinect data using ML techniques, potentially contributing to the development of exercise prescription strategies and personalized rehabilitation programs for patients with sarcopenia.

We employed four classification models to classify sarcopenia and demonstrated their significance in explaining the results of the classification models through variable importance extraction. This study provides insights into the interpretation of results generated by classification models through the extraction of feature importance.

Comparison with similar researches

Various studies have used ML for the sarcopenia classification (21,31,32). These studies used personal health data, abdominal CT images, or directly measured physical activity performance to classify sarcopenia (23,31,33). Previous research has also trained classification models (XGB and SVM) with high performance in classifying sarcopenia. The high performance of XGB and RF in classifying sarcopenia can be attributed to their distinctive features and advantages (33,34). Both models have added benefits of effectively handling missing data and mitigating the risk of overfitting.

However, several studies have used multidimensional data, specifically Kinect data. Abu Hassan et al. (35) evaluated correct movements using Kinect data, whereas Akbari et al. (36) used Kinect data with SVM and KNN classification models. Similarly, most studies that used multidimensional Kinect data, including ours, employed similar classification models and achieved high performance.

Previous studies that used the Kinect data followed similar procedures in terms of data collection, smoothing, and detection (36,37). However, most of these studies focused on analyzing and evaluating movements, which distinguishes them from the current research.

Feature importance is a metric that represents the relative importance of each feature in a classification model (38). By analyzing feature importance in this study, we identified features that considerably affected model performance: knee flexion (right) and squat time had the greatest impact on sarcopenia classification, indicating that they are variables for classifying sarcopenia.

Explanations of findings

The importance of knee flexion (right) aligns with findings of previous research on sarcopenia (39). Sarcopenia is characterized by a decrease in muscle mass, and reduced knee flexion can limit daily activities (40). The dominant limb often exhibits greater strength and neuromuscular coordination because it is used more frequently in daily activities and specific tasks such as exercise (41). According to the survey, only 2 of 38 participants were left-footed, whereas the rest were right-footed. Therefore, it is plausible that the dominant limb may experience greater muscle activation and subsequently exhibit more pronounced changes in muscle function. Additionally, investigating the impact of sarcopenia on both dominant and non-dominant limbs may provide valuable insights into the role of limb dominance in muscle deterioration and functional decline.

The significance of squat time implies its relationship with lower-limb strength and gait (42). A previous study has reported a correlation between lower-limb strength and gait, and squat time is commonly used to assess lower-limb strength (43). Thus, the importance of squat time emphasizes the relationship between lower-limb strength and gait, highlighting the importance of these variables in sarcopenia classification.

Sarcopenia primarily affects major muscle groups responsible for lower-limb function, including the quadriceps, hamstrings, and gluteal muscles (5). As individuals age or experience conditions such as breast cancer-related sarcopenia, the gradual loss of muscle mass and strength in these muscle groups can extensively impair mobility and balance, leading to difficulties in activities such as walking, climbing stairs, and standing from a seated position. Because knee flexion and squat time are important features, it can be inferred that lower-limb function plays a crucial role in sarcopenia classification. This allows us to prescribe exercises, such as performing deep squats, sitting on a low chair, and performing rapid squats with assistance, to effectively address the condition.

The importance of shoulder ROM can be related to limitations in shoulder mobility experienced by breast cancer survivors (44). Post-surgical changes in the tissues around the shoulder can lead to restricted shoulder mobility, which may be associated with physical inactivity (45). Therefore, the significance of shoulder ROM suggests a relationship between shoulder mobility limitations and sarcopenia classification, indicating a connection to research on physical inactivity.

Strengths and limitations

This is a relatively novel study that used Kinect data to classify sarcopenia. On the basis of these insights, exercise prescription strategies and tailored rehabilitation programs can be devised to optimize sarcopenia management with a focus on key variables such as knee flexion, squat time, and shoulder ROM. Furthermore, given that most participants in our study were right-footed, the emphasis on right knee flexion underscores the potential influence of the dominant limb dynamics on muscle function and strength, suggesting the need for targeted interventions tailored to individual limb dominance.

Despite these promising results, this study had some limitations. These limitations include the small sample size and participant representativeness, which hinder generalizability. Establishing causality in observational studies is challenging, and further experimental research is required. Additionally, variable selection was limited to 12 variables; however, expanding the selection to include more joint-related variables could improve the analysis. Further research using a wider range of variables would enhance our understanding.

Implications and actions needed

Nevertheless, this study represents a pioneering effort to use Kinect data for sarcopenia classification, advancing the understanding of this condition in healthcare research. By identifying key variables, such as knee flexion (right), squat time, and shoulder ROM, this study offers valuable insights into sarcopenia mechanisms, particularly in breast cancer survivors. These insights have immediate clinical applications, enabling healthcare providers to tailor exercise prescription strategies and rehabilitation programs based on specific movement parameters derived from Kinect data. Ultimately, this research holds promise for a more accurate and effective management of sarcopenia, offering innovative approaches to optimize healthcare delivery and improve outcomes for breast cancer survivors.

Conclusions

Among ML classification models trained on Kinectdata, XGB demonstrated the best performance. Through feature-importance analysis, we identified the features that had the greatest impact. These findings have the potential to guide interventions in breast cancer-related sarcopenia.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Seoul National University Institutional Review Board (No. 2310/002-012), and informed consent was obtained from all individual participants.

Data Sharing Statement

Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2024-2337/dss

tcr-14-07-4208-dss.pdf ^{(79.3KB, pdf)}

DOI: 10.21037/tcr-2024-2337