گروه‌بندی دوندگان براساس زوایای سه‌بعدی مفصل مچ پا با استفاده از الگوریتم خوشه‌بندی عمیق زمانی

نوع مقاله : مقاله پژوهشی

نویسندگان

1 گروه بیومکانیک ورزشی، دانشکده علوم ورزشی، دانشگاه مازندران، بابلسر، ایران.

2 واحد علوم کامپیوتر، دانشکده علوم ریاضی و آمار، دانشگاه مازندران، بابلسر، ایران.

10.32598/SJRM.14.3.3334

چکیده

مقدمه و اهداف روش‌های طبقه‌بندی ایستا باهدف بررسی ارتباط نوع و ساختار پا با عوامل آسیب‌زا و تجویز مداخلات پیشگیرانه و درمانی از قبیل ارتز و کفش استفاده شده است. باوجوداین، ارتباط ضعیفی بین اندازه‌گیری‌های ایستا و عملکرد حرکتی پا گزارش شده ‌است و مطالعات به سمت استفاده از روش‌های طبقه‌بندی پویای پا که مبتنی بر الگوی حرکتی است، هدایت شده است؛ بنابراین، هدف از این مطالعه، گروه‌بندی دونده‌ها براساس الگوی سینماتیک سه‌بعدی مچ پا در حین دویدن و بررسی تفاوت‌های معنادار الگوهای سینماتیک گروه‌های شناسایی‌شده است.
مواد و روش‌ها الگوریتم خوشه‌بندی عمیق زمانی باهدف شناسایی زیرگروه‌های همگن بر روی داده‌های زوایای سه‌بعدی مفصل مچ پا 108 بزرگسال سالم (سن: 2/42±22/45 سال،قد: 0/11±1/69 متر، توده بدن: 9/54±64/64 کیلوگرم، جنسیت: 55 مرد، 53 زن) در حالت دویدن با پای برهنه اجرا شد. پس از شناسایی گروه‌ها، نگاشت پارامتریک آماری برای بررسی تفاوت الگوهای سینماتیکی مفصل مچ پا در سراسر مرحله استقرار دویدن در گروه‌های شناسایی‌شده مورد استفاده قرار گرفت.
یافته‌ها سه زیرگروه مجزا شناسایی شدند. مقایسه سری زمانی منحنی‌ها نشان داد افراد خوشه 1 دارای میانگین دورسی فلکشن مفصل مچ پا بزرگتر در مقایسه با الگوی دو خوشه دیگر در بین 40 تا 80 درصد از مرحله استقرار دویدن بودند (0/004=P). همچنین، افراد خوشه 3 الگوی تغییرات اورشن مفصل مچ پا بیشتری را بین 60 تا 100 درصد مرحله استقرار دویدن در مقایسه با الگوی افراد دو خوشه دیگر نشان داد (0/038=P). باوجوداین، تغییرات زوایای مفصل مچ پا در صفحه افقی در حین دویدن در هر سه گروه الگوی مشابه داشتند.
نتیجه‌گیری مدل ارائه‌شده قادر است دوندگان را براساس الگوی سینماتیک مفصل مچ پا در حین دویدن به‌صورت خودکار طبقه‌بندی نماید؛ بنابراین، با شناسایی الگوی حرکتی گروه‌های همگن و تعیین مداخلات مناسب برای هریک از این گروه‌ها، می‌توان دستور‌العملی مناسب در تجویز مداخلات پیشگیرانه از قبیل کفش را توسعه داد.

کلیدواژه‌ها

موضوعات


عنوان مقاله [English]

Grouping Runners Based on Three-Dimensional Ankle Joint Angles Using a Deep Temporal Clustering Algorithm

نویسندگان [English]

  • Zaniar Mohamadi 1
  • Mansour Eslami 1
  • Rohollah Yousefpour 2
1 Department of Sports Biomechanics, Faculty of Sports Sciences, University of Mazandaran, Babolsar, Iran.
2 Unit of Computer Sciences, Department of Mathematical Sciences, University of Mazandaran, Babolsar, Iran.
چکیده [English]

Background and Aims Static classification methods have been used to investigate the relationship between foot type and structure with risk factors and to prescribe preventive and therapeutic interventions such as orthotics and shoes. However, a weak relationship between static measurements and foot movement performance has been reported, and studies have shifted towards using dynamic foot classification methods based on movement patterns. Therefore, this study aimed to group runners based on three-dimensional kinematic patterns of the ankle joint during running and to investigate significant differences in kinematic patterns among the identified groups.
Methods The deep temporal clustering algorithm was implemented during barefoot running to identify homogeneous subgroups on three-dimensional ankle joint angle data of 108 healthy adults (age: 22.45±2.42 years; height: 1.69±0.11 m; body mass: 64.66±9.54 kg; gender: 55 males, 53 females). After identifying the clusters, a parametric statistical mapping was used to examine the differences in ankle joint kinematic patterns across the stance phase of running in the identified clusters.
Results Three distinct subgroups were identified. A comparison of the time series curves showed that individuals in cluster 1 had a larger average ankle joint dorsiflexion range compared to the other two cluster patterns between 40% and 80% of the stance phase of running (P=0.004). Additionally, individuals in cluster 3 showed a greater range of ankle joint eversion between 60% and 100% of the stance phase of running compared to the patterns of individuals in the other two clusters (P=0.038). However, changes in ankle joint angles in the horizontal plane during running were similar in all three groups.
Conclusion The proposed model can automatically group runners based on the kinematic pattern of the ankle joint during running. By identifying the movement pattern of homogeneous groups and determining appropriate interventions for each of these groups, a suitable guideline for prescribing preventive interventions such as shoes can be developed.

کلیدواژه‌ها [English]

  • Running
  • Ankle joint angles
  • Functional grouping
  • Foot grouping
  • Deep temporal clustering

Introduction
Running is a popular and affordable sport, but it can lead to overuse injuries due to its repetitive nature and high participation rates [1]. Running-related injuries have an annual incidence rate ranging from 19% to 79%, with about 70% affecting the knee and lower leg regions [2]. Evaluating lower limb function involves considering foot type and functional movement, as certain foot types can increase injury risk [3]. Previous studies have mainly relied on static foot classification methods to predict foot movement function, but a weak correlation has been observed. To overcome this limitation, researchers are investigating dynamic foot classification methods that capture foot movement during activities like running [4].
Recent research has focused on identifying movement patterns in homogeneous groups using unsupervised clustering algorithms [5–7]. However, these algorithms have limitations when applied to biomechanical data due to their design for static data with small dimensions [8]. Principal component analysis is commonly used to capture data variance, but it may overlook subtle differences in movement patterns [9]. A more effective approach is the deep temporal clustering (DTC) algorithm, a deep learning method that combines dimensionality reduction and temporal clustering in an end-to-end framework. This algorithm optimizes clustering and dimensionality reduction objectives simultaneously, making it suitable for analyzing time series data [10].
This study aimed to use the DTC algorithm to group runners based on three-dimensional kinematic patterns of the ankle joint during running. We hypothesized that identifying distinct subgroups based on ankle joint kinematics during the stance phase of running would be feasible, and there are differences in kinematic patterns among the identified groups.

Methods
A total of 108 healthy participants (age: 22.45±2.42 years; height: 1.69±0.11 m; body mass: 64.66±9.54 kg; gender: 55 males, 53 females) ran barefoot at a speed of 3±0.3 m/s. Kinematic data were collected using six Simi Motion cameras from Germany, recording at 200 Hz. Three successful attempts were recorded for each participant. The stabilization phase of running was determined using vertical force reaction data from a force plate. Raw data were filtered with a fourth-order Butterworth low-pass filter at a cutoff frequency of 16 Hz, determined through residual analysis. Ankle joint angles were calculated in MATLAB software, version R2022b using the Cardan (XYZ) sequence method and normalized to 100% of the stabilization phase. Data were then transferred to Python for further analysis.
In Python software, version 3.8, the DTC algorithm was applied to cluster kinematic patterns. The process involved three stages: Dimension reduction and learning short-scale waveforms with a convolutional neural network, learning temporal relationships with a bi-directional long short-term memory (Bi-LSTM), and non-parametric clustering on the hidden representations from Bi-LSTM to identify spatiotemporal dimensions and create clusters (Figure 1).

 

The clusters were compared using one-dimensional parametric statistical mapping.
Input time series data is encoded into a latent space using a convolutional autoencoder followed by a BI-LSTM, forming a temporal autoencoder. The latent representation from the BI-LSTM is then input to a temporal clustering layer to generate cluster assignments (Figure 2).

 



​​​​​​​Results
Three distinct clusters were identified, with a silhouette index value of 0.74, indicating optimal clustering [11, 12]. Cluster 1 had the highest percentage of individuals (54%) compared to clusters 2(24%) and 3(26%). Individuals in cluster 1 exhibited larger average ankle dorsiflexion between 40% and 80% of the stance phase compared to the other clusters (P=0.004). Cluster 3 showed greater variation in ankle eversion between 60% and 100% of the stance phase (P=0.038). However, changes in ankle joint angles in the horizontal plane during running were similar across all three groups.

Conclusion
The main goal of this study was to use a DTC algorithm to categorize healthy runners based on their ankle joint movement patterns during running. Identifying three distinct clusters confirms the hypothesis that there are homogeneous subgroups with unique ankle joint movement patterns during the stance phase of running. These differences in movement patterns among the clusters could help identify individuals at risk of injury and inform the development of targeted preventive or therapeutic interventions. The second hypothesis of the research was confirmed through statistical analysis, showing significant differences in movement patterns on the sagittal and frontal planes among the identified groups. Group 1 showed a greater range of motion in the sagittal plane from 40% to 80% of the stance phase compared to the other groups, while group 3 exhibited more eversion in the final 40% of the stance phase compared to groups 1 and 2. This eversion pattern in group 3 is consistent with previous findings by Oliver et al. [1], who identified a pronation foot group characterized by increased eversion during the stance phase using the K-means clustering algorithm. Prolonged ankle pronation, indicated by increased eversion during running, is associated with injuries such as Achilles tendonitis and stress fractures. Recognizing the pronation foot movement pattern in healthy runners may be linked to their ability to control foot pronation dynamically. Research suggests that foot muscles are less activated in a static state, highlighting the importance of assessing foot anatomy under dynamic conditions to understand individual pronation control. Static measurements may only reflect the response of inactive foot structures, such as bone alignment and tendon support, to specific static loads.
Foot muscles are crucial in controlling pronation and altering the foot arch shape when bearing 50% to 100% of body weight, affecting compensatory mechanisms during activities like walking. However, the limited volume of foot muscles restricts their compensatory capacity. In this study, the peak vertical reaction force during the stance phase of running exceeded twice the body weight, surpassing the foot muscles’ control overpronation and arch shape. Kelly et al. studied the role of foot muscles in controlling pronation and altering the foot’s arch shape. They discovered that these muscles are crucial during loading up to 150% of body weight. This finding is consistent with Kelly et al.’s research, suggesting differences in running loads and the foot muscles’ inability to compensate for ankle joint pronation during running. Dynamic classification of runners’ foot types can identify injury-prone individuals based on ankle joint kinematics, supporting existing literature on running movement patterns. This study underscores the shift from static foot classification to dynamic assessment for understanding foot movement performance during activities.
Dynamic foot-type clustering through an unsupervised approach provides significant advantages over static classification methods. The model evaluates and clusters runners’ movement performance to recommend targeted preventive and therapeutic interventions, including appropriate sports shoes. By identifying similar movement patterns within subgroups using a DTC algorithm, this model reduces individual variability in response to interventions and provides optimal guidance for functional clustering and intervention prescription.

Ethical Considerations
Compliance with ethical guidelines

This study was approved by the Ethics Committee, University of Mazandaran (IR.UMZ.REC.1403.097). All ethical principles were considered in this article. The participants were informed about the study objectives and methods. They were assured of the confidentiality of their information and were free to leave the study at any time, and the research results would be available to them if desired

Funding
This study was extracted from the PhD thesis of Zaniar Mohamadi at the Department of Sports Biomechanics, University of Mazandaran. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for profit sectors.

Authors' contributions
All authors contributed equally to the conception and design of the study, data collection and analysis, interpretation of the results, and drafting of the manuscript. Each author approved the final version of the manuscript for submission.

Conflict of interest
The authors declared no conflict of interest.

Acknowledgments
The authors would like to thank all participants for their cooperation in this study.

مقدمه و اهداف
دویدن به‌عنوان یکی از رایج‌ترین فعالیت‌های ورزشی به‌دلیل دسترسی آسان و هزینه‌های پایین، به‌طور گسترده‌ای در میان مردم محبوبیت یافته است [1]. بااین‌حال، درنتیجه افزایش مشارکت و به‌دلیل ماهیت تکراری دویدن آسیب‌های ناشی از پرکاری رایج است [2]. در مطالعات اپیدمیولوژیک، نرخ بروز آسیب‌های مرتبط با دویدن 19-79 درصد در سال بوده که 70 درصد در زانو و بخش‌های تحتانی زانو گزارش شده ‌است. در بررسی عملکرد اندام تحتانی، نوع و عملکرد حرکتی پا از اهمیت ویژه‌ای برخوردار است [3]. برخی انواع پا توسط محققان به‌عنوان عوامل مستعدکننده بروز آسیب‌های اندام تحتانی شناسایی شده‌است [3-5]. این محققان با استفاده از روش‌های طبقه‌بندی ایستا به بررسی ارتباط نوع و ساختار پا با عوامل آسیب‌زا و تجویز مداخلات پیشگیرانه و درمانی ازقبیل ارتز و کفش پرداخته‌اند. در این پژوهش‌ها فرض بر این است که می‌توان عملکرد حرکتی پا را براساس اندازه‌گیری‌های ایستا پیش‌بینی کرد [3]. باوجوداین، ارتباط ضعیفی بین اندازه‌گیری‌های ایستا و عملکرد حرکتی پا گزارش شده است و افراد حاضر در این طبقه‌ها پاسخ متفاوتی به مداخلات نشان می‌دهند [6، 7]. به همین دلیل، مطالعات به سمت استفاده از روش‌های طبقه‌بندی پویای پا که عملکرد حرکتی پا را در حین اجرای تکالیف حرکتی مانند دویدن منعکس می‌سازند، هدایت شده است [6].
اخیراً پژوهش‌ها با استفاده از الگوریتم‌های خوشه‌بندی بدون نظارت به شناسایی الگوی حرکتی گروه‌های همگن پرداخته‌اند [7-9]. نقطه قوت شناسایی گروه‌های همگن استفاده از الگوریتم‌های خوشه‌بندی بدون نظارت بود [10]، بااین‌حال، استفاده از این الگوریتم‌ها با دو محدودیت‌ همراه است. اول، این الگوریتم‌ها اصولاً برای داده‌های استاتیک با ابعاد کوچک طراحی شده‌اند [11، 12]. دوم، این الگوریتم‌ها نمی‌توانند به‌خوبی با داده‌های بیومکانیکی که ماهیت سری زمانی دارند، سازگاری داشته باشند. بنابراین، استفاده از این الگوریتم‌ها برای پردازش داده‌های بیومکانیکی با چالش‌ بزرگی ابعاد داده مواجه است [13] و جهت رفع این چالش، معمولاً از روش تحلیل مؤلفه‌های اصلی استفاده می‌شود [11، 12]. به‌ این ‌ترتیب، مجموعه‌ای از مؤلفه‌های اصلی که دربرگیرنده 70-80 درصد از واریانس کل داده‌ها است، به‌عنوان ورودی الگوریتم استفاده می‌شود. باوجوداین، این روش تفاوت‌های ظریف در الگوی حرکتی را نادیده می‌گیرد [11]. برای‌مثال در یک پژوهش که گروه‌بندی عملکردی براساس سینماتیک پا و مچ پا در حین دویدن صورت گرفت، مقادیر پایین شاخص‌های اعتبار‌سنجی (سیلوئیت (0/13) و ضریب همبستگی کوفنتیک (0/38) نشان‌دهنده ناهمگنی گروه‌های شناسایی شده بود [8]. به‌طورکلی، ناتوانی روش‌های اخیر در تعیین گروه‌های همگن را می‌توان ناشی از محدودیت‌هایی از قبیل نادیده‌گرفتن پیوستگی زمانی، مقیاس‌بندی مجدد داده‌ها جهت تفسیر، نادیده‌ گرفتن روابط غیر‌خطی بین متغیرها، وابستگی به محورهای انتخاب‌شده و محدودشدن معیار‌های اندازه‌گیری تشابه و عدم تشابه دانست [11، 14-16].
به‌منظور بر‌طرف نمودن چالش‌های موجود در زمینه خوشه‌بندی داده‌های سری‌زمانی، روش یادگیری عمیق و به‌طور خاص الگوریتم خوشه‌بندی زمانی عمیق (DTC)، پیشنهاد می‌شود [17]. مزیت اصلی الگوریتم خوشه‌بندی زمانی عمیق، استفاده از خود‌رمزنگار عمیق است که اطلاعات غیرخطی و ویژگی‌های پیچیده سری‌زمانی را در یک فضای پنهان نگاشت می‌نماید [13، 17]. بعد از نگاشت داده‌ها در فضای پنهان، لایه خوشه‌بندی زمانی را می‌توان با استفاده از معیارهای مختلف اندازه‌گیری تشابه و عدم‌تشابه، جهت شناسایی الگوی حرکتی گروه‌های همگن، براساس نیاز و کاربرد، اجرا کرد. همچنین این الگوریتم امکان شناسایی موضعی رویدادها را در داده‌های سری زمانی بدون برچسب فراهم می‌کند و توضیحی (برخلاف رویکردهای جعبه سیاه) در مورد مهم‌ترین ویژگی‌های داده برای گروه‌بندی ارائه می‌دهد.
 در یک تحقیق که روش خوشه‌بندی مبتنی بر نگاشت سرتاسری داده‌ها، با روش‌های خوشه‌بندی مبتنی بر یادگیری ماشین مقایسه شد، نتایج نشان داد روش سرتاسری با میانگین نمودار مشخصه عملکرد 0/92 نسبت به روش‌های خوشه‌‌بندی مبتنی بر یادگیری ماشین با میانگین نمودار مشخصه عملکرد 0/88 برتری دارد [17]. الگوریتم خوشه‌بندی زمانی عمیق در 13 مجموعه داده سری‌زمانی متشکل از داده‌هایی با تعداد نمونه، توالی زمانی و نسبت توزیع خوشه گوناگون به‌طورکلی عملکرد بهتری نسبت به روش‌های خوشه‌بندی مبتنی بر یادگیری ماشین داشته است [19]. بنابراین الگوریتم خوشه‌بندی زمانی عمیق دارای ویژگی‌هایی است که این الگوریتم را برای خوشه‌بندی معنا‌دار و همگن، داده‌های بیومکانیکی به‌صورت خودکار و بدون هیچ‌گونه تنظیم پارامتر مناسب جلوه می‌دهد.
در تحقیقات گذشته جهت شناسایی الگوی حرکتی گروه‌های همگن از متغیر‌هایی ازقبیل فشار کف پا، مؤلفه‌های نیروی عکس‌العمل زمین، زاویه مفصل و الگوهای فعال‌سازی عضلات استفاده شده است [7-9، 18، 19]. برای‌مثال در یک پژوهش براساس زوایای سه‌بعدی مبتنی بر مدل نمودن پا به سه بخش جلویی، میانی و خلفی در حین دویدن، گروه‌بندی انجام شد [8]. یک گروه با آبداکشن جلوی پا و گروه دیگر با ترکیبی از آداکشن بخش خلفی پا و آبداکشن بخش میانی پا تعریف شد. علی‌رغم اینکه این مدل‌ها، الگوی حرکتی دقیقی از پا و مفصل مچ پا ارائه می‌نمایند، اما به‌دلیل پیچیدگی و دشواری در تفسیر خروجی به‌ندرت در خارج از محیط‌های تحقیقاتی استفاده می‌شوند [20]. پیشنهاد شده است هر مدل یا روش اندازه‌گیری، منطبق با سؤالات بالینی باشد. بررسی حرکت در صفحه عرضی و افقی مفصل مچ پا معمولاً برای شناسایی آسیب احتمالی در دویدن، طراحی و تجویز کفش پیشنهاد شده است [6، 21].
 همچنین یافته‌های پژوهش‌های اخیر نشان داد متغیر‌هایی ازقبیل حرکت مفاصل اندام تحتانی در صفحه عرضی، افقی و مؤلفه داخلی–خارجی نیروی عکس‌العمل زمین، جهت تمایز الگوی حرکتی افراد مناسب است [22]. بنابراین توسعه گروه‌بندی براساس مدل‌ نمودن یک‌بخشی پا و متغیر‌های توصیه‌شده ممکن است نتایج مفیدی را در زمینه بالینی و کاربردی ارائه دهد. برای تحلیل این نتایج از نگاشت آماری پارامتریک استفاده خواهد شد. نگاشت آماری پارامتریک یک روش مفید برای بررسی و تحلیل ویژگی‌ها و تفاوت‌های معنادار در الگوهای حرکتی با ماهیت سری زمانی است [23]. این روش براساس تئوری میدان تصادفی، قابلیت انجام آزمون‌های آماری را با حفظ پیوستگی داده‌ها فراهم می‌کند [24]. بنابراین، جهت استخراج اطلاعات مفید از الگوهای حرکتی و تفاوت‌های موجود بین گروه‌های شناسایی‌شده می‌توان از این روش بهره برد.
بنابراین، هدف از مطالعه حاضر به‌کارگیری الگوریتم خوشه‌بندی عمیق زمانی جهت گروه‌بندی دونده‌ها براساس الگوهای سینماتیک سه‌بعدی مفصل مچ پا در حین دویدن بود. ما فرض کردیم که 1. شناسایی زیرگروه‌های متمایز براساس سینماتیک سه‌بعدی مفصل مچ پا در مرحله استقرار دویدن امکان‌پذیر خواهد بود و 2. الگوی‌های سینماتیک سه بعدی مفصل مچ پا بین گروهای شناسایی‌شده تفاوت وجود دارد.

 مواد و روش‌ها
108 آزمودنی (سن: 2/42±22/45 سال، قد: 0/11±1/69 متر، جرم بدن: 9/54±64/64 کیلوگرم، جنسیت: 55 مرد، 53 زن) سالم با عدم سابقه جراحی، شکستگی، سوختگی، مشکلات عصبی -عضلانی، آسیب‌های حاد یا آسیب‌های ناشی از دویدن و ضربات جدی در اندام تحتانی در این پژوهش شرکت کردند (جدول شماره 1).

 

سالم با عدم سابقه جراحی، شکستگی، سوختگی، مشکلات عصبی-‌عضلانی، آسیب‌های حاد یا آسیب‌های ناشی از دویدن، ضربات جدی در اندام تحتانی و عدم استفاده از اندام مصنوعی در ران، زانو و مچ پا، نداشتن ناهنجاری‌های ساختاری و عملکردی مانند گودی کمر، صافی کمر، کف پای صاف و گود و همچنین زانوی ضربدری و پرانتزی، عدم استفاده از هر گونه کفی، نداشتن دیابت و بیماری‌های مربوط به اعصاب پیرامونی در 6 ماه قبل از آزمون و نداشتن هیچ‌گونه دردی در 3 ماه قبل از آزمون از شرایط ورود آزمودنی‌ها به تحقیق بود. اطلاعات مربوط به تست غربالگری توسط فیزیوتراپ متخصص کنترل شد.
یک دستگاه صفحه نیروسنج شرکت Kistler ساخت سوئیس با فرکانس نمونه‌برداری 1000 هرتز در میانه مسیر دویدن 15 متری جاسازی شد؛ به‌نحوی‌که آزمودنی‌ها قادر به تشخیص محل آن نبودند. برای کنترل سرعت آزمودنی‌ها، دو متخصص زمان عبور آزمودنی از خط 5 متری نسبت به نقطه شروع تا تماس با صفحه نیرو را با کرنومتر دستی Q&Q ساخت کشور ژاپن با دقت 0/01 ثبت می‌کردند. در صورتی که زمان به‌دست‌آمده از هر دو متخصص در دامنه قابل‌قبول (0/3±3 متر بر ثانیه) بود آن تلاش ثبت می‌شد.
 در این تحقیق، از مدل سینماتیکی دو بخشی برای مدل‌سازی ساق پا و پا به‌عنوان اجسام صلب بهره‌گیری شد که منطبق بر مدل‌های استاندارد 6 درجه آزادی است و به‌ویژه در مطالعات مرتبط با دویدن رایج است [10]. تعداد 7 نشانگر بازتابی بر روی برجستگی‌های استخوانی اپی‌کوندیل‌های داخلی و خارجی ران، قوزک‌های داخلی و خارجی، مرکز استخوان پاشنه در محل اتصال تاندون آشیل، انتهای دیستال استخوان‌های متاتارس 1 و 5 و همچنین پایه استخوان متاتارس دوم قرار داده شدند.
 همچنین یک کلاستر شامل سه نشانگر روی ساق، به‌عنوان نشانگرهای آناتومیک و رهگیری استفاده شد. داده‌های سینماتیکی با استفاده از 6 دستگاه دوربین فیلم‌برداری شرکت Simi Motion ساخت آلمان با سرعت تصویربرداری 200 هرتز جمع‌آوری شدند. برای هر آزمودنی 3 تلاش موفق ثبت شد. مرحله استقرار دویدن با استفاده از نیروی عمودی عکس‌العمل زمین حاصل از صفحه نیروسنج تعیین شد. داده‌های خام با استفاده از پالایه پایین گذر باترورث مرتبه چهارم در فرکانس برشی 16 هرتز (تعیین‌شده براساس روش تحلیل باقی‌مانده) پالایه شد (تصویر شماره 1).

 

با استفاده از مختصات، زوایای سه‌بعدی مفصل مچ پا به روش توالی (XYZ) کاردان در محیط نرم‌افزار متلب نسخه‌ 2022 محاسبه و داده‌ها ازنظر زمانی به 100 درصد مرحله استقرار دویدن هنجار شد [25]. درنهایت برای هر آزمودنی یک ماتریس به ابعاد 101×3 تشکیل و به‌عنوان ورودی به نرم‌افزار پایتون منتقل شد.
در محیط نرم‌افزار پایتون نسخه 3.8 الگوریتم خوشه‌بندی عمیق زمانی با هدف خوشه‌بندی الگوهای سینماتیک اجرا شد. بدین ترتیب رویکرد سه‌مرحله‌ای: مرحله اول با پیاده‌سازی شبکه عصبی پیچشی، ابعاد داده‌ها را کاهش می‌دهد و شکل موج‌های مقیاس کوتاه را می‌آموزد. مرحله دوم که به‌عنوان حافظه بلندمدت‌-کوتاه‌مدت دوطرفه (Bi-LSTM) پیاده‌سازی شده است، ابعاد داده را کاهش و ارتباطات زمانی بین منحنی‌ها را در تمام ‌مقیاس‌های زمانی یاد می‌گیرد. مرحله سوم، خوشه‌بندی ناپارامتری نمایش‌های نهفته (Bi-LSTM) را انجام می‌دهد و یک یا چند بعد مکانی-زمانی را شناسایی می‌کند که در طول آن داده‌ها به دو یا چند خوشه تقسیم می‌شوند (تصویر شماره 2) [17].

 

به این ترتیب خوشه‌بندی داده‌ها انجام شد. مقایسه الگوهای سینماتیک خوشه‌های شناسایی‌شده با استفاده از تحلیل واریانس اندازه‌گیری مکرر که در پکیج نگاشت آماری یک بعدی تعبیه شده است، انجام شد (سطح معناداری 05/0 ). به‌منظور بررسی تعقیبی و مقایسه دوبه‌دو الگوی خوشه‌های شناسایی‌شده از آزمون تی چندگانه (سطح معناداری اصلاح‌شده بونفرونی 0/017) استفاده شد [24]. 

یافته‌ها
پس از اجرای الگوریتم خوشه‌بندی زمانی عمیق بر روی داده‌های سینماتیک سه‌بعدی مفصل مچ پا، سه خوشه مجزا شناسایی شد. اعتبار خوشه‌بندی با استفاده از شاخص سیلوئیت محاسبه و مقدار 0/74 به دست آمد [26]. این مقدار نشان‌دهنده آن است که مدل با شناسایی الگوی سینماتیک همگن، خوشه‌بندی بهینه‌ای را اجرا کرده است [8، 26، 27]. درصد فراوانی افراد خوشه 1 54 درصد) در مقایسه با دو خوشه‌ دیگر (24 و 26درصد) بالاتر بوده است (تصویر شماره 3).

 

مقایسه سری زمانی منحنی نشان داد افراد خوشه 1 دارای میانگین دورسی فلکشن مفصل مچ پا بزرگتر در مقایسه با الگوی دو خوشه دیگر در فاز بین 40 و 80 درصد از مرحله استقرار دویدن بود (0/004=P). همچنین، افراد خوشه 3 الگوی تغییرات اورشن مفصل مچ پا بیشتری را بین 60 تا 100 درصد مرحله استقرار دویدن در مقایسه با الگوی افراد دو خوشه دیگر نشان داد (0/038=P) (تصویر شماره 4).

 

باوجوداین، تغییرات زوایای مفصل مچ پا در صفحه افقی در حین دویدن در هر سه گروه الگوی مشابه داشتند.

بحث
هدف اصلی این پژوهش به‌کارگیری الگوریتم خوشه‌بندی زمانی عمیق برای گروه‌بندی دونده‌های سالم براساس الگوهای حرکتی مفصل مچ پا در حین دویدن بود. شناسایی سه خوشه متمایز، فرضیه اول مبنی بر وجود زیرگروه‌های همگن الگوی حرکتی مفصل مچ پا در مرحله استقرار دویدن را تأیید می‌کند. گروه 1 دامنه حرکتی بیشتری در صفحه حرکتی ساجیتال نسبت به گروه 2 و 3 نشان داد، درحالی‌که گروه 3 اورشن بیشتری را نسبت به دو گروه دیگر داشت. تفاوت‌های معنادار در الگوهای حرکتی سه گروه، به شناسایی افراد در معرض خطر آسیب کمک می‌کند و می‌تواند به توسعه دستورالعمل‌های مناسب برای تجویز مداخلات پیشگیرانه یا درمانی منجر شود. 
مطابق با فرضیه دوم پژوهش، نتایج ارزیابی آماری نشان داد الگوی حرکتی گروه‌های شناسایی‌شده در صفحه حرکتی ساجیتال و فرونتال تفاوت معنادار وجود دارد. گروه 1 در مقایسه با گروه‌های دیگر دارای دامنه حرکتی بیشتری در صفحه حرکتی ساجیتال در 40 تا 80 درصد مرحله استقرار بود، درحالی‌که گروه 3 اورشن بیشتری را در 40 درصد پایانی مرحله استقرار دویدن نسبت به گروه 1 و 2 نشان داد. الگوی حرکتی مشاهده‌شده در گروه 3  بسیار شبیه به نتایج اولیور و همکاران است [10].
در این پژوهش با استفاده از الگوریتم خوشه‌بندی K means گروهی از دونده‌ها که اورشن بیشتری در مرحله استقرار داشتند به‌عنوان پای پرونیت معرفی شدند [10]. اورشن بیشتر مفصل مچ پا در حین دویدن نشان‌دهنده پرونیشن مفصل مچ پا است که با بروز آسیب‌هایی از قبیل تاندونوپاتی آشیل و شکستگی استرسی همراه است [10، 28]. شناسایی الگوی حرکتی پای پرونیت در دونده‌های سالم ممکن است با توانایی کنترل پویای پرونیشن پا در افراد مختلف مرتبط باشد [28]. نشان داده شده است عضلات کف پا در حالت ایستا سطوح پایینی از فعال شدن دارند [29]. بنابراین اندازه‌گیری آناتومیک پا در حالت ایستا و بدون کنترل عضلانی کافی صرفاً پاسخ ساختارهای غیرفعال پا ازجمله هم‌ترازی استخوانی و حمایت تاندونی را به یک حالت خاص از اعمال بار ایستا نشان می‌دهد [29]. 
عضلات کف‌پایی توانایی کنترل پرونیشن و تغییر شکل قوس کف پا در حین بارگیری با 50 تا 100 درصد وزن بدن را دارند [29، 30]. بنابراین ممکن است در حین اجرای تکالیف حرکتی مانند راه‌رفتن مکانیسم جبران عضلانی تأثیرگذار باشد. بااین‌حال، ظرفیت مکانیسم‌های جبران عضلانی به‌دلیل حجم نسبتاً کم عضلات کف پا محدود است [30]. در پژوهش حاضر، اوج نیروی عکس‌العمل عمودی در مرحله استقرار دویدن بیش از 2 برابر وزن بدن ثبت شده است که این مقدار بار فراتر از توانایی عضلات کف پا برای کنترل پرونیشن و تغییر شکل قوس پا است. کیلی و همکاران در تحقیق خود به بررسی توانایی عضلات کف پایی در کنترل پرونیشن و تغییر شکل قوس کف پا پرداخته و گزارش کردند که این عضلات در حین بارگیری تا 150 درصد وزن بدن، تأثیر قابل‌ملاحظه‌ای دارند [29]. به‌این‌ترتیب، همسو با نتایج کیلی و همکاران، شناسایی گروهی از افراد با الگوی حرکتی پای پرونیت می‌تواند ناشی از اختلاف بارهای اعمال‌شده در حین دویدن و عدم تأثیرگذاری مکانیسم‌های جبران عضلانی در کنترل حرکت پرونیشن مفصل مچ پا در حین دویدن باشد [29، 30]. باتوجه‌به این یافته‌ها می‌توان ادعا کرد، گروه‌بندی پویای نوع پای دونده‌ها، قادر است دوندگان در معرض آسیب را براساس الگوی سینماتیک مفصل مچ پا به‌خوبی تمیز دهد. یافته‌های این پژوهش تطابق خوبی با ادبیات موجود در زمینه بررسی الگوهای حرکتی در حین دویدن دارد [10، 28] و بر اهمیت گذر از طبقه‌بندی ایستای پا برای مطالعات مرتبط با عملکرد حرکتی پا در هنگام اجرای تکالیف حرکتی تأکید می‌کند.
این مطالعه با استفاده از الگوریتم خوشه‌بندی زمانی عمیق، به بررسی و گروه‌بندی شرکت‌کنندگان براساس الگوهای حرکتی همگن می‌پردازد. برخلاف روش‌های سنتی که عموماً از تحلیل مؤلفه‌های اصلی به‌منظور کاهش ابعاد داده‌ها استفاده می‌کنند، این الگوریتم به‌طور کامل تمامی داده‌های مرحله استقرار دویدن را مدنظر قرار می‌دهد تا شباهت‌ها میان الگوهای حرکتی را مشخص کند. بنابراین با لحاظ تفاوت‌های ظریف الگوهای حرکتی، این رویکرد قادر است تحلیل دقیق‌تری از الگوهای حرکت دویدن ارائه دهد و به گروه‌بندی عملکردی بهینه دست یابد. همچنین ترکیب کاهش ابعاد و خوشه‌بندی زمانی در یک چارچوب یادگیری سرتاسری بدون نظارت، این الگوریتم را به ابزاری مؤثر برای شناسایی الگوهای حرکتی همگن تبدیل می‌کند. بدین ترتیب، رویکرد ارائه‌شده، روشی بهبودیافته را برای پژوهشگران فراهم می‌آورد تا با دقت و کارایی بیشتر، الگوهای حرکتی را تجزیه‌وتحلیل و گروه‌بندی کنند.
استفاده از گروه‌بندی پویای نوع پا مبتنی بر رویکرد خوشه‌بندی بدون نظارت به‌جای روش‌های طبقه‌بندی سنتی ایستا، مزایای قابل‌توجهی را به همراه دارد. این مدل گروه‌بندی، با ارزیابی و گروه‌بندی مبتنی بر عملکرد حرکتی دونده‌ها در حین دویدن، به تجویز هدفمند مداخلات پیشگیرانه و درمانی مانند کفش ورزشی کمک می‌کند. همچنین روش ارائه‌شده در این پژوهش، زیرگروه‌های همگن الگوی حرکتی را براساس رویکرد خوشه‌بندی شناسایی می‌کند. این امر نه‌تنها چالش تفاوت‌های فردی را در پاسخ به مداخلات درمانی و پیشگیرانه کاهش می‌دهد، بلکه دستورالعمل‌ بهینه‌ای برای گروه‌بندی عملکردی و تجویز مداخلات فراهم می‌آورد.
پژوهش حاضر همانند سایر پژوهش‌ها عاری از محدودیت نبود. باوجوداینکه عوامل مختلف تأثیرگذار بر شناسایی الگوی حرکتی گروه‌های همگن در پژوهش حاضر کنترل شد. سینماتیک دویدن در آزمودنی‌ها با سطوح مهارتی متنوع دویدن متفاوت است و ممکن است گروه‌های شناسایی‌شده نماینده همه دوندگان نباشند [31]. بنابراین ضروری است این مدل بر روی نمونه بزرگتری از دونده‌ها با سطوح مهارتی متنوع آزمایش و تأیید شود. همچنین بررسی پاسخ گروه‌های شناسایی‌شده به مداخلات درمانی و پیشگیرانه می‌تواند این مدل را توسعه بخشد.

نتیجه‌گیری
این پژوهش مدل گروه‌بندی پویای نوع پا را براساس سینماتیک مفصل مچ پا در حین دویدن پیشنهاد می‌کند. این رویکرد محدودیت‌های ناشی از پیش‌بینی حرکت را در گروه‌بندی ایستای نوع پا بر طرف می‌نماید. نتایج نشان می‌دهد گروه‌های شناسایی‌شده دارای الگوی حرکتی متمایز با تفاوت معنادار در صفحه فرونتال و ساجیتال هستند. بنابراین، با شناسایی و تعیین مداخلات مناسب برای هریک از این گروه‌ها، می‌توان دستور‌العملی مناسب در تجویز مداخلات پیشگیرانه و درمانی از قبیل کفش را توسعه داد.

ملاحظات اخلاقی
پیروی از اصول اخلاق پژوهش

در اجرای پژوهش ملاحظات اخلاقی مطابق با دستورالعمل کمیته اخلاق دانشگاه مازندران در نظر گرفته شده و کد اخلاق به شماره (IR.UMZ.REC.1403.097) دریافت شده است. تمام اصول اخلاقی در این مقاله رعایت شده است. اهداف و روش‌های مطالعه به شرکت‌کنندگان اطلاع داده شد. به آن‌ها در مورد محرمانه بودن اطلاعاتشان اطمینان داده شد و آن‌ها آزاد بودند در هر زمان از مطالعه خارج شوند و در صورت تمایل، نتایج تحقیق در اختیار آن‌ها قرار خواهد گرفت.

حامی مالی
این مقاله برگرفته از رساله دکتری زانیار محمدی گروه بیومکانیک ورزشی دانشگاه مازندران می‌باشد و هیچ‌گونه کمک مالی از سازمان‌های دولتی، خصوصی و غیرانتفاعی دریافت نکرده است.

مشارکت نویسندگان
همه نویسندگان به‌طور یکسان در مفهوم و طراحی مطالعه، جمع‌آوری و تجزیه‌وتحلیل داده‌ها، تفسیر نتایج و تهیه پیش‌نویس مقاله مشارکت داشتند.

تعارض منافع
بنابر اظهار نویسندگان، این مقاله تعارض منافع ندارد. 

تشکر و قدردانی
از کلیه شرکت‌کنندگان گرامی، مسئولین و تمامی افرادی که در اجرای این پژوهش همکاری داشتند، تشکر می‌شود.

 

References

  1. Hamill J, Palmer C, Van Emmerik RE. Coordinative variability and overuse injury. Sports Medicine, Arthroscopy, Rehabilitation, Therapy & Technology: SMARTT. 2012; 4(1):45. [DOI:10.1186/1758-2555-4-45] [PMID]
  2. Hreljac A. Etiology, prevention, and early intervention of overuse injuries in runners: A biomechanical perspective. Physical Medicine and Rehabilitation Clinics of North America. 2005; 16(3):651-67, vi. [DOI:10.1016/j.pmr.2005.02.002][PMID]
  3. Razeghi M, Batt ME. Foot type classification: A critical review of current methods. Gait & Posture. 2002; 15(3):282-91. [DOI:10.1016/S0966-6362(01)00151-5][PMID]
  4. Cornwall MW, McPoil TG. Influence of rearfoot postural alignment on rearfoot motion during w The Foot. 2004; 14(3):133-8. [Link]
  5. Behling AV, Nigg BM. Relationships between the foot posture Index and static as well as dynamic rear foot and arch variables. Journal of Biomechanics. 2020; 98:109448. [DOI:10.1016/j.jbiomech.2019.109448][PMID]
  6. Nigg BM. Biomechanics of Sport Shoes. Calgary: University of Calgary;2010.[Link]
  7. Senevirathna AM, Pohl AJ, Jordan MJ, Edwards WB, Ferber R. Differences in kinetic variables between injured and uninjured rearfoot runners: A hierarchical cluster analysis. Scandinavian Journal of Medicine & Science in Sports. 2023; 33(2):160-8. [DOI:10.1111/sms.14249][PMID]
  8. Watari R, Suda EY, Santos JPS, Matias AB, Taddei UT, Sacco ICN. Subgroups of foot-ankle movement patterns can influence the responsiveness to a foot-core exercise program: A hierarchical cluster analysis. Frontiers in Bioengineering and Biotechnology. 2021; 9:645710. [DOI:10.3389/fbioe.2021.645710][PMID]
  9. Watari R, Kobsar D, Phinyomark A, Osis S, Ferber R. Determination of patellofemoral pain sub-groups and development of a method for predicting treatment outcome using running gait kinematics. Clinical Biomechanics (Bristol, Avon). 2016; 38:13-21. [DOI:10.1016/j.clinbiomech.2016.08.003][PMID]
  10. Chalmers O, Page R, Langley B. A step towards dynamic foot classification: Functional grouping using ankle joint frontal plane motion in running. Gait & Posture. 2022; 97:35-9. [DOI:10.1016/j.gaitpost.2022.07.005][PMID]
  11. Phinyomark A, Hettinga BA, Osis S, Ferber R. Do intermediate- and higher-order principal components contain useful information to detect subtle changes in lower extremity biomechanics during running? Human Movement Science. 2015; 44:91-101. [DOI:10.1016/j.2015.08.018][PMID]
  12. Jolliffe IT, Cadima J. Principal component analysis: A review and recent developments. Philosophical Transactions. Series A, Mathematical, Physical, and Engineering Sciences. 2016; 374(2065):20150202. [DOI:10.1098/rsta.2015.0202][PMID]
  13. Alqahtani A, Ali M, Xie X, Jones MW. Deep time-series clustering: A review. Electronics. 2021; 10(23):3001. [Link]
  14. Abdi H, Williams LJ. Principal component analysis. Wiley interdisciplinary reviews: Computational statistics. 2010; 2(4):433-59. [Link]
  15. Ghojogh B, Crowley M, Karray F, Ghodsi A. Principal component analysis. In: Elements of dimensionality reduction and manifold learning. Cham: Springer; 2023. [DOI:10.1007/978-3-031-10602-6_5]
  16. Aghabozorgi S, Shirkhorshidi AS, Wah TY. Time-series clustering–A decade review. Information Systems. 2015; 53:16-38. [Link]
  17. Sai Madiraju N, Sadat SM, Fisher D, Karimabadi H. Deep temporal clustering: Fully unsupervised learning of time-domain features. arXiv e-prints. 2018 Feb:arXiv-1802. [Link]
  18. De Cock A, Willems T, Witvrouw E, Vanrenterghem J, De Clercq D. A functional foot type classification with cluster analysis based on plantar pressure distribution during jogging. Gait & Posture. 2006; 23(3):339-47. [DOI:10.1016/j.gaitpost.2005.04.011][PMID]
  19. Turner A, Hayes S. The classification of minor gait alterations using wearable sensors and deep learning. IEEE Transactions on Bio-Medical Engineering. 2019; 66(11):3136-45. [DOI:10.1109/TBME.2019.2900863][PMID]
  20. Leardini A, Caravaggi P, Theologis T, Stebbins J. Multi-segment foot models and their use in clinical populations. Gait & Posture. 2019; 69:50-9. [DOI:10.1016/j.gaitpost.2019.01.022][PMID]
  21. Malisoux L, Theisen D. Can the "Appropriate" footwear prevent injury in leisure-time running? Evidence Versus Beliefs. Journal of Athletic Training. 2020; 55(12):1215-23.[DOI:10.4085/1062-6050-523-19][PMID]
  22. Hoitz F, von Tscharner V, Baltich J, Nigg BM. Individuality decoded by running patterns: Movement characteristics that determine the uniqueness of human running. Plos On 2021; 16(4):e0249657. [DOI:10.1371/journal.pone.0249657][PMID]
  23. Pataky TC, Mu T, Bosch K, Rosenbaum D, Goulermas JY. Gait recognition: Highly unique dynamic plantar pressure patterns among 104 individuals. Journal of The Royal Society, Interface. 2012; 9(69):790-800. [DOI:10.1098/rsif.2011.0430][PMID]
  24. Pataky TC, Robinson MA, Vanrenterghem J. Vector field statistical analysis of kinematic and force trajectories. Journal of Biomechanics. 2013; 46(14):2394-401. [DOI:10.1016/j.jbiomech.2013.07.031][PMID]
  25. Winter DA. Biomechanics and motor control of human movement. New Jersey: John Wiley & Sons; 2009. [Link]
  26. Dudek A. Silhouette Index as Clustering Evaluation Tool. In: Jajuga K, Batóg J, Walesiak M, editors. Classification and Data Analysis. SKAD 2019. Studies in Classification, Data Analysis, and Knowledge Organization. Cham: Springer; 2020. [DOI:10.1007/978-3-030-52348-0_2]
  27. Tomašev N, Radovanović Clustering Evaluation in High-Dimensional Data. In: Celebi M, Aydin K, editors. Unsupervised Learning Algorithms. Cham: Springer; 2016. [DOI:10.1007/978-3-319-24211-8_4]
  28. Zhang X, Yang F, Zhao K, Vanwanseele B. Symptomatic and asymptomatic pronated feet show differences in the forefoot abduction motion during jogging, but not in the arch deformation. Sports Biom 2024; 23(12):2521-32. [DOI:10.1080/14763141.2022.2109506][PMID]
  29. Kelly LA, Cresswell AG, Racinais S, Whiteley R, Lichtwark G. Intrinsic foot muscles have the capacity to control deformation of the longitudinal arch. Journal of the Royal Society, 2014; 11(93):20131188. [DOI:10.1098/rsif.2013.1188][PMID]
  30. Kelly LA, Lichtwark G, Cresswell AG. Active regulation of longitudinal arch compression and recoil during walking and running. Journal of The Royal Society, Interface. 2015; 12(102):20 [DOI:10.1098/rsif.2014.1076][PMID]
  31. Quan W, Zhou H, Xu D, Li S, Baker JS, Gu Y. Competitive and recreational running kinematics examined using principal components analysis. Healthcare (Basel). 2021; 9(10):1321 [DOI:10.3390/healthcare9101321][PMID]
  1. Hamill J, Palmer C, Van Emmerik RE. Coordinative variability and overuse injury. Sports Medicine, Arthroscopy, Rehabilitation, Therapy & Technology: SMARTT. 2012; 4(1):45. [DOI:10.1186/1758-2555-4-45] [PMID]
  2. Hreljac A. Etiology, prevention, and early intervention of overuse injuries in runners: A biomechanical perspective. Physical Medicine and Rehabilitation Clinics of North America. 2005; 16(3):651-67, vi. [DOI:10.1016/j.pmr.2005.02.002][PMID]
  3. Razeghi M, Batt ME. Foot type classification: A critical review of current methods. Gait & Posture. 2002; 15(3):282-91. [DOI:10.1016/S0966-6362(01)00151-5][PMID]
  4. Cornwall MW, McPoil TG. Influence of rearfoot postural alignment on rearfoot motion during w The Foot. 2004; 14(3):133-8. [Link]
  5. Behling AV, Nigg BM. Relationships between the foot posture Index and static as well as dynamic rear foot and arch variables. Journal of Biomechanics. 2020; 98:109448. [DOI:10.1016/j.jbiomech.2019.109448][PMID]
  6. Nigg BM. Biomechanics of Sport Shoes. Calgary: University of Calgary;2010.[Link]
  7. Senevirathna AM, Pohl AJ, Jordan MJ, Edwards WB, Ferber R. Differences in kinetic variables between injured and uninjured rearfoot runners: A hierarchical cluster analysis. Scandinavian Journal of Medicine & Science in Sports. 2023; 33(2):160-8. [DOI:10.1111/sms.14249][PMID]
  8. Watari R, Suda EY, Santos JPS, Matias AB, Taddei UT, Sacco ICN. Subgroups of foot-ankle movement patterns can influence the responsiveness to a foot-core exercise program: A hierarchical cluster analysis. Frontiers in Bioengineering and Biotechnology. 2021; 9:645710. [DOI:10.3389/fbioe.2021.645710][PMID]
  9. Watari R, Kobsar D, Phinyomark A, Osis S, Ferber R. Determination of patellofemoral pain sub-groups and development of a method for predicting treatment outcome using running gait kinematics. Clinical Biomechanics (Bristol, Avon). 2016; 38:13-21. [DOI:10.1016/j.clinbiomech.2016.08.003][PMID]
  10. Chalmers O, Page R, Langley B. A step towards dynamic foot classification: Functional grouping using ankle joint frontal plane motion in running. Gait & Posture. 2022; 97:35-9. [DOI:10.1016/j.gaitpost.2022.07.005][PMID]
  11. Phinyomark A, Hettinga BA, Osis S, Ferber R. Do intermediate- and higher-order principal components contain useful information to detect subtle changes in lower extremity biomechanics during running? Human Movement Science. 2015; 44:91-101. [DOI:10.1016/j.2015.08.018][PMID]
  12. Jolliffe IT, Cadima J. Principal component analysis: A review and recent developments. Philosophical Transactions. Series A, Mathematical, Physical, and Engineering Sciences. 2016; 374(2065):20150202. [DOI:10.1098/rsta.2015.0202][PMID]
  13. Alqahtani A, Ali M, Xie X, Jones MW. Deep time-series clustering: A review. Electronics. 2021; 10(23):3001. [Link]
  14. Abdi H, Williams LJ. Principal component analysis. Wiley interdisciplinary reviews: Computational statistics. 2010; 2(4):433-59. [Link]
  15. Ghojogh B, Crowley M, Karray F, Ghodsi A. Principal component analysis. In: Elements of dimensionality reduction and manifold learning. Cham: Springer; 2023. [DOI:10.1007/978-3-031-10602-6_5]
  16. Aghabozorgi S, Shirkhorshidi AS, Wah TY. Time-series clustering–A decade review. Information Systems. 2015; 53:16-38. [Link]
  17. Sai Madiraju N, Sadat SM, Fisher D, Karimabadi H. Deep temporal clustering: Fully unsupervised learning of time-domain features. arXiv e-prints. 2018 Feb:arXiv-1802. [Link]
  18. De Cock A, Willems T, Witvrouw E, Vanrenterghem J, De Clercq D. A functional foot type classification with cluster analysis based on plantar pressure distribution during jogging. Gait & Posture. 2006; 23(3):339-47. [DOI:10.1016/j.gaitpost.2005.04.011][PMID]
  19. Turner A, Hayes S. The classification of minor gait alterations using wearable sensors and deep learning. IEEE Transactions on Bio-Medical Engineering. 2019; 66(11):3136-45. [DOI:10.1109/TBME.2019.2900863][PMID]
  20. Leardini A, Caravaggi P, Theologis T, Stebbins J. Multi-segment foot models and their use in clinical populations. Gait & Posture. 2019; 69:50-9. [DOI:10.1016/j.gaitpost.2019.01.022][PMID]
  21. Malisoux L, Theisen D. Can the "Appropriate" footwear prevent injury in leisure-time running? Evidence Versus Beliefs. Journal of Athletic Training. 2020; 55(12):1215-23.[DOI:10.4085/1062-6050-523-19][PMID]
  22. Hoitz F, von Tscharner V, Baltich J, Nigg BM. Individuality decoded by running patterns: Movement characteristics that determine the uniqueness of human running. Plos On 2021; 16(4):e0249657. [DOI:10.1371/journal.pone.0249657][PMID]
  23. Pataky TC, Mu T, Bosch K, Rosenbaum D, Goulermas JY. Gait recognition: Highly unique dynamic plantar pressure patterns among 104 individuals. Journal of The Royal Society, Interface. 2012; 9(69):790-800. [DOI:10.1098/rsif.2011.0430][PMID]
  24. Pataky TC, Robinson MA, Vanrenterghem J. Vector field statistical analysis of kinematic and force trajectories. Journal of Biomechanics. 2013; 46(14):2394-401. [DOI:10.1016/j.jbiomech.2013.07.031][PMID]
  25. Winter DA. Biomechanics and motor control of human movement. New Jersey: John Wiley & Sons; 2009. [Link]
  26. Dudek A. Silhouette Index as Clustering Evaluation Tool. In: Jajuga K, Batóg J, Walesiak M, editors. Classification and Data Analysis. SKAD 2019. Studies in Classification, Data Analysis, and Knowledge Organization. Cham: Springer; 2020. [DOI:10.1007/978-3-030-52348-0_2]
  27. Tomašev N, Radovanović Clustering Evaluation in High-Dimensional Data. In: Celebi M, Aydin K, editors. Unsupervised Learning Algorithms. Cham: Springer; 2016. [DOI:10.1007/978-3-319-24211-8_4]
  28. Zhang X, Yang F, Zhao K, Vanwanseele B. Symptomatic and asymptomatic pronated feet show differences in the forefoot abduction motion during jogging, but not in the arch deformation. Sports Biom 2024; 23(12):2521-32. [DOI:10.1080/14763141.2022.2109506][PMID]
  29. Kelly LA, Cresswell AG, Racinais S, Whiteley R, Lichtwark G. Intrinsic foot muscles have the capacity to control deformation of the longitudinal arch. Journal of the Royal Society, 2014; 11(93):20131188. [DOI:10.1098/rsif.2013.1188][PMID]
  30. Kelly LA, Lichtwark G, Cresswell AG. Active regulation of longitudinal arch compression and recoil during walking and running. Journal of The Royal Society, Interface. 2015; 12(102):20 [DOI:10.1098/rsif.2014.1076][PMID]
  31. Quan W, Zhou H, Xu D, Li S, Baker JS, Gu Y. Competitive and recreational running kinematics examined using principal components analysis. Healthcare (Basel). 2021; 9(10):1321 [DOI:10.3390/healthcare9101321][PMID]
دوره 14، شماره 3
مرداد و شهریور 1404
صفحه 418-433
  • تاریخ دریافت: 01 بهمن 1403
  • تاریخ بازنگری: 09 بهمن 1403
  • تاریخ پذیرش: 16 فروردین 1404
  • تاریخ اولین انتشار: 16 فروردین 1404