서브메뉴
검색
Integrating Continuous IMU Monitoring With Transformer-Based Variational Autoencoders to Predict Cartilage Health and Clinical Outcomes in Knee Osteoarthritis
Integrating Continuous IMU Monitoring With Transformer-Based Variational Autoencoders to Predict Cartilage Health and Clinical Outcomes in Knee Osteoarthritis
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202103502
- ISBN
- 9798280729841
- DDC
- 610
- 서명/저자
- Integrating Continuous IMU Monitoring With Transformer-Based Variational Autoencoders to Predict Cartilage Health and Clinical Outcomes in Knee Osteoarthritis
- 발행사항
- [Sl] : University of California, San Francisco, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 77 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-12, Section: B.
- 주기사항
- Advisor: Souza, Richard.
- 학위논문주기
- Thesis (Ph.D.)--University of California, San Francisco, 2025.
- 초록/해제
- 요약Knee osteoarthritis (KOA) is a major contributor to disability worldwide, with rising prevalence driven by changes in lifestyle and physical activity patterns. It is a multifactorial disease, arising from a complex interplay of genetic, structural, biochemical and biomechanical factors. This dissertation centers on physical activity as a key modifiable factor with potential for intervention. While movement is essential for joint health, both excessive and insufficient knee joint loading can accelerate disease progression. Existing lab-based tools to estimate joint loading are impractical for everyday use, and consumer activity trackers offer limited biomechanical insight. To address this, we present a novel framework that uses a single thigh-mounted 6-axis inertial measurement unit (IMU) and a transformer-based variational autoencoder (VAE) to continuously monitor knee joint loading in natural conditions. In this study, individuals with patellofemoral KOA wore an IMU for one week. The recorded acceleration and gyroscope signals were segmented into 1.5-second windows with 50% overlap and used as input to a VAE model. The 32-dimensional latent space of the VAE was designed to encode biomechanically meaningful features of daily activity. We analyzed the distribution of latent representations across participants and quantified their time spent in distinct activity clusters. These loading profiles were then compared to quantitative magnetic resonance imaging (qMRI) metrics of knee cartilage health, including T1ρ and T2 relaxation times. Our results suggest that specific activity patterns encoded in the latent space are associated with healthier cartilage composition. This approach offers a scalable, non-invasive method for assessing mechanical loading and activity exposure in individuals with KOA. It opens the door for personalized recommendations that balance mobility with joint protection to potentially delay disease progression.
- 일반주제명
- Bioengineering
- 일반주제명
- Medical imaging
- 일반주제명
- Biomechanics
- 기타저자
- University of California, San Francisco Bioengineering
- 기본자료저록
- Dissertations Abstracts International. 86-12B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017357372
■00520260202103502
■006m o d
■007cr#unu||||||||
■020 ▼a9798280729841
■035 ▼a(MiAaPQ)AAI32001521
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a610
■1001 ▼aCarbajal Mendez, Hector Abramn.▼0(orcid)0000-0001-9724-5140
■24510▼aIntegrating Continuous IMU Monitoring With Transformer-Based Variational Autoencoders to Predict Cartilage Health and Clinical Outcomes in Knee Osteoarthritis
■260 ▼a[Sl]▼bUniversity of California, San Francisco▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a77 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-12, Section: B.
■500 ▼aAdvisor: Souza, Richard.
■5021 ▼aThesis (Ph.D.)--University of California, San Francisco, 2025.
■520 ▼aKnee osteoarthritis (KOA) is a major contributor to disability worldwide, with rising prevalence driven by changes in lifestyle and physical activity patterns. It is a multifactorial disease, arising from a complex interplay of genetic, structural, biochemical and biomechanical factors. This dissertation centers on physical activity as a key modifiable factor with potential for intervention. While movement is essential for joint health, both excessive and insufficient knee joint loading can accelerate disease progression. Existing lab-based tools to estimate joint loading are impractical for everyday use, and consumer activity trackers offer limited biomechanical insight. To address this, we present a novel framework that uses a single thigh-mounted 6-axis inertial measurement unit (IMU) and a transformer-based variational autoencoder (VAE) to continuously monitor knee joint loading in natural conditions. In this study, individuals with patellofemoral KOA wore an IMU for one week. The recorded acceleration and gyroscope signals were segmented into 1.5-second windows with 50% overlap and used as input to a VAE model. The 32-dimensional latent space of the VAE was designed to encode biomechanically meaningful features of daily activity. We analyzed the distribution of latent representations across participants and quantified their time spent in distinct activity clusters. These loading profiles were then compared to quantitative magnetic resonance imaging (qMRI) metrics of knee cartilage health, including T1ρ and T2 relaxation times. Our results suggest that specific activity patterns encoded in the latent space are associated with healthier cartilage composition. This approach offers a scalable, non-invasive method for assessing mechanical loading and activity exposure in individuals with KOA. It opens the door for personalized recommendations that balance mobility with joint protection to potentially delay disease progression.
■590 ▼aSchool code: 0034.
■650 4▼aBioengineering
■650 4▼aMedical imaging
■650 4▼aBiomechanics
■653 ▼aKnee osteoarthritis
■653 ▼aVariational autoencoder
■653 ▼aMagnetic resonance imaging
■653 ▼aInertial measurement unit
■690 ▼a0202
■690 ▼a0574
■690 ▼a0648
■71020▼aUniversity of California, San Francisco▼bBioengineering.
■7730 ▼tDissertations Abstracts International▼g86-12B.
■790 ▼a0034
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17357372▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
