Comparative Performance Evaluation of Federated and Centralized Learning for Velum and OTE Segmentation in Sleep Endoscopy Images.

Accurate segmentation of upper airway structures such as the velum and OTE (oropharynx, tongue base, epiglottis) in drug-induced sleep endoscopy (DISE) images is crucial for predicting the degree and location of obstruction to determine treatment options for obstructive sleep apnea (OSA). This study...

Descripción completa

Detalles Bibliográficos
Publicado en:Journal of Imaging Informatics in Medicine Vol. 39; no. 4; pp. 3773 - 3784
Autores principales: Yeom, Jong Chan, Kim, Jin Youp, Kim, Young Jae, Kim, Kwang Gi, Rhee, Cha-Seo
Formato: pictorial research tables/charts Journal Article
Publicado: Springer Nature Aug2026
Acceso en línea:Ver este registro en EBSCOhost
fields @attributes:
  recordID: 1
pdfLink:
plink: https://search.ebscohost.com/login.aspx?direct=true&db=ccm&AN=196241822&site=ehost-live
header:
  @attributes:
    shortDbName: ccm
    uiTerm: 196241822
    longDbName: CINAHL Complete
    uiTag: AN
  controlInfo:
    bkinfo:
    dissinfo:
    jinfo:
      jid:
        29482925
        NR3A
      jtl: Journal of Imaging Informatics in Medicine
      issn: 29482925
      maglogo: N
    pubinfo:
      dt: Aug2026
      vid: 39
      iid: 4
      pid: 237
      pub: Springer Nature
      place: New York, New York
    artinfo:
      ui:
        196241822
        189902488
        196241822
        196241822
        10.1007/s10278-025-01756-4
        196241822
      ppf: 3773
      ppct: 11
      formats:
      tig:
        atl: Comparative Performance Evaluation of Federated and Centralized Learning for Velum and OTE Segmentation in Sleep Endoscopy Images.
      aug:
        au:
          Yeom, Jong Chan
          Kim, Jin Youp
          Kim, Young Jae
          Kim, Kwang Gi
          Rhee, Cha-Seo
        affil: https://ror.org/005nteb15 Department of Bio-health Medical Engineering, Gachon University, Gil Medical Center, Incheon, Republic of Korea
      sug:
        subj:
          Federated Learning
          Endoscopy
          Image Processing, Computer Assisted
          Sleep Apnea, Obstructive Diagnosis
          Tongue Radiography
          Sensitivity and Specificity
          Human
          Comparative Studies
          Videorecording
          Funding Source
          Epiglottis Radiography
          Oropharynx Radiography
          Machine Learning
          Deep Learning
          Reproducibility of Results
          Radiographic Image Interpretation, Computer-Assisted
          Convolutional Neural Networks
          Predictive Validity
          Academic Medical Centers
          South Korea
          Oropharynx Anatomy and Histology
          Confidence Intervals
          Artifacts
          Airway Obstruction Radiography
          Sedation
          Retrospective Design
          Record Review
          T-Tests
          Descriptive Statistics
          Multicenter Studies
      ab: Accurate segmentation of upper airway structures such as the velum and OTE (oropharynx, tongue base, epiglottis) in drug-induced sleep endoscopy (DISE) images is crucial for predicting the degree and location of obstruction to determine treatment options for obstructive sleep apnea (OSA). This study systematically compares centralized learning (CL) and federated learning (FL) approaches for the semantic segmentation of these regions using multi-institutional DISE video data. A convolutional neural network (CNN)-based segmentation model was trained and evaluated for both learning paradigms. The results consistently showed that the CL approach achieved statistically significantly higher segmentation performance across all metrics—precision, recall, accuracy, and Dice similarity coefficient (DSC)—for both the velum and OTE regions compared with FL. For the velum region, CL achieved a DSC of 85.91 ± 1.01%, compared with FL's 81.78 ± 0.58%. Similarly, for the OTE region, CL achieved an average DSC of 87.04 ± 0.41%, whereas FL achieved 85.20 ± 0.25%. Further analysis revealed that while both models struggled with ambiguous boundaries and anatomical variability—particularly for the tongue base—the epiglottis and oropharynx lateral wall were segmented with higher accuracy. These findings underscore the need for advanced techniques in FL, such as improved optimization algorithms and methods to address data heterogeneity, to narrow the performance gap with CL. This study provides foundational insights for developing more robust and clinically applicable deep learning models for upper airway analysis, emphasizing the importance of future research into advanced FL strategies and real-world validation.
      pubtype: Academic Journal
      doctype:
        pictorial
        research
        tables/charts
        Journal Article
      ougenre: Article
    language: English
    refInfo:
    holdings:
      @attributes:
        islocal: N