Scholarlink

Deep learning techniques such as Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) have demonstrated strong capabilities in region-of-interest (ROI) detection and disease severity scoring, aiming to reduce the manual workload for clinicians. However, studies encompassing an end-to-end method for ROI detection, their subsequent scoring, and aggregation to a patient level score in psoriatic arthritis (PsA) have been limited. To address this gap, we have developed an end-to-end multistep approach for automated radiographic assessment in PsA that detects joints-of-interest from X-rays, assigns structural damage scores to the identified joints, and aggregates them to an extremity level, and subsequently to a patient level score. Furthermore, we compare our approach with state-of-the-art methods and human experts. Our results demonstrate the strong performance of our joint detection models, achieving an average intersection over union (IoU) value of 0.88 for foot joints and 0.76 for hand joints. The subsequent scoring models for structural damage show excellent (intra-class correlation coefficient (ICC) > 0.90) agreement with an ICC of 0.94, 0.98, and 0.97 for joint level erosion, JSN, and patient level scores, respectively, when compared to human readers.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Vision Transformer Model for Automated End-to-End Radiographic Assessment of Joint Damage in Psoriatic Arthritis

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Darshana Govind,
  • Zijun Gao,
  • Chaitanya Parmar,
  • Kenneth Broos,
  • Nicholas Fountoulakis,
  • Lenore Noonan,
  • Shinobu Yamamoto,
  • Natalia Zemlianskaia,
  • Craig S. Meyer,
  • Emily Scherer,
  • Michael Deman,
  • Pablo Damasceno,
  • Philip S. Murphy,
  • Terence Rooney,
  • Elizabeth Hsia,
  • Anna Beutler,
  • Robert Janiczek,
  • Stephen S. F. Yip,
  • Kristopher Standish

摘要

Deep learning techniques such as Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) have demonstrated strong capabilities in region-of-interest (ROI) detection and disease severity scoring, aiming to reduce the manual workload for clinicians. However, studies encompassing an end-to-end method for ROI detection, their subsequent scoring, and aggregation to a patient level score in psoriatic arthritis (PsA) have been limited. To address this gap, we have developed an end-to-end multistep approach for automated radiographic assessment in PsA that detects joints-of-interest from X-rays, assigns structural damage scores to the identified joints, and aggregates them to an extremity level, and subsequently to a patient level score. Furthermore, we compare our approach with state-of-the-art methods and human experts. Our results demonstrate the strong performance of our joint detection models, achieving an average intersection over union (IoU) value of 0.88 for foot joints and 0.76 for hand joints. The subsequent scoring models for structural damage show excellent (intra-class correlation coefficient (ICC) > 0.90) agreement with an ICC of 0.94, 0.98, and 0.97 for joint level erosion, JSN, and patient level scores, respectively, when compared to human readers.