Scholarlink

<p>Systemic sclerosis (SSc) is a heterogeneous autoimmune disease characterized by fibrosis, vascular damage, and immune dysregulation. In this study, we evaluated the potential of Fourier-transform infrared (FTIR) spectroscopy of whole blood samples combined with multivariate and machine learning approaches to differentiate between disease subtypes and the presence of interstitial lung disease (ILD). Subtle but consistent spectral differences were observed in the amide I/II and lipid-associated regions (~ 1500–1700&#xa0;cm<sup>−1</sup> and ~ 2900&#xa0;cm<sup>−1</sup>). Principal Component Analysis (PCA) revealed clear clustering along the first principal component (PC1). Subsequently, we developed and evaluated several supervised machine learning models to classify the serum spectra according to SSc subtype . In the classification between diffuse and limited SSc, the Random Forest (RF) model achieved the optimal overall performance. Our results demonstrated the potential of FTIR spectroscopy, particularly when combined with machine learning, as a non-invasive tool for disease stratification and biomarker discovery in SSc. Further improvements in model optimization and spectral feature extraction are needed to enhance clinical applicability.</p>

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

Spectroscopic and machine learning approaches for clinical subtyping in systemic sclerosis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月2日

查看全文
  • Bartosz Miziołek,
  • Justyna Miszczyk,
  • Wiesław Paja,
  • Michał Kępski,
  • Monika Bultrowicz,
  • Beata Bergler-Czop,
  • Aleksandra Frątczak,
  • Joanna Depciuch

摘要

Systemic sclerosis (SSc) is a heterogeneous autoimmune disease characterized by fibrosis, vascular damage, and immune dysregulation. In this study, we evaluated the potential of Fourier-transform infrared (FTIR) spectroscopy of whole blood samples combined with multivariate and machine learning approaches to differentiate between disease subtypes and the presence of interstitial lung disease (ILD). Subtle but consistent spectral differences were observed in the amide I/II and lipid-associated regions (~ 1500–1700 cm−1 and ~ 2900 cm−1). Principal Component Analysis (PCA) revealed clear clustering along the first principal component (PC1). Subsequently, we developed and evaluated several supervised machine learning models to classify the serum spectra according to SSc subtype . In the classification between diffuse and limited SSc, the Random Forest (RF) model achieved the optimal overall performance. Our results demonstrated the potential of FTIR spectroscopy, particularly when combined with machine learning, as a non-invasive tool for disease stratification and biomarker discovery in SSc. Further improvements in model optimization and spectral feature extraction are needed to enhance clinical applicability.