Scholarlink

<p>Despite continuously evolving medical advances, CVD risk in Rheumatoid Arthritis (RA) remains paradoxically high to date. Carotid intima-media thickness (cIMT) is a widely used surrogate marker for atherosclerosis. However, issues related to operator-dependent assessment, availability and cost of carotid ultrasound are barriers to its wide implementation as an aid to cardiovascular risk assessment in RA. We aimed to develop a computational artificial intelligence (AI) model for cIMT prediction in RA. The recently proposed DERGA algorithm (Data Ensemble Refinement Greedy Algorithm) was employed in a database of datasets from 101 patients with RA, utilizing information on a wide range of clinical and laboratory variables, classical cardiovascular risk factors, disease-related parameters, and vascular assessments obtained with nailfold videocapillaroscopy (NVC). A total of 13,917,800 models were designed and trained. Among the four evaluated regression metaheuristic algorithms, the best predictive performance was achieved by the DERGA–Extra Trees model. The optimal model utilized only 8 of the 52 available input variables, while maintaining excellent predictive accuracy. Eventually, the 8 most important parameters predicting cIMT, listed from the most influential to the least influential, were white blood count, age, high density lipoprotein cholesterol, capillary density, systolic blood pressure, microhemorrhages, inhibitors of the renin-angiotensin-aldosterone, and methotrexate. A very strong positive linear correlation was observed between predicted and actual (measured) cIMT values (<i>R</i> = 0.9843), supporting the high predictive capability of the proposed computational intelligence model. Pending external validation in larger cohorts, the findings of the present study should be considered preliminary. Nevertheless, they provide further evidencesupporting the potential utility of AI applications for the assessment of subclinical vascular involvement in RA. While the role of NVC as an indicator of cardiovascular health is beginning to unfold, these findings underscore its promise as an adjunctive modality to facilitate more effective CVD risk stratification in RA.</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

Computational intelligence using nailfold videocapillaroscopy for the prediction of carotid intima-media thickness in rheumatoid arthritis: a cohort-based study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月4日

查看全文
  • Danial J. Armaghani,
  • Elena Angeloudi,
  • Amir H. Gandomi,
  • Panagiota Anyfanti,
  • Eleni Gavriilaki,
  • Stergios Soulaidopoulos,
  • Eleni Pagkopoulou,
  • Michael Doumas,
  • George D. Kitas,
  • Ahmed G. Gad,
  • Sanjog Chhetri Sapkota,
  • Georgios A. Drosopoulos,
  • Konstantina V. Leontari,
  • Markos Z. Tsoukalas,
  • Leonidas Triantafyllidis ,
  • Ahmed Salih Mohammed,
  • Abidhan Bardhan,
  • Pijush Samui,
  • Gai-Ge Wang,
  • Panagiotis G. Asteris,
  • Theodoros Dimitroulas

摘要

Despite continuously evolving medical advances, CVD risk in Rheumatoid Arthritis (RA) remains paradoxically high to date. Carotid intima-media thickness (cIMT) is a widely used surrogate marker for atherosclerosis. However, issues related to operator-dependent assessment, availability and cost of carotid ultrasound are barriers to its wide implementation as an aid to cardiovascular risk assessment in RA. We aimed to develop a computational artificial intelligence (AI) model for cIMT prediction in RA. The recently proposed DERGA algorithm (Data Ensemble Refinement Greedy Algorithm) was employed in a database of datasets from 101 patients with RA, utilizing information on a wide range of clinical and laboratory variables, classical cardiovascular risk factors, disease-related parameters, and vascular assessments obtained with nailfold videocapillaroscopy (NVC). A total of 13,917,800 models were designed and trained. Among the four evaluated regression metaheuristic algorithms, the best predictive performance was achieved by the DERGA–Extra Trees model. The optimal model utilized only 8 of the 52 available input variables, while maintaining excellent predictive accuracy. Eventually, the 8 most important parameters predicting cIMT, listed from the most influential to the least influential, were white blood count, age, high density lipoprotein cholesterol, capillary density, systolic blood pressure, microhemorrhages, inhibitors of the renin-angiotensin-aldosterone, and methotrexate. A very strong positive linear correlation was observed between predicted and actual (measured) cIMT values (R = 0.9843), supporting the high predictive capability of the proposed computational intelligence model. Pending external validation in larger cohorts, the findings of the present study should be considered preliminary. Nevertheless, they provide further evidencesupporting the potential utility of AI applications for the assessment of subclinical vascular involvement in RA. While the role of NVC as an indicator of cardiovascular health is beginning to unfold, these findings underscore its promise as an adjunctive modality to facilitate more effective CVD risk stratification in RA.