Scholarlink

<p>Artificial intelligence (AI) is transforming neuroradiological practice, yet multiple sclerosis (MS) diagnosis remains challenged by qualitative MRI assessment with significant inter-observer variability. While advanced AI-based quantitative methods show promise, translation into standardized clinical tools aligned with diagnostic criteria remains limited, creating gaps between computational capabilities and neuroradiological application. To develop and clinically validate an AI-driven decision support system for automated, multi-parametric quantitative MS lesion characterization that transforms qualitative neuroradiological assessments into standardized, reproducible metrics aligned with McDonald spatial dissemination 2017 criteria. We developed an AI-powered framework analyzing T2-FLAIR data from 300 MS patients using conditional generative adversarial networks. The system implements context-aware lesion refinement and neuroanatomically informed classification to distinguish periventricular, juxtacortical, and paraventricular hyperintensities. Multi-parametric AI analysis quantifies lesion characteristics including count, area, and spatial position. Three board-certified neuroradiologists validated fifteen stratified cases across different scanners, imaging protocols, and lesion burden categories. The AI system achieved excellent correlation with expert neuroradiological assessments: periventricular lesions (<i>r</i> = 0.929, 95% CI [0.796, 0.977], MAE = 3.3), paraventricular lesions (<i>r</i> = 0.932, 95% CI [0.804, 0.978], MAE = 3.9), and juxtacortical lesions (<i>r</i> = 0.927, 95% CI [0.790, 0.976], MAE = 1.7). Clinical workflow efficiency improved with 74.2% reduction in analysis time (11.3 ± 2.7 to 3.0 ± 1.1&#xa0;min per case, <i>p</i> = 0.017) while maintaining high diagnostic accuracy. This AI-driven clinical decision support system successfully bridges advanced computational analysis and neuroradiological practice, providing standardized metrics aligned with McDonald 2017 criteria while improving workflow efficiency. The framework demonstrates potential for clinical implementation in neuroradiology departments.</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

AI-Driven Multi-parametric MS Lesion Analysis from T2-FLAIR Imaging: a Clinical Decision Support Framework for Neuroradiology

  • 期刊论文

  • 发表日期：

    2026年4月20日

查看全文
  • Mahdi Bashiri Bawil,
  • Mousa Shamsi,
  • Abolhassan Shakeri Bavil

摘要

Artificial intelligence (AI) is transforming neuroradiological practice, yet multiple sclerosis (MS) diagnosis remains challenged by qualitative MRI assessment with significant inter-observer variability. While advanced AI-based quantitative methods show promise, translation into standardized clinical tools aligned with diagnostic criteria remains limited, creating gaps between computational capabilities and neuroradiological application. To develop and clinically validate an AI-driven decision support system for automated, multi-parametric quantitative MS lesion characterization that transforms qualitative neuroradiological assessments into standardized, reproducible metrics aligned with McDonald spatial dissemination 2017 criteria. We developed an AI-powered framework analyzing T2-FLAIR data from 300 MS patients using conditional generative adversarial networks. The system implements context-aware lesion refinement and neuroanatomically informed classification to distinguish periventricular, juxtacortical, and paraventricular hyperintensities. Multi-parametric AI analysis quantifies lesion characteristics including count, area, and spatial position. Three board-certified neuroradiologists validated fifteen stratified cases across different scanners, imaging protocols, and lesion burden categories. The AI system achieved excellent correlation with expert neuroradiological assessments: periventricular lesions (r = 0.929, 95% CI [0.796, 0.977], MAE = 3.3), paraventricular lesions (r = 0.932, 95% CI [0.804, 0.978], MAE = 3.9), and juxtacortical lesions (r = 0.927, 95% CI [0.790, 0.976], MAE = 1.7). Clinical workflow efficiency improved with 74.2% reduction in analysis time (11.3 ± 2.7 to 3.0 ± 1.1 min per case, p = 0.017) while maintaining high diagnostic accuracy. This AI-driven clinical decision support system successfully bridges advanced computational analysis and neuroradiological practice, providing standardized metrics aligned with McDonald 2017 criteria while improving workflow efficiency. The framework demonstrates potential for clinical implementation in neuroradiology departments.