Scholarlink

Purpose <p>Accurately detecting Acute bilirubin encephalopathy (ABE) from non-ABE neonates with hyperbilirubinemia (HB) condition remains a challenge in clinical practice. In this study, an automatic ABE diagnosing system based on multi-modal MRI images and non-image clinical metadata is proposed to address the issue.</p> Methods <p>A total of 75 ABE neonates and 75 non-ABE neonates with HB are included in the study. Each patient has 3 multi-modal magnetic resonance images and 8 non-image clinical features. To investigate the diagnosing model’s performance, 3 different feature sets, namely deep features from multi-modal MRI images, non-image clinical features, and fusion features, are extracted, respectively, and then further classified by a support vector machine (SVM), respectively.</p> Results <p>The results indicated the SVM classifier built on the fusion features achieved the best classification performance with an accuracy of 93.24 ± 2.35, specificity of 91.38 ± 4.45%, sensitivity of 95.11 ± 2.97%, precision of 91.87 ± 3.88%, area-under-the-curve (AUC) of 98.08 ± 1.16%, F1_score of 93.38 ± 2.23%. The performance of the SVM classifier built on the deep features was better than that built on the non-image clinical features.</p> Conclusion <p>Our study demonstrated that ABE diagnostic performance based on deep features from multi-modal MRI images could be significantly improved by incorporating clinical features. The proposed strategy may potentially be applicable to clinical practice to facilitate clinical management.</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

Detecting acute bilirubin encephalopathy in neonates based on multimodal MRI images and non-image clinical data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年5月26日

查看全文
  • Miao Wu,
  • Qian Liu,
  • Can Lai

摘要

Purpose

Accurately detecting Acute bilirubin encephalopathy (ABE) from non-ABE neonates with hyperbilirubinemia (HB) condition remains a challenge in clinical practice. In this study, an automatic ABE diagnosing system based on multi-modal MRI images and non-image clinical metadata is proposed to address the issue.

Methods

A total of 75 ABE neonates and 75 non-ABE neonates with HB are included in the study. Each patient has 3 multi-modal magnetic resonance images and 8 non-image clinical features. To investigate the diagnosing model’s performance, 3 different feature sets, namely deep features from multi-modal MRI images, non-image clinical features, and fusion features, are extracted, respectively, and then further classified by a support vector machine (SVM), respectively.

Results

The results indicated the SVM classifier built on the fusion features achieved the best classification performance with an accuracy of 93.24 ± 2.35, specificity of 91.38 ± 4.45%, sensitivity of 95.11 ± 2.97%, precision of 91.87 ± 3.88%, area-under-the-curve (AUC) of 98.08 ± 1.16%, F1_score of 93.38 ± 2.23%. The performance of the SVM classifier built on the deep features was better than that built on the non-image clinical features.

Conclusion

Our study demonstrated that ABE diagnostic performance based on deep features from multi-modal MRI images could be significantly improved by incorporating clinical features. The proposed strategy may potentially be applicable to clinical practice to facilitate clinical management.