Scholarlink

Neonatal Hypoxic-Ischemic Encephalopathy is a form of brain injury caused by insufficient oxygen supply in the blood, which can lead to severe consequences such as death, lifelong disabilities, and long-term neurodevelopmental complications. Commonly used modalities for assessing HIE in neonates include Magnetic Resonance Imaging, Computed Tomography, and Electroencephalography. Although MRI remains the gold standard for diagnosing this condition and other imaging modalities, it poses a significant challenge for healthcare professionals to confirm the diagnosis accurately through mere visual assessment. This study specifically employed transcranial ultrasound imaging to facilitate the identification of HIE in neonates. Here, we have used four DCNN architectures: VGG19, InceptionResNetV2, EfficientNetv2-S, and DenseNet121, as these models have performed well in previous state-of-the-art work related to medical image analysis. To ensure unbiased training, three modes were used: fine-tuning, transfer learning, and scratch. The models are trained on resized images (224 × 224) and evaluated with an 80:20 train-test ratio. The results showed that EfficientNetv2-S and DenseNet121 achieved a remarkable accuracy of 100%, followed by VGG19 and InceptionResNetV2 at 99.9% and 99.8%, respectively. The fine-tuning mode outperforms the other modes, demonstrating promising results for HIE detection. The proposed model can help healthcare professionals diagnose HIE so it can be treated in a timely manner.

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

Deep Learning Approaches for Hypoxic-Ischemic Encephalopathy Detection in Neonates Using Ultrasound Imaging

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • K. Sridhar Patnaik,
  • Itu Snigdh,
  • Rajeev Mishra,
  • Rajeev Kumar Ranjan,
  • M. Rajesh Kumar Rao,
  • Chitranjan Kumar Rai,
  • Saket Kumar Singh,
  • Manisha Oraon,
  • Riya Agarwal,
  • Harish Shivprasad Gupta,
  • Md. Shahrukh,
  • Prakhar Suresh Srivastava,
  • Ruchi Pandey

摘要

Neonatal Hypoxic-Ischemic Encephalopathy is a form of brain injury caused by insufficient oxygen supply in the blood, which can lead to severe consequences such as death, lifelong disabilities, and long-term neurodevelopmental complications. Commonly used modalities for assessing HIE in neonates include Magnetic Resonance Imaging, Computed Tomography, and Electroencephalography. Although MRI remains the gold standard for diagnosing this condition and other imaging modalities, it poses a significant challenge for healthcare professionals to confirm the diagnosis accurately through mere visual assessment. This study specifically employed transcranial ultrasound imaging to facilitate the identification of HIE in neonates. Here, we have used four DCNN architectures: VGG19, InceptionResNetV2, EfficientNetv2-S, and DenseNet121, as these models have performed well in previous state-of-the-art work related to medical image analysis. To ensure unbiased training, three modes were used: fine-tuning, transfer learning, and scratch. The models are trained on resized images (224 × 224) and evaluated with an 80:20 train-test ratio. The results showed that EfficientNetv2-S and DenseNet121 achieved a remarkable accuracy of 100%, followed by VGG19 and InceptionResNetV2 at 99.9% and 99.8%, respectively. The fine-tuning mode outperforms the other modes, demonstrating promising results for HIE detection. The proposed model can help healthcare professionals diagnose HIE so it can be treated in a timely manner.