Scholarlink

<p>Segmentation of lesions in medical resonance imaging (MRI) is the most helpful method for diagnosing multiple sclerosis (MS) patients. Despite existing challenges, this study aims to enhance MS lesion segmentation using an innovative deep-learning approach. We propose a convolutional neural network (CNN) based architecture leveraging an auto-encoder with data augmentation and a high-resolution integration block (HRIB). This architecture enables end-to-end segmentation, enhancing accuracy by incorporating information from up-sampling. Moreover, the proposed architecture is evaluated using the medical image computing and computer-assisted intervention (MICCAI) dataset at the University Medical Center Ljubljana (UMCL) of MS patients. For the MICCAI-2008 dataset, our model achieved an 88.9% dice similarity coefficient (DSC), 98.2% precision, and 84.5% accuracy across T1, T2, and FLAIR modalities. Similarly, for MICCAI-2016 and White Matter MS (WMMS) datasets, our model attained DSC scores of 62.7% and 78.3%, respectively, with high precision and accuracy. Also, comparative analysis has been done for the proposed model for both datasets of MICCAI. This approach leads to better segmentation of MS lesions.</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

Medical Lesion Segmentation Using Deep Learning Technique for Multiple Sclerosis Disease

  • 期刊论文

  • 发表日期：

    2025年5月25日

查看全文
  • Abhilasha Joshi,
  • K. K. Sharma

摘要

Segmentation of lesions in medical resonance imaging (MRI) is the most helpful method for diagnosing multiple sclerosis (MS) patients. Despite existing challenges, this study aims to enhance MS lesion segmentation using an innovative deep-learning approach. We propose a convolutional neural network (CNN) based architecture leveraging an auto-encoder with data augmentation and a high-resolution integration block (HRIB). This architecture enables end-to-end segmentation, enhancing accuracy by incorporating information from up-sampling. Moreover, the proposed architecture is evaluated using the medical image computing and computer-assisted intervention (MICCAI) dataset at the University Medical Center Ljubljana (UMCL) of MS patients. For the MICCAI-2008 dataset, our model achieved an 88.9% dice similarity coefficient (DSC), 98.2% precision, and 84.5% accuracy across T1, T2, and FLAIR modalities. Similarly, for MICCAI-2016 and White Matter MS (WMMS) datasets, our model attained DSC scores of 62.7% and 78.3%, respectively, with high precision and accuracy. Also, comparative analysis has been done for the proposed model for both datasets of MICCAI. This approach leads to better segmentation of MS lesions.