Scholarlink

<p>Brain tumor imposes a serious threat to patients’ lives because of its aggressive nature and variations in its shape, size and location. But if detected well on time, it can even be cured. Consequently, researchers have been constantly working to develop state-of-the-art techniques to identify the stage and form of tumor. Magnetic Resonance Imaging (MRI) is an effective way to study the anatomy of brain and has been considered duly to perform segmentation of brain tumors in the field of neuro-oncology, facilitating diagnosis, treatment planning, and longitudinal disease assessment. However, the precise segmentation of brain tumors is a challenging task due to tumor heterogeneity, indistinct and irregular boundaries, intensity inhomogeneity and the scarcity of well-annotated, high-quality datasets. Convolutional neural network (CNN)-based methods have shown excellent prospects in segmentation and classification of brain tumors over the past years significantly advancing the field of automated brain tumor segmentation. Using a Preferred Reporting Items for Systematic Reviews and Meta Analyses (PRISMA) inspired review methodology, the paper systematically reviews around 200 scientific papers selected from reputed databases and proposes an architecture-based nomenclature of CNN-based deep learning models focusing on pathway models, fully convolutional networks (FCNs), U-Net and its variants, hybrid models, federated learning, transformers and attention mechanisms across different segmentation tasks, including whole tumor, tumor core, and enhancing tumor regions. The survey also extensively covers the technical aspects of these methods highlighting their advantages and limitations. The study further put forwards the key challenges faced by current approaches like interpretability and model transparency, class imbalance problem, real-time deployment challenge, computational efficiency and scalability, data availability and annotation issue and suggest possible solutions. Finally, the survey discusses the various open research problems and specifies future research directions for readers to develop more robust segmentation techniques for brain tumor analysis.</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

A Systematic Review of Convolutional Neural Networks for Multimodal MRI Brain Tumor Segmentation and Classification

  • 期刊论文

  • 发表日期：

    2026年7月28日

查看全文
  • Pratibha Maurya

摘要

Brain tumor imposes a serious threat to patients’ lives because of its aggressive nature and variations in its shape, size and location. But if detected well on time, it can even be cured. Consequently, researchers have been constantly working to develop state-of-the-art techniques to identify the stage and form of tumor. Magnetic Resonance Imaging (MRI) is an effective way to study the anatomy of brain and has been considered duly to perform segmentation of brain tumors in the field of neuro-oncology, facilitating diagnosis, treatment planning, and longitudinal disease assessment. However, the precise segmentation of brain tumors is a challenging task due to tumor heterogeneity, indistinct and irregular boundaries, intensity inhomogeneity and the scarcity of well-annotated, high-quality datasets. Convolutional neural network (CNN)-based methods have shown excellent prospects in segmentation and classification of brain tumors over the past years significantly advancing the field of automated brain tumor segmentation. Using a Preferred Reporting Items for Systematic Reviews and Meta Analyses (PRISMA) inspired review methodology, the paper systematically reviews around 200 scientific papers selected from reputed databases and proposes an architecture-based nomenclature of CNN-based deep learning models focusing on pathway models, fully convolutional networks (FCNs), U-Net and its variants, hybrid models, federated learning, transformers and attention mechanisms across different segmentation tasks, including whole tumor, tumor core, and enhancing tumor regions. The survey also extensively covers the technical aspects of these methods highlighting their advantages and limitations. The study further put forwards the key challenges faced by current approaches like interpretability and model transparency, class imbalance problem, real-time deployment challenge, computational efficiency and scalability, data availability and annotation issue and suggest possible solutions. Finally, the survey discusses the various open research problems and specifies future research directions for readers to develop more robust segmentation techniques for brain tumor analysis.