Scholarlink

<p>Brain tumor segmentation is a crucial aspect of medical image analysis that distinguishes brain tumors from adjacent normal tissues in magnetic resonance imaging (MRI) scans. The objective of this task is to generate a binary or multi-class segmentation mask, meticulously delineating the precise location and extent of the identified tumor within the imaging data. Current paradigms for medical image segmentation, exemplified by variations of encoder–decoder architectures such as fully convolutional networks (FCNs) and a spectrum of U-shaped networks (e.g., U-Net and UNet++), have achieved notable success. However, they suffer from three persistent challenges: (1) the confinement of multi-scale feature extraction, (2) the encounter with feature map sparsity, and (3) the neglect of long-range dependencies. To address these limitations, we propose DensePyConvNet, a densely connected pyramid convolutional network tailored for accurate and efficient brain tumor segmentation in MRI. The model integrates multi-scale feature extraction, dense skip connection, and strip pooling strategy, achieving state-of-the-art performance on both Kaggle_3m and BraTS2019 datasets. Given the high-resolution, multi-modal nature of medical imaging data and the clinical imperative for real-time diagnosis, DensePyConvNet is designed with high-performance computing (HPC) deployment in mind. The proposed architecture exhibits strong parallelism and low-latency inference, making it well-suited for GPU acceleration, large-scale processing, and integration into time-critical diagnostic pipelines.</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 dense pyramid convolutional neural network for MRI brain tumor segmentation

  • 期刊论文

  • 发表日期：

    2025年9月5日

查看全文
  • Wei Xiong,
  • Haina Song,
  • Honggang Xie,
  • Yan Li

摘要

Brain tumor segmentation is a crucial aspect of medical image analysis that distinguishes brain tumors from adjacent normal tissues in magnetic resonance imaging (MRI) scans. The objective of this task is to generate a binary or multi-class segmentation mask, meticulously delineating the precise location and extent of the identified tumor within the imaging data. Current paradigms for medical image segmentation, exemplified by variations of encoder–decoder architectures such as fully convolutional networks (FCNs) and a spectrum of U-shaped networks (e.g., U-Net and UNet++), have achieved notable success. However, they suffer from three persistent challenges: (1) the confinement of multi-scale feature extraction, (2) the encounter with feature map sparsity, and (3) the neglect of long-range dependencies. To address these limitations, we propose DensePyConvNet, a densely connected pyramid convolutional network tailored for accurate and efficient brain tumor segmentation in MRI. The model integrates multi-scale feature extraction, dense skip connection, and strip pooling strategy, achieving state-of-the-art performance on both Kaggle_3m and BraTS2019 datasets. Given the high-resolution, multi-modal nature of medical imaging data and the clinical imperative for real-time diagnosis, DensePyConvNet is designed with high-performance computing (HPC) deployment in mind. The proposed architecture exhibits strong parallelism and low-latency inference, making it well-suited for GPU acceleration, large-scale processing, and integration into time-critical diagnostic pipelines.