Scholarlink

<p>Accurate segmentation and identification of hippocampal subregions are essential for understanding spatial memory, neuronal plasticity, and disease-related alterations in brain architecture. While fluorescent immunohistochemistry (IHC) enables detailed visualization of subregion-specific molecular markers, automated segmentation and classification remain challenging due to staining variability, morphological complexity, and low signal-to-noise ratios. Moreover, the absence of benchmark datasets has hindered the development of computational approaches for the automatic segmentation and identification of hippocampal regions in histological images, which are crucial for streamlining downstream analyses. To address these limitations, we introduce a novel multiplexed murine hippocampal dataset containing images stained with cFos, NeuN, and either ΔFosB or GAD67, capturing neuronal activity, structural features, and plasticity-associated signals. In parallel, we propose a multitask UNet-based generative adversarial network (MT-UGAN) that simultaneously performs segmentation and classification of hippocampal subregions from murine IHC images. The model leverages a UNet-based GAN architecture with a shared encoder, allowing for feature reuse across both segmentation and classification tasks. Together, the dataset and MT-UGAN establish the first integrated system for automated hippocampal subregion analysis. Experimental results demonstrate that the proposed MT-UGAN significantly outperforms conventional single-task models in both region-wise segmentation and classification performance, achieving a Dice score of 0.82 and classification accuracy of 91.0%. By unifying subregion segmentation and identification, this framework provides a scalable and interpretable solution for automated hippocampal analysis, thereby facilitating downstream analyses in neuroimaging research. To democratize our funding and support further research, we made the curated hippocampal subregion datasets and the proposed MT-UGAN model as a standalone tool publicly available at: <a href="https://github.com/MLBC-lab/MT-UGAN">https://github.com/MLBC-lab/MT-UGAN</a>.</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

Segmentation and classification of hippocampal subregions using multi-task generative adversarial networks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月28日

查看全文
  • Sayed Mehedi Azim,
  • Renuka Kumar,
  • Brian Corbett,
  • Iman Dehzangi

摘要

Accurate segmentation and identification of hippocampal subregions are essential for understanding spatial memory, neuronal plasticity, and disease-related alterations in brain architecture. While fluorescent immunohistochemistry (IHC) enables detailed visualization of subregion-specific molecular markers, automated segmentation and classification remain challenging due to staining variability, morphological complexity, and low signal-to-noise ratios. Moreover, the absence of benchmark datasets has hindered the development of computational approaches for the automatic segmentation and identification of hippocampal regions in histological images, which are crucial for streamlining downstream analyses. To address these limitations, we introduce a novel multiplexed murine hippocampal dataset containing images stained with cFos, NeuN, and either ΔFosB or GAD67, capturing neuronal activity, structural features, and plasticity-associated signals. In parallel, we propose a multitask UNet-based generative adversarial network (MT-UGAN) that simultaneously performs segmentation and classification of hippocampal subregions from murine IHC images. The model leverages a UNet-based GAN architecture with a shared encoder, allowing for feature reuse across both segmentation and classification tasks. Together, the dataset and MT-UGAN establish the first integrated system for automated hippocampal subregion analysis. Experimental results demonstrate that the proposed MT-UGAN significantly outperforms conventional single-task models in both region-wise segmentation and classification performance, achieving a Dice score of 0.82 and classification accuracy of 91.0%. By unifying subregion segmentation and identification, this framework provides a scalable and interpretable solution for automated hippocampal analysis, thereby facilitating downstream analyses in neuroimaging research. To democratize our funding and support further research, we made the curated hippocampal subregion datasets and the proposed MT-UGAN model as a standalone tool publicly available at: https://github.com/MLBC-lab/MT-UGAN.