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

Area of Science:

• Neuroscience
• Computational Biology
• Medical Imaging

Background:

• Accurate segmentation of hippocampal subregions is crucial for studying spatial memory, neuroplasticity, and neurological diseases.
• Automated analysis of histological images is hindered by staining variability and lack of benchmark datasets.
• Current methods struggle with the complexity of identifying hippocampal subregions in immunohistochemistry (IHC) images.

Purpose of the Study:

• To create a novel multiplexed murine hippocampal dataset for IHC analysis.
• To develop an automated system for simultaneous segmentation and classification of hippocampal subregions.
• To establish a benchmark for computational approaches in hippocampal subregion analysis.

Main Methods:

• Generated a multiplexed murine hippocampal dataset with cFos, NeuN, and ΔFosB or GAD67 staining.
• Proposed a multitask UNet-based generative adversarial network (MT-UGAN) for joint segmentation and classification.
• Utilized a shared encoder in the UNet-GAN architecture for efficient feature reuse.

Main Results:

• The MT-UGAN achieved a Dice score of 0.82 for segmentation and 91.0% accuracy for classification.
• The proposed multitask model significantly outperformed single-task models in both segmentation and classification.
• The integrated dataset and MT-UGAN provide a scalable solution for automated hippocampal analysis.

Conclusions:

• The developed dataset and MT-UGAN represent the first integrated system for automated hippocampal subregion analysis.
• This framework facilitates downstream neuroimaging research by providing accurate and efficient analysis.
• The dataset and model are publicly available to support further research and development.