A Deep Learning-based Pipeline for Segmenting the Cerebral Cortex Laminar Structure in Histology Images

Area of Science:

• Neuroscience
• Computational Biology
• Medical Imaging

Background:

• Understanding brain structure, particularly the cerebral cortex's laminar organization, is crucial for deciphering neural processing and neurological disorders.
• Accurate identification of cortical layers in neuroimaging data is essential for studying brain connectivity and function.

Purpose of the Study:

• To present a novel computational framework for segmenting cerebral cortical layers using AI and deep learning.
• To evaluate the framework's performance against existing methods using marmoset brain slice images and the BigBrain dataset.

Main Methods:

• Utilized AI-based tools for cortical label acquisition followed by a deep learning model for precise cerebral cortical layer segmentation.
• Employed Nissl-stained and myelin-stained brain slice images from the common marmoset (Callithrix jacchus).
• Compared segmentation accuracy using metrics like Euclidean distance, 95th percentile Hausdorff distance (95HD), and Jaccard Index.

Main Results:

• Achieved an acceptable Euclidean distance ($$) for cortical label acquisition.
• Demonstrated superior performance with a mean 95HD of $$ compared to Wagstyl et al.'s $$.
• Obtained a higher Jaccard Index ($$) on the BigBrain dataset, indicating better segmentation quality than the comparative method ($$).

Conclusions:

• The novel computational framework offers improved accuracy for cerebral cortical layer segmentation.
• This advancement facilitates detailed analysis of brain anatomical structure and connectivity.
• The findings support the use of advanced computational tools in neuroscience research for understanding brain function and disease.