Automatic Accurate Infant Cerebellar Tissue Segmentation with Densely Connected Convolutional Network.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• The cerebellum is vital for motor control and cognitive functions.
• Infant brain development studies require accurate cerebellar tissue segmentation.
• Challenges include poor tissue contrast, complex folding, and partial volume effects in infant brains.

Purpose of the Study:

• To develop an automated and accurate method for infant cerebellum segmentation.
• To address the limitations of existing methods in segmenting infant cerebellar tissues.

Main Methods:

• Proposed a densely connected convolutional network (CNN) for robust feature learning.
• Developed a novel deep neural network architecture ensuring maximum information flow between layers.
• Utilized contextual features from all preceding layers to guide segmentation.

Main Results:

• The proposed method demonstrated superior performance compared to state-of-the-art techniques.
• Achieved accurate segmentation on infant brain images from the Baby Connectome Project (BCP).
• Successfully segmented brains of 6- and 12-month-old infants.

Conclusions:

• The novel densely connected CNN offers an effective solution for challenging infant cerebellum segmentation.
• This advancement facilitates quantitative analysis of early brain development.
• The method shows promise for improving neuroimaging research in infants.