Automated MRI cerebellar size measurements using active appearance modeling

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• The cerebellum is crucial for diagnosing various neurological and developmental disorders.
• Manual segmentation of the cerebellum is costly and time-consuming.
• Existing automated tools for cerebellar segmentation lack reliability and availability.

Purpose of the Study:

• To introduce the Cerebellar Analysis Toolkit (CATK), an automated tool for reliable cerebellar segmentation.
• To improve the accuracy and efficiency of cerebellar measurements in neuroimaging studies.
• To facilitate the diagnosis of disorders associated with cerebellar abnormalities.

Main Methods:

• Developed CATK based on a Bayesian framework (FMRIB's FIRST) using Active Appearance Models (AAMs).
• Employed linear registration and Point Distribution Models (PDM) with stellar sampling.
• Utilized T1-weighted MRI data from 63 subjects for training and validation.

Main Results:

• CATK demonstrated high reliability with average test-retest Intraclass Correlation Coefficients (ICCs) of 0.96.
• Achieved excellent agreement with manual labels, showing an average validity ICC of 0.87.
• Outperformed the atlas-based SUIT approach in reliability and validity.

Conclusions:

• CATK provides a reliable and valid automated solution for cerebellar segmentation.
• The toolkit can significantly aid neuroimaging studies in diagnosing cerebellar-related disorders.
• Future extensions aim to include cerebellar hemisphere parcel segmentation.