A Comparative Study of Automated and Manual Segmentation Methods for the Neonatal Brain

Quantitative assessment of Magnetic Resonance images (MRI) is known to be a useful computational tool to aid in the medical diagnosis of brain diseases and more generally of diseases related to the central nervous system. The method relies on the identification of brain tissue, and segmentation of the main anatomical brain structures to extract their volumes. While there are numerous accurate automatic segmentation techniques for adult brains, their efficacy often diminishes when applied to infant brains. Therefore, the purpose of this research is to address this disparity by conducting a comprehensive comparison of three existing automatic segmentation software tools for infant brains, employing manual segmentation as the ground truth. In particular, the study focuses on the segmentation of four anatomical structures in the neonatal brain: the Hippocampus, Thalamus, Amygdala, and Cerebellum. The chosen automatic segmentation tools, namely infant FreeSurfer, MANTiS, and iBEAT V2.0, are used to delineate these structures. The results obtained are rigorously analyzed and compared, providing information on the strengths and limitations of each method. This investigation is expected to contribute valuable knowledge to the field of neonatal brain image analysis, helping to enhance the accuracy and reliability of specific segmentation techniques for infant brains.