The objective of this study is the development of an automatic segmentation framework for measuring volume changes in the white matter tissue from premature infant MRI data. The early stage of the brain development presents several major computational challenges such as structure and shape variations between patients. Furthermore, a high water content is present in the brain tissue, that leads to inconsistencies and overlapping intensity values across different brain structures. Another problem lies in low-frequency multiplicative intensity variations, which arises from an inhomogeneous magnetic field during the MRI acquisition. Finally, the segmentation is influenced by the partial volume effects which describe voxels that are generated by more than one tissue type. To overcome these challenges, this study is divided into three parts with the intention to locally segment the white matter tissue without the guidance of an atlas. Firstly, a novel brain extraction method is proposed with the aim to remove all non-brain tissue. The data quality can be improved by noise reduction using an anisotropic diffusion filter and intensity variations adjustments throughout the volume. In order to minimise the influence of missing contours and overlapping intensity values between brain and nonbrain tissue, a brain mask is created and applied during the extraction of the brain tissue. Secondly, the low-frequency intensity inhomogeneities are addressed by calculating the bias field which can be separated and corrected using low pass filtering. Finally, the segmentation process is performed by combining probabilistic clustering with classification algorithms. In order to achieve the final segmentation, the algorithm starts with a pre-segmentation procedure which was applied to reduce the intensity inhomogeneities within the white matter tissue. The key element in the segmentation process is the classification of diffused and missing contours as well as the partial volume voxels by performing a voxel reclassification scheme. The white matter segmentation framework was tested using the Dice Similarity Metric, and the numerical evaluation demonstrated precise segmentation results.