This study developed a local semi-supervised approach to classify brain tissue in infant MRI scans under 5 months of age. The method first delineates myelinated white matter globally using a difference image between PD-weighted and T1-weighted templates. It then extracts unmyelinated white matter and grey matter from a T2-weighted reference, initialized with partial volume estimation. The approach successfully segments four tissue classes and improves understanding of neurodevelopment from early infant brain MRI.