MCMxxxVI (1936): A New Versatile Automatic Technique for Brain Lesion Segmentation and Volume Analysis

Maria Valdes Hernandez¹, Mark E. Bastin², Joanna M. Wardlaw¹

We describe an alternative approach to automatically segment brain tissues, haemosiderin deposits (HDs) including brain microbleeds (BMBs) and white matter lesions (WMLs). Two types of structural images are registered, modulated in frequency to the red/green space and fused. Minimum variance quantization quantifies and segments tissues and lesion volumes. The intra-observer reliability and intra-class correlation coefficient for WMLs segmentation was 0.99, with a similarity index of 0.9. HDs and BMBs are all detected along with high-iron-content areas like small vessels. This approach is fast, accurate, observer independent and increases the reliability and repeatability of WML, BMBs and HDs volume measurements.