Newborn Brain MR Image Segmentation Using Deformable Surface Model Based on Fuzzy Knowledge Models

Abstract

Hypoxic-ischemic encephalopathy is a typical cerebral disorder of newborn babies. For diagnosis of the neonatal cerebral disorders, the measurement of cerebral volume and surface area is effective for quantitatively evaluating the morphological change. The measurement needs brain segmentation process in magnetic resonance (MR) images. However, there are few studies for newborn brain segmentation. This study proposes an automated brain segmentation method for the newborn brain in MR images. The automated method can improve the diagnosis of the newborn cerebral disorders with efficiency and high accuracy. The proposed method first constructs fuzzy models using learning dataset. The fuzzy models express brain features by fuzzy membership functions. Next, the proposed method applies the deformable surface model based on the fuzzy models to subject's head MR images, and estimates the subject's brain region. To validate the proposed method, it has been applied to 10 newborn subjects (revised ages are − 1 month and 1 month), and compare the segmentation result with those of the conventional methods. Leave-one-out-cross validation (LOOCV) test was conducted. The mean accuracy was 93.6 ± 3.7%, the mean sensitivity was 98.9 ± 0.5, and the mean G-metrics was 4.4 ± 0.8.

Keywords::

• Component
• Newborn Brain Segmentation
• Magnetic Resonance Imagings
• Fuzzy Knowledge Model
• Deformable Surface Model