Year: 2023
Type: conference-abstract
Abstract: The manual segmentation of fetal brain tissues is a time-consuming task that requires high expertise and is prone to errors. As a result, an automatic segmentation approach is considered a crucial tool for diagnosing the developing brain. Hence, deep learning techniques are the most advanced segmentation methods for medical images. However, a considerable number of training images are needed to represent the quick development of the fetus' transient brain structures. To address this, we collected ultrasound videos from several hospitals, Annotation was accomplished by specialist doctors who manually segmented 22 fetal ultrasound videos, which provided us with 6,631 frames and their corresponding masks. Previous methods focused on segmenting the brain into the three main tissue classes: WM, cortical gray matter, and ventricles. However, characteristics of other tissue classes, such as Cerebellum, choroid plexus, Cisterna Magna, Hippocampal gyri box, left ventricle Segment, Right ventricle, Thalamus, Cavum septum pellucidum, are important to understand and predict healthy or aberrant brain development in preterm infants of similar gestational age as fetuses. The cerebellum is particularly of clinical interest. Segmentation of the fetal cerebellum is crucial to identify potential abnormalities and disorders early on and monitor the effects of maternal drug exposure or environmental factors on fetal brain development. Therefore, we propose to first automatically segment the cerebellum and the other brain tissues. We employed a dense yet fast learning architecture to shorten the computing time so our model can be applied in real-life cases. According to the final test result, the proposed approach has an IOU rate of 89%, and a DC of 90%, it can efficiently predict and segment the cerebellum, which establishes a crucial foundation for clinical studies on the automatic segmentation method, the segmentation of the other brain tissues is in progress.
DOI: 10.1002/uog.26695
Authors: A. Zanbouaa, E. Bouyakhf, J. Bassma, S. Slimani
Institution: Université Mohammed VI des Sciences et de la Santé
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Keywords: fetal brain segmentation, prenatal ultrasound, cerebellum segmentation, brain tissue classification, deep learning segmentation, fetal brain development, preterm infants, choroid plexus, cisterna magna, thalamus, cavum septum pellucidum, white matter, cortical gray matter, ventricles, training data annotation
Open Access status: bronze
OpenAlex ID: W4387260239
