Automated Classification and Size Estimation of Fetal Ventriculomegaly from MRI Images: A Comparative Study of Deep Learning Segmentation Approaches

Abstract

Fetal ventriculomegaly is one of the major risks in prenatal diagnosis, which is an enlargement of the ventricles of the developing fetus's brain. Timely prediction of these brain disorders helps patients and healthcare providers make informed decisions and treatment plans. This paper investigates the classification of ventriculomegaly from MRI images using a deep learning approach. Segmentation of ventricles is performed using DeeplabV3+ and U-Net architectures and the accuracy, dice coefficient, and loss of these two approaches are compared. Additionally, a novel approach for size estimation was developed, utilizing contour detection to delineate ventricle boundaries and compute their areas by enclosing them within rectangles. A threshold-based classification method was then applied to differentiate between abnormal ventriculomegaly and normal cases. This research contributes to the assessment and comparison of the segmentation algorithms for aiding in the identification and classification of ventriculomegaly cases, offering potential implications for clinical diagnosis and treatment planning.

Ähnliche Arbeiten