Abstract:
|
Ongoing Machine Learning advancements in the field of Medical Imaging have promoted development of Computer-Aided based Systems assisting medical professionals to make better decisions about patient's health. In particular, Fetal Ultrasound screening plays an important role in antenatal care providing diagnosis information and optimizing best delivery outcomes for mother and fetus. The ultrasound standard plane acquisition is a pre-requisite to subsequent biometric measurements and diagnosis. In this regard, there is an emerging need to automatically identify Fetal Biometric Patterns for fetal measurement as a tool to help the efficiency of clinical experts.
This thesis aims to automatically classify Fetal Biometric Standard Planes performed in the First Trimester Ultrasound Examination Routine. The proposed framework relies on applying a supervised learning workflow with two different extracted features to build a Fetal Biometric Planes classification model. The first alternative relies on training a model based on extracted Bag-of-Words SURF descriptors. The second alternative is inspired by the famous Eigenfaces Face Recognition Algorithm that operated by extracting PCA features to build the classification model. Finally, a comparative evaluation of classification will determine which of the two approaches achieves best results to develop the Automatic Classification System of Fetal Biometric Planes. |