Detection and Classification of Pregnancy StateUsing Deep Learning Technique

Abstract

This work aims to design and develop a model that detects and classifies pregnancy health status. Ultrasound is one of the most prevalent developments in clinical imaging, as it enables a doctor to evaluate, analyze and treat diseases. Most complications from pregnancy lead to serious problems that restrict healthy growth, causing weakness or death. In this work, an image processing system was developed to recognize the health during pregnancy and classify it for all stages of its development. The technique in deep learning has been implemented, as CNN (Resnet50) image recognition model was applied to detect and classify fetal health status from ultrasound images. The proposed model contributed to providing an integrated solution for each pregnancy period that works to identify all stages of fetal development, starting from the pre-pregnancy stage (here it is known about the suitability of the uterus for pregnancy, the size of the ovum, and its ability to form the fetus) and up to the stage of birth, through training, verification and testing using the cross-verification technique that five folds of the diagnostic rudder were used under the patterns that distinguish each stage from the other and to verify that it is sound or unsound in the concerning stage. This study enhanced diagnostic accuracy by using transfer learning and novel accessory images that were not trained as feedback. The model achieved an accuracy of 96.5% in detecting the fetus and classifying it into any of the stages that were divided according to the features that appear from one stage to the next to eleven categories.