Amniotic Fluid and Placental Segmentation, and Preterm Prognostication AI Algorithms

This thesis comprises three projects utilizing AI techniques to advance prenatal care via segmentation and prognostication. Two initial projects focus on segmenting the Amniotic Fluid (AF) and placenta, critical for fetal growth, and detecting pathologies. The novel AFNet architecture, employing Convolutional Neural Networks, achieves a 93.38% mean IOU similarity score for AF segmentation and 88.39% 2D Dice, and 81.42% 3D Dice scores for placental segmentation. The final project prognosticates preterm birth in twin pregnancies using a Random Forest Classifier on binned gestational age data, achieving 89.4% recall and 74.1% F1 score for birth prediction under 37 weeks. These projects showcase AI's potential and limitations in advancing medical segmentation and classification.<p></p>