AUTOMATIC LOCALIZATION OF FETAL HEAD CIRCUMFERENCE ON TWO-DIMENSIONAL ULTRASONOGRAPHY IMAGES USING YOLOV8

• Main Author: Aisyah, Siti
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/86057
• Institution: Institut Teknologi Bandung

Fetal head circumference (HC) is one of thr biometric parameters that is often used to determining gestational ages and to assess fetal size in the womb. In practice, until now, fetal head circumference localization is still done manually by doctors or sonographers by drawing an elliptical line that encloses the fetal head circumference. This can cause errors and observer variations. Along with the development of science and technology, localization can be done automatically. Currently, deep learning has revolutionized rapidly in various fields including the medical field. Therefore, in this final project, an automatic localization method based on deep learning will be applied. The deep learning chosen is YOLOv8 (You Only Look Once version 8). Then, the performance of YOLOv8 will be evaluated using the average value of the cross-validation results with evaluation metrics of precision, recall, mAP50, mAP50-95, and F1-score. On data covering all trimesters with a total of 999 2D USG images of the fetal head, the performance results were 0.9841 ± 0.0418; 0.9840 ± 0.0338; 0.9888 ± 0.0234; 0.9127 ± 0.0658; and 0.9839 ± 0.0365. Better localization results were obtained in the second and third trimesters. This is because the ultrasound images in the second and third trimesters have a fetal skull that is larger and denser than in the first trimester, so it looks clearer. With the results of localization performance in the second trimester sequentially were 0.9929 ± 0.0169; 0.9891 ± 0.0280; 0.9932 ± 0.0054; 0.9259 ± 0.0532; and 0.9909 ± 0.0206. And the results of localization performance in the third trimester sequentially were 0.9057 ± 0.2740; 0.9701 ± 0.1318; 0.9645 ± 0.1209; 0.8294 ± 0.1681; and 0.8910 ± 0.2840.