A Comparative Study of Different Segmentation and Regression Models for Fetal Head Circumference Measurement

Constant fetal health monitoring during pregnancy becomes crucial in assessing and diagnosing natural fetal growth. Out of all the methods, the ultrasound imaging-based approach for fetal aspect measurement is quite commonly used in modern days. An extensive comparative approach of various regression and segmentation models has been taken in this paper in the measurement of fetal head circumference. This research focuses on optimal model selection out of CNN regression models and segmentation models. Out-of-regression CNN models ResNet-50, InceptionV3, EfficientNet, and out-of-segmentation models U- net and Attention-Unet have been considered and implemented. Utilizing evaluation matrices such as MAE, RMSE, accuracy, and standard deviation for comparative analysis, the model designs were examined. This research work was carried out on the public dataset named HC-18 containing 805 multi-scaled ultrasound images. With room for improvement in this research work, it is highly evident that Inception V3 performed the best having an accuracy of about 87.04% whereas U-net performed the worst, having an accuracy of about 73.85%.