Automatic Identification of Early Miscarriage Based on Multiple Features Extracted From Ultrasound Images

Ultrasound is one of the most widely used multipurpose imaging modalities that is ideal for monitoring and diagnosing early pregnancy events. The first sign and measurable element of an early pregnancy is the Gestational Sac (GS). Currently, the size of GS is manually measured from an ultrasound image of the GS. This paper argues that the Mean Sac Diameter (MSD) derived from the manual measurements results in interand intra-observer variations, which may lead to difficulties in diagnosis. The paper proposes a fully automated diagnosis solution to accurately identify miscarriage cases in the first trimester of pregnancy based on currently used MSD as well as alternative geometric features extracted from the image. Our experimental results show that the perimeter and volume of the GS are effective features where the perimeter can outperform the MSD. Furthermore, our study shows that the identification accuracy of early miscarriage can be further improved by combining the perimeter, volume and MSD of the GS.