The Comparison Analysis of the Cervical Features Between Second-and Third-Trimester Pregnancy in Ultrasound Images Using eXplainable AI

Yeong-Eun Jeon, Ga-Hyun Son, Ho-Jung Kim, Jae-Jun Lee, Dong-Ok Won

Abstract: The fetus is maintained in the uterus by the cervix throughout pregnancy. In the last period of pregnancy, the cervix softens, shortens, and expands in preparation for labor. When most clinicians check for changes in the cervix, like softness, they use palpation. However, since this palpation causes discomfort to the mother, we want to compare the status of the area around the cervical canal according to the pregnancy period through ultrasound images only. Therefore, we trained the deep learning network model and obtained high performance for the second and third-trimester classifications. Further, we used explainable Artificial Intelligence (XAI) techniques (i.e., Grad-CAM/Grad-CAM++, Score-CAM, Eigen-CAM) in order to find which areas were important features during the deep learning network training. As a result, in the third-trimester period images, it was seen that the fetal head was a major feature, however, it was found that the cervix and cervical border were also affected without a fetal head. Also, it was determined that the classification in the second-trimester images was based on the potential region toward the lower uterus from the internal os. By analyzing the deep learning network result using the XAI approaches, this might be used as a new feature to describe the cervical change.