TY - EJOU AU - Mori, Kentaro AU - Kitaya, Kotaro AU - Ishikawa, Tomomoto AU - Hata, Yutaka TI - A Pregnancy Prediction System based on Uterine Peristalsis from Ultrasonic Images T2 - Intelligent Automation \& Soft Computing PY - 2021 VL - 29 IS - 2 SN - 2326-005X AB - In infertility treatment, it is required to improve a success rate of the treatment. A purpose of this study is to develop a prediction system for pregnancy outcomes using ultrasonic images. In infertility treatment, it is typical to evaluate the endometrial shape by using ultrasonic images. The convolutional neural network (CNN) system developed in the current study predicted pregnancy outcome by velocity information. The velocity information has a movement feature of uterine. It is known that a uterine movement is deep related to infertility. Experiments compared the velocity-based and shape-based systems. The shape-based systems predict the optimal uterine features for pregnancy success based on endometrial shape by inputting original ultrasonic images to CNN model. The current findings revealed that the velocity-based system provided similar accuracy to the shape-based systems. However, the output of the velocity-based system, the area under curve (AUC) for the receiver operating characteristic (ROC) curve, provided a higher value than the shape-based systems. The AUC values of the shape-based and velocity-based systems were 0.65, and 0.72, respectively. These results showed that the analysis of the velocity of uterine movements was effective for pregnancy outcome prediction. Previous clinical evaluation did not target the uterine movement but only the endometrial shape. Therefore, this study has revealed a new treatment approach for infertility. KW - Convolutional neural network (CNN); Infertility; Pregnancy prediction; Ultrasonic image; Uterine peristalsis DO - 10.32604/iasc.2021.01010