FETAL SPINA BIFIDASEGMENTATION AND CLASSIFICATION USING DED WITH FM2DCN

Abstract

The gold standard for monitoring foetuses is generally acknowledged to be ultrasound imaging. In this study, two different deep learning algorithms were used to segment and classify ultrasound images of the foetal spine into normal and pathological categories. Between November 2015 and November 2020, 300 expectant women were enrolled, and at a hospital, they underwent three-dimensional ultrasound examinations. An adaptive bilateral filter (ABF) is used to remove any observable noise from the images. The dilated encoder-decoder (DED) method is used to segment images of foetal spina bifida, and a feature map-based differential convolutional network (FMDDCN) is used to classify the images. The suggested model was 96% accurate in terms of pixels, according to segmentation analysis, and 87% accurate in terms of mean intersection over union (MIoU). The suggested model's accuracy was 96.5 at the end of the classification evaluation, compared to 95.0 for the state-of-the-art methods.