Prognosis of Anomaly Detection in Tiny Life using AI based Fetal Segmentation Approach

Medical imaging for prenatal health monitoring has advanced significantly in recent years, especially with deep learning approaches. These developments make it possible to automate the analysis and segmentation of fetal features using MRI and ultrasound (US) scans. These advancements have the potential to have a significant influence on prenatal diagnosis by facilitating early detection of anomalies and thorough evaluation of fetal health. Convolutional neural network SegNet is used for segmentation tasks. This network architecture follows an encoder-decoder design in which the encoder extracts features and the decoder reconstructs segmented images. Methods such as threshold, region-based, edge-based, semantic, instance, and k-means clustering are applied for segmentation. The primary model employed is U-Net, known for its effectiveness in semantic segmentation tasks. The foetus is segmented by its structure and the foetus’s head circumference (HC) can be detected by this segmentation process by using U-Net and by other segmentation methods like threshold, region-based, edge-based, semantic, instance, and k-means clustering the foetus structure can be detected. Using U-net the dice score is obtained as 96.51 for training images with a 648.65 loss and 95.29 for validation images with a 1260.94 validation loss. The paper provides a systematic technique for automating the segmentation of foetus features in medical images. Using deep learning techniques, notably the U-Net architecture, encouraging results have been made in accurately highlighting foetus areas in ultrasound images. The models have been increased and segmentation accuracy has increased through the application of several segmentation techniques and performance evaluation standards.