Yolact-based Approach for Real-Time Fetal Heart Segmentation

The detection and identification of anomalies in fetal hearts require extensive expertise, clinical experience, and significant time. In this context, deep learning (DL) technology based on instance segmentation has become increasingly popular and effective in accelerating the detection of fetal cardiac anomalies. This research aims to develop a DL model based on instance segmentation using the YOLACT approach to detect fetal cardiac anomalies. Specifically, it focuses on anomalies in the septal wall of the fetal heart using ultrasound (USG) images. Fetal cardiac USG images are used as the dataset to train the DL model. The test results demonstrate that the proposed DL model based on the YOLACT approach is capable of detecting defects in the septal wall of the fetal heart with a mean average precision (mAP) of 93.44% and an intersection over union (IoU) of 82.02%. In this regard, the YOLACT technique enables the identification of abnormalities at the pixel level, providing detailed visualization of the affected areas. This research holds the potential to be a breakthrough in the field of fetal cardiac anomaly detection, as it can reduce dependence on experts and expedite the diagnostic process. By utilizing deep learning technology and the YOLACT approach, it is expected that there will be improvements in accuracy and efficiency in identifying fetal cardiac anomalies, thereby enabling early and better medical interventions for patients.