Effective Feature Technique for Heart Disease [FETUS] Prediction in Machine Learning

Perinatal mortality is a big issue in countries like India due to infections, low birth weight, malnutrition, fetal malformations and most importantly fetal heart defects (FHD). It is very important to detect FHD at an early stage to reduce mortality and do interventions on time. This work uses fetal ultrasound images from clinical settings using a robust preprocessing pipeline with a median filter having tophat model to improve image quality and contrast. Segmentation is done using Multi-Scale Region of Interest (MSROI) to segment the structures in the fetal heart. Convolutional Neural Networks (CNNs) are used for fetal heart abnormality detection and classification as they can learn complex patterns from raw data. Although Support Vector Machines (SVMs) are good with small datasets, CNNs are good with high dimensional, heterogeneous data. The approach is to detect early diagnosis of FHD especially in resource constrained healthcare settings and eventually perinatal improvement.