An adaptive neuro fuzzy methodology for the diagnosis of prenatal hypoplastic left heart syndrome from ultrasound images

Abstract

Congenital heart defect (CHD) is one of the most serious congenital deformities in a fetus. About 31% to 55% of CHDs are the primary cause that leads to life-threatening problem among neonates, hence sonographers emphasize the importance of prenatal CHD screening. Among 18 types of CHDs, the asymmetric appearance of the heart seems to be a challenging part. Hypoplastic left heart syndrome (HLHS) is a critical and rare CHD, with an underdeveloped left heart chamber of the fetus. This prenatal CHD can be diagnosed between 17 to 21 weeks of gestation period. Though ultrasound provides a good diagnostic result, prenatal diagnosis is still a challenging area due to its speckle noise and irregular appearance of the heart chambers. In this context, the basic step is to appropriately select the pre-processing algorithm, one such algorithm is the Fuzzy based maximum likelihood estimation technique (FMLET). Right ventricle left ventricle ratio (RVLVR) and cardiac thoracic ratio (CTR) are the two important features required for manual diagnosis of the ultrasound images. Hence, morphological operations such as open, close, thinning and thickening helps to extract the diagnostically important features inherent in the images. Finally, the computer aided decision support (CADS) system is designed with pre-processing module, morphological module and adaptive neuro fuzzy (ANFC) classifier module. ANFC is investigated as the good classifiers to help the experts in terms of self-learning with higher diagnostic rate. The proposed CADS proven with 91% of diagnostic accuracy and the standardized area under the ROC curve obtained was 0.9137.