Improved Non-Invasive Detection of Congenital Heart Disease with Sparse Domain Kalman Filtering for Fetal ElectrocarDiogram Denoising

Electrocardiography has potential for the diagnosis of congenital heart disease in fetuses, but this potential is still challenged by large amounts of noise present on non-invasive measurements. In recent work, Kalman-LISTA was introduced as a tool for sparse domain Kalman filtering, which was shown to be an adequate for suppressing this noise. In this work, Kalman-LISTA is compared to per-sample median filtering, and both denoising techniques are used as an input to a classifier network for the detection of congenital heart disease. Kalman-LISTA was shown to raise the method's sensitivity by 10% while hardly affecting specificity. More research may further improve denoising and classification performance, and the potential benefits to clinical practice should be evaluated.