An integrated fractional stockwell transform with atrous convolutions aided vision transformer based capsule network for fetal ECG arrhythmia detection.

One of the most difficult but important steps in assessing the fetus's health is the diagnosis of fetal cardiac abnormalities using fetal electrocardiograms (FECG). In order to provide accurate information regarding the fetus's condition, FECG monitoring is required. Severe fetal arrhythmia can cause heart failure or even death. This paper presents a Non-Causal Adaptive Filter that extracts the FECG through multiple error estimation. The maternal channel ECG in the chest will be used as the reference input, and the abdominal ECG will be used as the primary input for this filter. The clean FECG signals will be transformed into time-frequency (T-F) images using a fractional Stockwell transform after the FECG signals have been extracted. Using the Stockwell and fractional Fourier transforms, it can simultaneously display the time and fractional-frequency data in the time-fractional-frequency plane. The ability to detect fetal ECG arrhythmias with a clear physical interpretation is more significant. The resulting images are fed into the Atrous Convolutions aided Vision Transformer based Capsule Network (AConvVTCapNet) model, which detects fetal ECG arrhythmias. In this instance, Atrous convolutions efficiently compute dense feature maps, allowing the network to have wider receptive fields. The proposed model's parameters are adjusted using a new Opposition based Fire Hawk Optimization (OFHO) technique, which is carried out by the capsule network during the classification process. The proposed method obtained 98.23 % accuracy and 98.25 % specificity in the fetal ECG arrhythmia detection process.