Dataset: Impact of pulsed-wave-Doppler velocity-envelope extraction techniques on classification of complete fetal cardiac cycles
The pulsed-wave Doppler (PWD) signal is used as a reference for the mechanical activity of the heart and several validated parameters are used in clinical practice, most of which are evaluated by cardiologists based on visual inspection alone. Specific datasets have also been established of the simultaneous recordings of Doppler ultrasound and fetal electrocardiography. These all require the development of software tools that can automatically identify complete and meaningful fetal cardiac cycles to enable measurements. Several works in the literature have reported investigations of the extraction of the PWD velocity envelopes. The work "Impact of pulsed-wave-Doppler velocity-envelope tracing techniques on classification of complete fetal cardiac cycles", by the same authors, focus on the selection and comparison of different steps in the signal processing chain that lead to the PWD envelopes, which are used as inputs for the detection of complete heartbeats. Here, we presents the dataset that has been used for the validation of the methodology. The data were collected from the fetal echocardiographic examination of 25 low-risk pregnant volunteers at gestational weeks ranging from the 21st to the 27th, from which we obtained a total of 43 PWD traces. All images and data are anonymous. The PWD images were collected at the Division of Pediatric Cardiology, San Michele Hospital, Cagliari, Italy. The study was approved by the Independent Ethics Committee of the Cagliari University Hospital (AOU Cagliari) and performed following the principles outlined in the Helsinki Declaration of 1975, as revised in 2000. Each volunteer signed a form acknowledging their informed consent to the research protocol. Based on the best clinical practice, we chose the five-chamber apical window. Using the apical five-chamber view, the four cardiac chambers and the first part of the aorta (assumed as a fifth chamber) can be investigated, and the diastolic and systolic functions have distinguishable morphologies in the PWD spectrum. Overall, the EA-wave indicates the mitral inflow and the V-wave the aortic outflow. Here, the 43 images, named as imgXX.bmp. An expert pediatric cardiologist labeled all the complete and measurable fetal cardiac cycles using a custom MATLAB graphical interface, as described in "Automatic detection of complete and measurable cardiac cycles in antenatal pulsed-wave Doppler signals". The annotations are saved in the annotations.mat file, that that consists of 2 columns: the first lists the image indexes at which the cardiologist annotated the fetal cardiac cycle onset, the second specifies the image number at which the annotation is related to. A MATLAB function allows visualizing the cardiologiest's annotations on a selected PWD image or all the cardiologiest's annotations in every PWD images in the dataset.
Steps to reproduce
The PWD signals were recorded using a Philips iE33 ultrasound machine (Philips, The Netherlands). We set the sweep speed to 75 mm/s and did not change the machine settings (e.g., gain, axis scaling, baseline, etc.) during the acquisition period (ranging from 6.4 s to 119.8 s). All the frames were captured using the video frame grabber USB3HDCAP USB3.0 (by StarTech, Ontario, Canada) connected to the DVI output of the ultrasound device. To ensure that no frame was dropped, the frames were collected at a rate of 60 fps, which is higher than both the screen refresh rate and the DVI output (30 fps). Due to the variable position of the foetus within the maternal uterus, the mitral blood inflow can be moving either towards or away from the ultrasound transducer. Therefore, the EA-V pattern can have a positive balance (positive EA wave, negative V wave) or a negative balance (negative EA wave, positive V wave). Using a custom Matlab graphical interface, we converted the whole video into a single wide image. To work only with images exhibiting a positive balance, we flipped the negative-balanced images upside-down. In the dataset created for "Impact of pulsed-wave-Doppler velocity-envelope tracing techniques on classification of complete fetal cardiac cycles", for each manually labelled complete and meaningful fetal cardiac cycle, the software tool also labelled as readable other PWD windows up to 15 samples before and after the starting point of the annotated one. Overall, we obtained 87,736 windows representing meaningful cycles. The other 86,583 windows representing incomplete or malformed foetal cardiac cycles were also randomly taken, with at least 40-samples distance from each annotation. This approach led to a balanced dataset.