To test our hypothesis, 7 female Landrace swine (Prairie Swine, Saskatoon, SK, CAN) were used in a prospective, repeated measures study. All animal procedures were conducted in accordance with the National Institutes of Health policy on the use of animals in research and approved by the Research Ethics Board Animal Use and Care Committee at the University of Saskatchewan (Saskatoon, SK, CAN; Animal Use Protocol #20200042). This study exploring rapid ultrasonographic identification of the heart was one of several concurrent investigation protocols performed in these swine with these 7 study subjects representing a convenience sample based on compatibility of protocols and resources. Following a minimum 7-day acclimatization, swine were anesthetized, intubated, mechanically ventilated, instrumented, heparinized, and placed supine in a V-shaped holder [25]. Continuous heart rate and cardiac activity were monitored by electrocardiogram (ECG) using a traditional 3-lead limb system (ADI, Colorado Springs, CO, USA).

Experimental sonographic localization of the heart was performed using a Verathon BladderScan Prime Plus bladder scanner (Verathon, WA, USA). With the operator standing on the animal’s left side, the transducer was placed perpendicular to the chest wall along the left sternal border. Pre-arrest scans were performed from the 4th to the 8th ICSs, with scanner-generated images recorded for each space (see Fig. 1). Echocardiography was then performed by author PO using a GE Vivid i (General Electric, NY, USA) to describe the underlying cardiac anatomy of each animal at each ICS (aortic root, LV outflow tract, LV, and apex). When concurrent study protocols permitted, additional pre-arrest scans were obtained in an effort to increase the number of images available for analysis. Asphyxiated cardiac arrest was then induced and cardiac standstill was confirmed with transthoracic echocardiography (TTE). At the time of arrest, the same ICSs (as described above) were again scanned with the bladder scanner. Finally, when possible within the confines of concurrent protocols, the aforementioned ICSs were scanned during late arrest with several scans performed at each ICS. For each ICS scan, the BladderScan Prime Plus recorded 12 B-mode images (the transducer’s biplane imaging generates 6 orthogonal image pairs). To determine volume, the bladder scanner’s AI generates border tracings which are superimposed on each of the 12 saved images (see Fig. 2). A single scan (12 images) and volume calculation is completed in under 5 s.

Fig. 1

Experimental sonographic localization of the heart with the operator standing on the animal’s left side, the transducer placed perpendicular to the chest wall along the left sternal border. Image on the left shows the transducer is directly over a fluid-filled structure (heart). Echocardiography was performed to describe the underlying cardiac anatomy at each intercostal space (right image)

Fig. 2.

5th intercostal space in pre-arrest (left column), arrest (center column) and late arrest (right column). Top row displaying image capture by the bladder scanner (ac, tracing obscured), middle row displaying tracing by bladder scanner’s AI (df), and bottom row showing human reviewer tracing (gi). Images c, f, i show an area of echogenicity that the human reviewer interpreted as within the heart, in contrast to the bladder scanner tracing (image F)

To determine the performance of the bladder scanner’s AI in identifying and tracing a heart, its tracings were compared to 3 expert reviewers certified in focused cardiac ultrasound. The effect that the physiologic state (pre-arrest, arrest and late arrest) had on bladder scanner performance after recording 263 image sets (3156 images captured) was also explored. To approximate the cardiac view as seen along the left sternal border in humans, image analysis was narrowed to those intercostal spaces where the LV or LVOT were identified during TTE. The resulting sample included 69 image sets (828 images) for randomization (see Fig. 3). Assuming a normal distribution of data, we determined 150 images (50 images per state) would provide a sufficient sample (1.97 M pixels/image) while avoiding reviewer fatigue. Images were randomly selected and screened for inclusion as images with significant rib shadows obstructing the view of the heart were excluded (10/160 selected images). The images were coded according to the specific animal, state of arrest, intercostal space, and scan number. These coded images were then shuffled to ensure that each of the human reviewers were not aware of the state of arrest. The tracings created by the bladder scanner were obscured (Adobe Photoshop, CA, USA) to allow for a free trace by the reviewers. Three authors (PA, PO and RW) independently reviewed and then traced their own borders onto the images (Fig. 2, images G-I). The tracings of each human reviewer were compared to those generated by the bladder scan (D-F). Image segmentation analysis between the two sets of images (bladder scanner AI and pooled human reviewers) was determined via Sørensen–Dice coefficients [26]. The Sørensen–Dice index (SDI) is a formula that compares a known-truth sample against an estimated-truth sample. The formula (2 TP/(2 TP + FP + FN)) outputs a single value indicating how closely the two samples match up. A value of 1 represents perfect agreement, and a value of 0 means no agreement.

Fig. 3

Image selection process: of 3156 images, 828 met criteria for inclusion according to visualization of LV and/or LVOT. Images were randomized and then screened for improper transducer placement (i.e., are of interest obstructed by rib shadow) until there were 50 images for each physiologic state. A total of 160 images were screened, with 10 being excluded due to improper transducer placement. (LV: left ventricle; LVOT: left ventricular outflow tract)

Reviewer and bladder scanner tracings were processed using software developed and tested by co-author CA [27]. The software was programmed to crop tracings to the bladder scan image area as seen in Fig. 2. Next, the software filled the tracing outline and converted the images to black and white (white = pixel within tracing, black = pixel not within tracing). Finally, the software compared each black and white filled human reviewer tracing against its bladder scanner counterpart pixel-by-pixel to calculate the SDI. The image analysis is represented as true positive (yellow), true negative (white), false positive (red), false negative (green). The overlapping portion captured by the human reviewer and the bladder scanner is shown in yellow. The segment that was captured by the bladder scanner but not the human-reviewer is red, and the segment traced by the human reviewer but was not captured by the bladder scanner is green (see Fig. 4).

Fig. 4

The left shows a single image with both the bladder scanner tracing (blue) and a human reviewer tracing (magenta). The right shows the same image with a visual representation of true positive (yellow, 11.304%), true negative (white, 83.450%), false positive (red, 0.172%) and false negative (green, 5.074%) areas

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit


This article is autogenerated using RSS feeds and has not been created or edited by OA JF.

Click here for Source link (