Utility of 3D Printed Models Versus Cadaveric Pathology for Learning: Challenging Stated Preferences

The study employed a mixed methods approach of structured observation [16] and computer vision analysis [17]. Ethics approval was received from the Sydney Children’s Hospitals Network Human Research Ethics Committee (protocol LNR/18/SCHN/337).

Context

This research took place during an existing cardiac pathology practical skills workshop that focused on congenital heart deformities. The workshop was taught during an 8-week pediatric term, which is held four times per year for a rotating cohort as part of a 4-year graduate-entry medical program. At the time of the study, students undertook the term at the end of their third year or beginning of their fourth year. Each term, approximately a 70-student rotation attended the hospital for their structured teaching week, in which the workshop was held. The students had experience of learning with cadaveric specimens during their first and second year, but we are not aware of any formal teaching using 3D printed models.

Recruitment

The study took place during four workshops, held over two terms, from April to June 2019, with each workshop containing different students. Students were invited to participate in the study or attend the usual workshop. A total of 90 students consented to be a part of the study, which included an initial pilot study of 23 students to establish the protocol for the subsequent sessions (such as room layout and how the heart models were circulated) and were not included in the final analyses. No participants withdrew after consenting.

Study Design

Conventionally, the 30-min cardiac pathology practical skills workshops involved Pathology Department tutors (senior paediatric pathologists) using cadaveric heart samples to demonstrate the three-dimensional pathology. The 18 cadaveric congenital heart deformity specimens are regularly used by the Department of Pathology at the Children’s Hospital at Westmead for a practical workshop for medical students from the University of Sydney. These hearts are fixed and waxed, enabling interaction with the models, and were prepared with windows cut to highlight key features. The teaching collection includes examples of pathologies such as tetralogy of Fallot, ventricular septal defect, patent ductus arteriosus, and atrial septal defects (and combinations of pathologies).

For our study, 3D printed replicas of these cadaveric heart models were used alongside the current teaching practice. CT scans of the hearts were performed at 0.4-mm slice thickness using a Siemens SOMATOM Force. The CT scans were exported as DICOM files and segmented using Mimics v19.0 and converted into STL format. Hearts were printed at scale using fused deposition modelling (FDM) on the Stratasys Fortus 450mc in ASA using T12 tips at a layer height of 0.1778 mm with caustic acid–soluble support material SR-30. A total of 18 cadaveric hearts were scanned and replicated using 3D printing.

Students were informed about the study and consented to participate and be video recorded but were not told that the researchers were analyzing their handling of the heart models. This ensured that students acted authentically with the educational models and that their behavior was not influenced by the study’s objective.

Following a didactic lecture on the formation of congenital heart deformities from the lead tutor (a pathologist), cadaveric heart models and their 3D printed replicas (18 paired samples in total) were passed around the students (Fig. 1a). For each type of congenital heart deformity, the paired hearts were passed in plastic trays so that students could access whichever type of heart model they preferred. The paired hearts each had an identical tag, taken from the original cadaveric hearts, which described the patient and heart deformity. The students undertook unstructured exploratory learning as they inspected each heart type and passed along the trays as they finished their observations. Students were encouraged to use a pipe cleaner and their mobile phone lights to inspect the anatomic features of the heart models and identify defects in internal heart chamber walls.

Fig. 1figure 1

The experimental setup, including a the paired 3D printed (left) and cadaveric (right) hearts in a tray, b the room and recording configuration, and c a sample frame from a video recording

Data Collection

The primary mode of data collection was through video recordings of the participants as they interacted with the 3D printed and cadaveric models during the workshop. The room was set up so that tables were in a square shape with a camera rig in the center (Fig. 1b). A custom 3D printed jig housed five GoPro video cameras and was configured to capture all the students sitting around the tables (Fig. 1c). A total of 700 min of video footage was captured across the 4 workshops, each having a duration of approximately 40 min, with the students handling the hearts for approximately 20 min. All videos were filmed with 1920 × 1080 definition at a frame rate between 24 and 48 frames per second (depending on the GoPro used).

Data CodingStructured Observation

Structured observation was used by the researchers to systematically observe the behavior of students during their cardiac pathology practical skills workshop as they interacted with the 3D printed and cadaveric hearts. Most qualitative methods of observation require retrospective reports by participants (e.g., a post-workshop focus group or survey), or represent scenarios in which behavior can only be inferred [18]. Structured observation was selected for this study as it offers a way to validate findings from previous research through direct observation, rather than subjective measures such as inference, reflection, or participants’ opinions. The observation followed a set protocol to observe and record behavior, allowing the researchers to observe behavior directly. The researchers used a coding scheme to capture the actions and behaviors of students. The coding scheme captured interactions with the 3D printed and cadaveric hearts, including the sequence in which hearts were used by students (i.e., cadaveric or 3D printed heart first, where both were available), particular types of interaction (i.e., used a mobile phone light to highlight features of a heart, poked with a pipe cleaner or fingers to explore the deformities of a heart, and rough handling, e.g., dropping a heart on the table), and instances of collaboration (e.g., where a student discussed model features with another student, or consulted with a tutor for clarification). Each of these interactions was captured in a separate column of the coding scheme, based on the type of interaction and whether it occurred with a 3D printed or cadaveric heart. Instances where students did not have free choice for which heart to use were excluded from the data set for structured observation, i.e., where a pair of students could only reach one tray (as seen in Fig. 1) and each student’s decision to use a cadaveric or 3D printed model was forced as the other student had already picked up one of the available models.

Interactions were observed through recordings of the workshops, with four of the researchers (two PhDs, one Bachelor of Design/PhD student, one Bachelor of Engineering) collectively coding the data. During the initial coding, each researcher individually coded data for one participant. Codes were then compared between researchers, and discrepancies and outlying events were discussed. Triangulation between the researchers in this way assisted in validating the coding and helped the researchers establish a consistent coding style and align results. Once the researchers had aligned their coding practices, further coding was completed so that each participant’s data was coded by only one researcher. The researchers were co-located to maintain rigorous and consistent coding. This allowed the researchers to collectively discuss and interpret any ambiguous interactions that were observed.

In reporting the structured observation, each participant’s heart choice from each tray was represented by the letters P (printed model only), C (cadaveric only), PC (printed then cadaveric), and CP (cadaveric then printed). The choice of heart type was coded each time a tray was available for the student. For example, “CP, CP, PC, C” illustrates the choices made by a participant across four trays.

Computer Vision Analysis

Computer vision analysis was used to quantitatively investigate how students used the hearts. Object detection is a category of computer vision algorithm used to positively identify pre-defined objects within an image or data source. Object detection as a field of study has been accelerated by developments in deep learning and data analytics [19]. In the context of this study, object detection was used to identify two objects, cadaveric hearts, and 3D printed models, as seen in Fig. 1a.

The object detection algorithm employed in this study was YOLOV5 pre-configured object detection model, which was trained on 418 manually coded bounding boxed images of the hearts [17]. Once trained, the system was able to identify the location of either category of object, 3D printed or cadaveric, in the videos of the participants. Several iterations of the dataset’s design were required to achieve a final mean average precision of 0.93 measured at an intersection of union value of 0.5 over 300 epochs, including two augmentations to the dataset (brightness and blur) [20]. The outcome of this process is a model that is 94% accurate at positively identifying 3D printed or cadaveric hearts in the videos of the participants. This model was then applied to all 700 min of the footage and used to generate the analyses presented in the results. Notably, this allowed the researchers to measure the quantity of time that each model type was used. Unlike the structured observation, the computer vision analysis was unable to discriminate between when students had free choice over their heart type.

Data Analysis

Wilcoxon matched pairs signed rank test was used to compare student preferences (time held) in the four sessions and was run on GraphPad Prism version 9.1.2 for Windows (San Diego, CA, USA). To assess student interactions and collaboration, we assumed a null hypothesis that students would choose 3D printed and cadaveric hearts at equal rates (i.e., 50%:50%). We used IBM SPSS version 25 for Windows (Armonk, NY, USA) and tested goodness of fit using the chi-squared statistic to compare observed vs. expected counts based on whether students first picked up a 3D printed or cadaveric heart.

留言 (0)

沒有登入
gif