This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Medical Education, is properly cited. The complete bibliographic information, a link to the original publication on https://mededu.jmir.org/, as well as this copyright and license information must be included.
Accurate interpretation of a 12-lead electrocardiogram (ECG) demands high levels of skill and expertise. Early training in medical school plays an important role in building the ECG interpretation skill. Thus, understanding how medical students perform the task of interpretation is important for improving this skill.
We aimed to use eye tracking as a tool to research how eye fixation can be used to gain a deeper understanding of how medical students interpret ECGs.
In total, 16 medical students were recruited to interpret 10 different ECGs each. Their eye movements were recorded using an eye tracker. Fixation heatmaps of where the students looked were generated from the collected data set. Statistical analysis was conducted on the fixation count and duration using the Mann-Whitney U test and the Kruskal-Wallis test.
The average percentage of correct interpretations was 55.63%, with an SD of 4.63%. After analyzing the average fixation duration, we found that medical students study the three lower leads (rhythm strips) the most using a top-down approach: lead II (mean=2727 ms, SD=456), followed by leads V1 (mean=1476 ms, SD=320) and V5 (mean=1301 ms, SD=236). We also found that medical students develop a personal system of interpretation that adapts to the nature and complexity of the diagnosis. In addition, we found that medical students consider some leads as their guiding point toward finding a hint leading to the correct interpretation.
The use of eye tracking successfully provides a quantitative explanation of how medical students learn to interpret a 12-lead ECG.
The electrocardiogram (ECG) is a graph that represents the electrical activity of the heart. The 12-lead ECG showcases this activity from 12 different “viewpoints” called leads. Although more than 300 million ECG tests are performed annually in the United States [
We present a quantitative study of ECG interpretation by 16 medical students. We requested from them to interpret 10 ECGs each while we collected their eye-tracking data using the eye-tracking methodology. We based our study on related works that looked at similar behavior in the general medical population. We sought to find a confirmation of our hypothesis that the order and the duration with which medical students look at specific areas across the ECG directly affect their final interpretation. We based our data analysis on understanding ECG interpretation within medical students using a number of different features, mainly areas of interest (AOIs), fixation count, fixation duration, time to first fixation (TTFF), and fixation revisitations. The results of the study uncover ECG interpretation insights among the population of medical students that confirm our hypothesis.
When eye tracking was first used to understand ECG interpretation, the aim was to gain insight into experts’ interpretation procedures. Bond et al [
However, a quantitative study alone might not be enough to unveil complex visual behaviors, such as ECG interpretation. Davies et al [
Other studies, such as the studies done by Wood et al [
We asked our research questions based on the reviewed body of work. We then formed our hypothesis accordingly. We based our hypothesis on the results of previous related studies, as well as on our experience in teaching ECG interpretation to medical students.
Research question 1:
Identifying patterns that medical students follow to interpret an ECG may explain the correct/incorrect interpretations they provide. Bond et al [
In their study, Bond et al [
Research question 2:
Davies et al [
Research question 3:
Several studies have analyzed the eye-tracking data at the ECG lead level [
Our hypothesis is that
As commented above on research questions 1 and 2, waveform abnormalities catch the attention of the interpreter. We speculate that these abnormalities certainly catch the attention of medical students, and especially that recognizing them is at the foundation of the ECG interpretation curriculum. What then differentiates an accurate interpretation from an inaccurate one is whether students focus on the right abnormalities. This therefore influences the order with which students look at different areas of the ECG across their interpretation period, as well as how much time they spend fixating on these areas. Choosing the right data analysis tools and methodologies would be critical in obtaining accurate results to confirm or deny our hypothesis. As commented above on research question 3, finding the most suitable level of specificity to analyze the eye-tracking data for the interpreter is also part of the analysis. This is especially pertinent to understanding the dynamics of attention at the ECG lead level as well as the waveform level.
We curated 10 ECGs of different heart arrhythmias with the assistance of a cardiology consultant. Since the experiment was designed for medical students, we made sure that these arrhythmias were all studied in traditional medical school curricula. Although some ECGs may be difficult to interpret, the students participating in the study had all the necessary medical background to make the correct diagnosis. The ECG set included in the experiment was either from the personal collection of the cardiology consultant involved in the design of the experiment or from the open ECG data set available from the PhysioNet database [
We involved an expert in ECG interpretation as well as an expert in human-computer interaction in designing the experiment. We also referred to similar experiments described in the Related Works section. Involving these stakeholders in the design of the experiment ensured the refinement and fine-tuning of certain parameters important to lower its inherent bias. The experiment design was in the form of a digitized, short multiple-choice-question (MCQ) miniquiz, whereby medical students were exposed to the 10 ECGs on a computer screen one after the other, followed by the respective diagnosis question. This format is important as medical students are accustomed to it in their examinations. The exchange between the stakeholders involved in the design of the experiment yielded the following critical points in the experiment design:
Randomizing the order in which the ECGs are displayed for each participant. This contributes to offsetting learning effects. This was fundamentally pertinent since (1) the task was brief (as discussed in the next point) and (2) there were many repetitions of the same task (mainly 10 repetitions of the interpretation task) [
Restricting the duration for which participants are allowed to look at the ECG to 30 seconds per ECG. Deciding on this timing stemmed from the following reasons:
The primary aim of the study was not to measure participants' performance over the accuracy of interpretation but rather to unveil visual scanning dynamics that would explain the medical students’ reasoning behind their final diagnosis provided after performing the ECG interpretation task.
The time allowed for scanning an ECG was found to have no statistically significant effect over the accuracy of interpretation per Davies [
The quality of the collected eye-tracking data: Restricting the time allowed for students to look at an ECG forces them to spend their allowed timespan wisely focusing on selected areas of the ECG. The students are therefore more attentive and wander less around the different areas of the ECG. This improves the quality of the data obtained, as will be explained more in the forthcoming sections.
Medical students in the classes of 2022 and 2023 enrolled in the cardiology module within the 4-year medical program curriculum were invited to participate in the study. One of the main learning objectives was the 12-lead ECG interpretation in accordance with the US medical curriculum [
The students listened first to a 5-minute briefing, where they were introduced to the goal of the experiment, the number of ECGs they would interpret, and how they should interact with the interface. The interface was minimalistic in the sense that the 12-lead ECG image took up the whole computer screen and automatically changed to the next ECG after 30 seconds elapsed. The MCQ miniquiz asking about the ECG diagnosis was also minimalistic, whereby each question was centered in the screen and everything around it was grayed except the Next button to move to the next ECG. The students were allowed as much time as they needed to think about the ECG diagnosis, but in all cases, the time did not exceed 10 seconds to give their answer. The students were also encouraged to ask any question regarding the experiment or the setup. Once all the doubts were cleared, the students were set to familiarize themselves with the setup.
The students familiarized themselves with the setup by calibrating their eyes with the eye tracker's infrared sensors. Once their eyes were calibrated to the eye tracker, they proceeded to the interpretation task. The students were allowed 30 seconds to look at an ECG before being prompted for its diagnosis. Once they provided their diagnosis for the 10 ECGs, we performed an informal interview where we asked them about the overall difficulty of the interpretation task, the experiment design, and how they thought they did on the questions. Some of the students requested to check their answers.
A Tobii Pro X2-60 eye tracker and iMotions version 8.1 software (iMotions) were used to record eye movements with a sampling rate of 60 Hz (±1 Hz). Key strokes and mouse clicks were recorded to collect the students’ responses to the MCQs. The eye-tracking experiment was conducted on a 25-inch diagonal laptop monitor with a resolution of 1366 × 768 pixels.
The institutional review board approval for this study was granted by the ethics board of both the Qatar Biomedical Research Institute at Hamad Bin Khalifa University [
We generate fixation heatmaps using the collected eye-tracking data as a preliminary step of data exploration. Heatmaps helped us visualize the general trends in how medical students proceed with an ECG interpretation. The observation of heatmaps contributed toward finding an answer to both research questions 1 and 2, since they indicated where medical students focus more and what they ignore on an ECG. The heatmap images of all the ECGs are available in
We defined two AOI types. These two types were informed by the previous work that Davies et al [
Grid-based AOI applied to the NSR ECG. AOI: area of interest; aVF: augmented vector foot; aVL: augmented vector left; aVR: augmented vector right; ECG: electrocardiogram; NSR: normal sinus rhythm.
Long vs short AOI applied to the NSR ECG. AOI: area of interest; aVF: augmented vector foot; aVL: augmented vector left; aVR: augmented vector right; ECG: electrocardiogram; NSR: normal sinus rhythm.
To systematically understand how medical students proceed to interpreting an ECG, we used eye tracking. Eye tracking is a methodology that aims to quantify visual behavior when performing a specific task in order to understand the nuances in locus and levels of attention of interpreters [
To measure the significance in the fixation behavior of the medical students around different AOIs on the 12-lead ECG, we use two different statistical tests. These tests were curated in accordance with the nature of the collected eye-tracking data, as well as the areas being compared.
Demographics of the eye-tracking study participants.
Feature | Demographics |
Age (years) | 21 (n=5); 22 (n=1); 23 (n=5); 24 (n=5) |
Gender | Male: 15; female: 1 |
Country | Palestine (n=4); Jordan (n=6); South Korea (n=1); Egypt (n=2); Lebanon (n=2); Libya (n=1) |
Class | 2023 (n=6); 2022 (n=10) |
We present the results for the fixation count, the time spent fixating, the average TTFF, and the average fixation revisitations on each ECG lead over all 160 interpretations (16 participants, with 10 ECGs to interpret for each student) in
Results of the fixation count, the time spent fixating, the average TTFF,a and the average fixation revisitations on each ECGb lead for all interpreters.
ECG lead | Fixation count | Time spent fixating (msc) | TTFF (ms) | Fixation revisitation count | |||||||
|
Mean (μ) | SD (σ) | Mean (μ) | SD (σ) | Mean (μ) | SD (σ) | Mean (μ) | SD (σ) | |||
II | 54 | 9 | 2727 | 456 | 4225 | 213 | 2 | .85 | |||
V5 | 28 | 5 | 1301 | 236 | 12,639 | 447 | 1 | .82 | |||
V1 | 24 | 6 | 1476 | 320 | 9552 | 557 | 2 | .73 | |||
V3 | 16 | 5 | 974 | 103 | 10,634 | 557 | 4 | .51 | |||
V2 | 15 | 3 | 904 | 54 | 9898 | 447 | 4 | .65 | |||
aVFd | 14 | 3 | 890 | 51 | 6233 | 88 | 4 | .62 | |||
V4 right | 13 | 2 | 972 | 66 | 7035 | 222 | 3 | .14 | |||
3 | 11 | 1 | 886 | 54 | 9222 | 111 | 3 | .86 | |||
V3 right | 11 | 1 | 868 | 47 | 6064 | 93 | 3 | .37 | |||
2 | 11 | 1 | 1017 | 199 | 8683 | 38 | 4 | .47 | |||
aVLe | 10 | 1 | 723 | 47 | 3602 | 103 | 3 | .82 | |||
V7 | 8 | 1 | 437 | 48 | 10,513 | 601 | 2 | .92 | |||
1 | 8 | 1 | 707 | 42 | 11,393 | 587 | 2 | .80 | |||
V8 | 8 | 1 | 435 | 53 | 13,076 | 598 | 1 | .51 | |||
aVRf | 8 | 1 | 456 | 49 | 11,569 | 615 | 1 | .85 | |||
V6 | 7 | 1 | 325 | 49 | 18,641 | 566 | 1 | .59 | |||
V4 | 6 | 1 | 312 | 32 | 19,460 | 333 | 1 | .20 | |||
Info |
4 | 1 | 64 | 16 | 20,394 | 887 | 0 | .57 |
aTTFF: time to first fixation.
bECG: electrocardiogram.
cms: milliseconds.
daVF: augmented vector foot.
eaVL: augmented vector left.
faVR: augmented vector right.
Percentage (%) of correct interpretations per ECG for all medical students. AFib: atrial fibrillation; AV block: complete heart block; ECG: electrocardiogram; LBBB: left bundle branch block; NSR: normal sinus rhythm; STEMI: ST-segment elevation myocardial infarction; VTach: ventricular tachycardia; WPW: Wolf-Parkinson-White.
We analyzed the collected eye-tracking data. We started by defining key parameters on which the data analysis was founded, mainly AOIs. This contributed to answering research question 3. We then proceeded to find an answer to research question 1 by analyzing the eye-tracking parameters. These parameters are referred to in the hypothesis as the “duration for which medical students look at specific areas of the ECG.” They are the fixation count and the fixation duration. “Specific areas of the ECG,” as mentioned in the hypothesis therefore refer to the defined AOIs. We finally analyzed the correlation between these four eye-tracking features to test our hypothesis.
We selected the fixation count and the fixation duration as the main features guiding our analysis to answer research question 1.
Density distribution of the average fixation duration for all ECG interpreters and all ECGs. AFib: atrial fibrillation; AV block: complete heart block; ECG: electrocardiogram; LBBB: left bundle branch block; ms: milliseconds; NSR: normal sinus rhythm; STEMI: ST-segment elevation myocardial infarction; VTach: ventricular tachycardia; WPW: Wolf-Parkinson-White.
(1) Between the rhythm strip vs the short-lead signals: We used the Mann-Whitney U test, as discussed in the Statistical Analysis section. As observed in
(2) Between the 24 defined AOIs in selected ECGs: We used the Kruskal-Wallis test, as discussed in the Statistical Analysis section. The results from applying the test indicated that all the ECGs had a
Mann-Whitney U test conducted on the fixation count parameter of selected ECGs.a
ECG | Mean fixation count in rhythm strips | Mean fixation count in shorter strips | |
NSRb |
|
169.875 | 98.750 |
Atrial flutter |
|
182.625 | 54.125 |
VTachc | 1 | 118.000 | 131.875 |
WPWd | .958 | 144.500 | 145.625 |
Ventricular paced rhythm | .279 | 126.375 | 162.250 |
LBBBe | .161 | 143.250 | 115.75 |
STEMIf | .721 | 119.25 | 125.25 |
AV blockg | .248 | 157.875 | 117.875 |
*Values in italic are significant.
aECG: electrocardiogram.
bNSR: normal sinus rhythm.
cVTach: ventricular tachycardia.
dWPW: Wolf-Parkinson-White.
eLBBB: left bundle branch block.
fSTEMI: ST-segment elevation myocardial infarction.
gAV block: complete heart block.
We used the Pearson correlation coefficient, as discussed in the Materials and Methods section.
Pearson correlation coefficient for eye-tracking parameters over the 10 studied ECGs.a
|
Fixation count | Fixation duration | TTFFb | AOIc fixation revisitations |
Fixation count | —d | .81 | –.4 | .78 |
Fixation duration | — | — | –.36 | .53 |
TTFF | — | — | — | –.44 |
AOI fixation revisitations | — | — | — | — |
aECG: electrocardiogram.
bTTFF: time to first fixation.
cAOI: area of interest.
dNot applicable.
Here, we discuss the Results section. We mainly discuss the insights from the heatmaps as well as the analysis of the eye-tracking features.
As observed in all the heatmaps, medical students tend to lean toward one of two distinct behaviors for ECG interpretation. The nuance lies primarily in the areas where the fixations are concentrated.
1. In some ECGs, such as the NSR and the atrial flutter, eye fixations are concentrated over the lower area. This area contains the three leads (V1, II, and V5) referred to as rhythm strips.
2. In other ECGs, such as the ventricular paced rhythm, eye fixations are concentrated in the upper half, precisely over leads V1, V2, and V3.
The generated heatmaps indicate how complex and different the ECG interpretation approach is depending on the heart pathology observed. They also indicate that medical students adapt their interpretation behavior, depending on the nature of the ECG. The preliminary observations from the heatmaps show that medical students fixate more on abnormal areas of the ECG, such as atrial flutter waves, wide QRSs, elevated ST segments, and V-pacing spikes, regardless of whether their interpretation of the ECG is correct. These observations provide an answer to research question 2.
The eye-tracking features reported in the Results section and analyzed previously confirm the observations made when looking at heatmaps. The fixation count as well as the average fixation duration features demonstrate that medical students fixate the most on rhythm strips, mainly lead II, followed by leads V1 and V5. This might be because rhythm strips provide the best longitudinal analysis of the rate and rhythm over the entirety of the ECG. According to the TTFF and the fixation revisitations per lead, medical students’ natural tendency is to start their interpretation from the middle of the ECG. They then move to whatever lead they believe guides them to a correct interpretation. Specificities of their interpretation changes from one ECG to the other. Amidst the interpretation process, waveform abnormalities may catch their attention and divert their attention from one lead to another, provoking an increase in lead revisitations. These observations answer research question 1.
Research question 1: We recognized two main eye-tracking patterns that medical students often repeat during the process of ECG interpretation. The first pattern is that they start from the leads in the center of the ECG. The second pattern is that across the interpretation period, they refer primarily to the rhythm strips as the guiding leads toward finding an indicator for the appropriate diagnosis, as they spend most of the interpretation time fixating on these rhythm strips. This behavior of interpretation was also found by Davies et al [
Research question 2: The presence of one or multiple abnormal waveforms in the ECG does affect the ECG interpretation behavior of medical students. The abnormalities may be due to an actual heart rhythm irregularity or due to noise. These abnormalities usually derail the students from a systematic approach of interpretation. The abnormal signs catch their attention and therefore make them fixate longer and often switch between leads where these abnormalities occur. Davies et al [
Research question 3: We could not identify one suitable level of specificity to analyze the eye-tracking data. This is because the ECGs’ complexity of interpretation varies depending on the heart abnormality being tracked. However, the statistical analysis that we conducted offered an understanding of how to optimize AOIs to get meaning out of the eye-tracking data. The level of granularity varies between comparing rhythm strips against short-lead signals at one end of the spectrum. On the other end of the spectrum, the comparison is done by dividing the whole ECG into equal grid sizes in order to compare the eye-tracking feature among the grids. The challenge with grid-based analysis is that it is extremely granular and that the difference between grids is usually always significant. Grid-based analysis is suitable for understanding the-eye tracking behavior around specific waveform abnormalities.
We confirmed our hypothesis that the order with which and the duration for which medical students look at specific areas across the ECG directly affect their final interpretation. This was demonstrated through the distribution of the fixation count as well as the fixation duration across different AOIs over the ECG. Additionally, the TTFF as well as the AOI fixation revisitations proved that some ECG leads catch the attention of interpreters more than others. This is especially true when the lead contains an abnormal signal.
A major limitation that affected our work was the confounding effect of a common training framework for all participants. This was because all the participants recruited to the study received the same ECG interpretation training. This effect needs to be acknowledged as influencing the results, and therefore addressed. The study conducted by Breen et al [
The reported results are considered the first phase of a larger study. In future works, we plan on expanding the study by directly addressing the limitations mentioned in the previous section. We aim to increase the number of participants to be more diverse in terms of their educational background, demographics, and expertise in ECG interpretation. We aim to include more medical students, nurses, technicians, cardiology fellows, residents, and consultants. This increase in the number of participants will enable the examination of nuances in interpretation behavior across the whole medical practitioners’ spectrum. The prospects of this participant expansion include the ability to compare the results against related studies. This data will contribute to the development of a more detailed road map of how health practitioners proceed to interpreting ECGs.
We presented a quantitative analysis of the ECG interpretation behavior of medical students. We collected eye-tracking data of 16 medical students interpreting 10 ECGs each. This enabled us to analyze these data using a number of different eye-tracking features, mainly AOIs, fixation count, fixation duration, TTFF, and fixation revisitations. This was with the aim of answering three research questions: (1) whether we can recognize any interpretation patterns within the medical students’ population, (2) what elements in the ECG catch the attention of medical students during their interpretation process, and (3) and, finally, what the most suitable level of granularity is to analyze the eye-tracking data for optimal insights. We found that medical students often start their interpretation by fixating at the center of the ECG. This may be due to the central bias tendency toward fixating at the center of the observed image. This bias is widely observed in eye-tracking research, as reported by Bindemann et al [
Electrocardiograms and their definitions.
Heatmap for a Wolf-Parkinson-White syndrome electrocardiogram interpreted by medical students.
Heatmap for a ventricular tachycardia electrocardiogram interpreted by medical students.
Heatmap for a ventricular paced rhythm electrocardiogram interpreted by medical students.
Heatmap for an ST-segment elevation myocardial infarction electrocardiogram interpreted by medical students.
Heatmap for a normal sinus rhythm electrocardiogram interpreted by medical students.
Heatmap for a left bundle branch block electrocardiogram interpreted by medical students.
Heatmap for a hyperkalemia electrocardiogram interpreted by medical students.
Heatmap for a complete heart block electrocardiogram interpreted by medical students.
Heatmap for an atrial flutter electrocardiogram interpreted by medical students.
Heatmap for an atrial fibrillation electrocardiogram interpreted by medical students.
Eye-tracking parameter definitions.
Interpretation answers by medical students.
atrial fibrillation
area of interest
complete heart block
augmented vector foot
augmented vector left
atrioventricular nodal reentry tachycardia
augmented vector right
electrocardiogram
left bundle branch block
multiple-choice question
normal sinus rhythm
ST-segment elevation myocardial infarction
time to first fixation
ventricular tachycardia
Wolf-Parkinson-White
MTS would like to thank Dr Yahya Sqalli Houssaini and Ahmed Kachkach for the discussion and insights. MTS would also like to thank Kiara Heide from iMotions for the onboarding training. The authors would like to thank all the volunteering participants who contributed with their electrocardiogram interpretations.
MTS contributed to conceptualizing, designing, and conducting the practical experimentation. MTS also contributed to analyzing the data and writing the paper. DAT, MBE, and ME contributed to participants’ recruitment. DAT, MBE, and ME also contributed to the study’s conception, research methodology, and validation of the results. Finally, DAT, MBE, and ME revised the final manuscript.
None declared.