Article
Eye-tracking: Comparison of gaze pattern behavior between orthopedists and pediatricians during identification of pediatric fractures
Search Medline for
Authors
-
Amal Chidda
- Cantonal Hospital Fribourg, Fribourg, Switzerland
-
Samy Rima
- University of Fribourg, Fribourg, Switzerland
-
Ines Raabe
- Cantonal Hospital Fribourg, Fribourg, Switzerland
-
Moritz Tannast
- Cantonal Hospital Fribourg, Fribourg, Switzerland
-
Angela Seidel
- Cantonal Hospital Fribourg, Fribourg, Switzerland
| Published: | October 21, 2024 |
|---|
Outline
Text
Objectives: As analyzing pediatric radiographs are always a challenge due to the growth plate and ossification centers. We wanted to compare the gaze pattern behaviors of orthopedists and pediatricians when interpreting pediatric foot and ankle radiographs and their diagnostic accuracy.
Our second aim was to assess the diagnostic accuracy of orthopedists and pediatricians with different levels of experience and Artificial Intelligence (AI) when interpreting pediatric foot and ankle radiographs.
Methods: Participants had to analyze 23 pediatric foot and ankle radiographs and had to assess if there was a fracture. We selected the radiographs from patients who consulted in the emergency room. There was 10 normal x-rays and 13 x-rays with fracture of the foot and ankle. Their gaze pattern behaviors were assessed through an eye-tracker Eyelink 1000 Plus and the data collected in EvdentIDE (Okazolab). We analyzed the following eye-tracking data: amplitude of saccades, fixation duration of saccades, and count of saccades per second. We evaluated the diagnostic accuracy via the sensitivity and the specificity. Our artificial intelligence program was Gleamer.
Results and conclusion: We had 25 participants (17 resident in orthopedic et 8 resident in pediatric). The orthopedists had a mean amplitude of saccades of 3.88 and a mean duration of saccades of 32.92 ms. The pediatricians had a mean amplitude of saccades of 3.14 and a mean duration of saccades of 28.62 ms. The difference between the two groups were significant (p<0.0001). The orthopedists had a number of 3.07 fixations per second while the pediatricians had a number of 2.38 fixations per second.
Figure 1 [Fig. 1] is an exemple of a scanpath plot of a participant.
The orthopedists mean diagnostic accuracy was 74.4%: sensitivity of 76.5% and specificity of 76.3% and the mean diagnostic accuracy of the pediatricians was 72.92% with a sensitivity of 72.1% and specificity of 73.75%. For the artificial intelligence, the sensitivity was 92% and the specificity 80%, with a mean of 87%.
In conclusion, there is a difference in the way of analyzing radiographs between orthopedic surgeon and pediatric physician. It would appear that larger amplitudes of saccades are associated with a higher diagnostic accuracy.
The artificial intelligence doesn't have a 100% diagnostic accuracy rate.
