To see, or not to see, that is the question: Applying a psychological perspective to supporting medical image processing of students

Abstract

The interpretation of medical images is an error-prone process (Pinto & Brunese, 2010) that can have serious consequences for the patients. For example, overlooked tumors can be life threatening. Also dental radiographs, which account for the largest proportion of radiographs in Germany (Bundesamt für Strahlenschutz, 2016), can contain such serious anomalies, for example, calcifications of the carotid artery (Friedlander et al., 2005). When physicians fail to identify anomalies, the errors may result from not looking at the anomalies (detection error), not recognizing the features of an anomaly (recognition error), or deciding against the relevance of suspicious features of the anomaly (decision error) (Kundel et al., 1978; Wu & Wolfe, 2019). To date, there are only few evaluated training methods to improve medical image processing and resulting diagnostic performance, and the evaluation of specific training methods is lacking (Kok et al., 2017). Therefore, this dissertation aims at developing and evaluating different training methods to support dental students in reading panoramic radiographs (Orthopantomogramm; OPT). Three studies evaluated three training methods that were expected to result in the improved detection of anomalies and more intense visual processing. This intense visual processing should be reflected in sooner, longer and more frequent fixations on anomalies. Unexpectedly, only the training method in Study 2 improved the detection of anomalies and none of the training methods led to the expected intensification of visual processing. Study 1 examined an individualized full coverage training to help dental students to search in all areas of the OPTs and thereby reduce the number of missed anomalies. Dental students either received the training for five OPTs in the intervention group (n = 38) or diagnosed five OPTs in the control group (n = 23) in a pre- and post-test setting. The training consisted of gaze feedback comparisons with eye movement visualizations of a peer model showing full coverage and their own gaze behavior. The results showed only small and not meaningful improvements in the detection rate for anomalies. Similarly, the training had a very small positive effect on the visual coverage rate. Gaze behavior regarding anomalies changed with training towards expanded visual search with shorter and fewer fixations on anomalies. The time to first fixation indicated a minor shift in attention towards anomalies located in the periphery. An exploratory analysis revealed that the dental students made five times more recognition and decision errors compared to detection errors, suggesting that detection errors addressed in this training are only a small part of the problem. Study 2 evaluated training that aimed to reduce recognition and decision errors by focusing on anomaly identification. In a crossover design, two training sessions that addressed either anomalies located in the periphery, for example maxillary sinuses and the neck, (peripheral anomalies) or the dentition (central anomalies) were tested simultaneously. In one group, dental students (n = 39) first received training for the recognition of peripheral anomalies and second training for the recognition of central anomalies, while the other group (n = 39) received the training sessions in the reversed order. The training method compared two OPTs with and without disease and two OPTs with the same disease. Additionally, colored highlights tagged the relevant areas in the OPTs and the instruction contained a verbal description of the characteristic features of anomalies. The results showed that the training was effective by improving diagnostic performance. The order of the training sessions seemed to affect the effectiveness and the learning times of the training seemed to influence the output. The training did not change the visual search behavior. Study 3 investigated training with eye movement modeling examples (EMME) designed to combine visual search strategies and object identification. In the intervention group, dental students (n = 42) saw three EMME videos from experts with didactical verbal explanations between the pre- and the post-tests. Dental students in the control group (n = 41) only performed the pre- and the post-test. In an online study of 31 dental students, the study was replicated with a second evaluation of the training without measuring eye movements. However, the training did not improve the detection of anomalies in either of the experiments. Students’ visual search behavior did not change for visual coverage rate, but the intervention led to shorter, fewer, and later fixations on anomalies. Exploratory analysis confirmed the findings from Study 1 on the distribution of error types with less than 20% detection errors. The results of these studies suggest that the training method which focuses on anomaly identification (Study 2) is effective when recognition and decision errors dominate. This training method provides knowledge about the visual properties of anomalies and their discrimination (Study 2). In contrast, training methods that teach visual search strategies (Studies 1 and 3) do not appear to be beneficial under these circumstances. However, further research is needed to investigate possible long-term effects of search strategy training (Kramer et al., 2019). Changes in eye movements indicate that the training may trigger a change in cognition that could lead to improved diagnostic performance after some time. This dissertation implies that the type of errors needs to be considered before applying training in dental education. Comparison of radiographic images, as examined in Study 2, provides a supportive training method that could be easily implemented in university teaching and improves diagnostic performance.