Natural Language Processing Algorithm Used for Staging Pulmonary Oncology from Free-Text Radiological Reports: "Including PET-CT and Validation Towards Clinical Use".

Natural language processing (NLP) can be used to process and structure free text, such as (free text) radiological reports. In radiology, it is important that reports are complete and accurate for clinical staging of, for instance, pulmonary oncology. A computed tomography (CT) or positron emission tomography (PET)-CT scan is of great importance in tumor staging, and NLP may be of additional value to the radiological report when used in the staging process as it may be able to extract the T and N stage of the 8th tumor-node-metastasis (TNM) classification system. The purpose of this study is to evaluate a new TN algorithm (TN-PET-CT) by adding a layer of metabolic activity to an already existing rule-based NLP algorithm (TN-CT). This new TN-PET-CT algorithm is capable of staging chest CT examinations as well as PET-CT scans. The study design made it possible to perform a subgroup analysis to test the external validation of the prior TN-CT algorithm. For information extraction and matching, pyContextNLP, SpaCy, and regular expressions were used. Overall TN accuracy score of the TN-PET-CT algorithm was 0.73 and 0.62 in the training and validation set (N = 63, N = 100). The external validation of the TN-CT classifier (N = 65) was 0.72. Overall, it is possible to adjust the TN-CT algorithm into a TN-PET-CT algorithm. However, outcomes highly depend on the accuracy of the report, the used vocabulary, and its context to express, for example, uncertainty. This is true for both the adjusted PET-CT algorithm and for the CT algorithm when applied in another hospital.

Active monitoring versus an abduction device for treatment of infants with centered dysplastic hips: study protocol for a randomized controlled trial (TReatment with Active Monitoring (TRAM)-Trial).

BACKGROUND: Developmental Dysplasia of the Hip (DDH) is one of the most common pediatric orthopedic disorders, affecting 1-3% of all newborns. The optimal treatment of centered DDH is currently under debate. This randomized controlled trial aims to study the (cost-)effectiveness of active monitoring versus abduction treatment for infants with centered DDH. METHODS: This is a multicenter, parallel-group, open-label, non-inferiority randomized controlled trial studying the (cost-)effectiveness of active monitoring versus abduction treatment for infants with centered DDH in fourteen hospitals in the Netherlands. In total, 800 infants with centered DDH (Graf IIa-/IIb/IIc), aged 10-16 weeks, will be randomly allocated to the active monitoring or abduction treatment group. Infants will be followed up until the age of 24 months. The primary outcome is the rate of normal hips, defined as an acetabular index lower than 25 degrees on an antero-posterior radiograph, at the age of 12 months. Secondary outcomes are the rate of normal hips at the age of 24 months, complications, time to hip normalization, the relation between baseline patient characteristics and the rate of normal hips, compliance, costs, cost-effectiveness, budget impact, health-related quality of life (HRQoL) of the infant, HRQoL of the parents/caregivers, and parent/caregiver satisfaction with the treatment protocol. DISCUSSION: The outcomes of this randomized controlled trial will contribute to improving current care-as-usual for infants with centered DDH. TRIAL REGISTRATION: Dutch Trial Register, NL9714, registered September 6, 2021. https://clinicaltrialregister.nl/en/trial/29596.

How Natural Language Processing Can Aid With Pulmonary Oncology Tumor Node Metastasis Staging From Free-Text Radiology Reports: Algorithm Development and Validation.

BACKGROUND: Natural language processing (NLP) is thought to be a promising solution to extract and store concepts from free text in a structured manner for data mining purposes. This is also true for radiology reports, which still consist mostly of free text. Accurate and complete reports are very important for clinical decision support, for instance, in oncological staging. As such, NLP can be a tool to structure the content of the radiology report, thereby increasing the report's value. OBJECTIVE: This study describes the implementation and validation of an N-stage classifier for pulmonary oncology. It is based on free-text radiological chest computed tomography reports according to the tumor, node, and metastasis (TNM) classification, which has been added to the already existing T-stage classifier to create a combined TN-stage classifier. METHODS: SpaCy, PyContextNLP, and regular expressions were used for proper information extraction, after additional rules were set to accurately extract N-stage. RESULTS: The overall TN-stage classifier accuracy scores were 0.84 and 0.85, respectively, for the training (N=95) and validation (N=97) sets. This is comparable to the outcomes of the T-stage classifier (0.87-0.92). CONCLUSIONS: This study shows that NLP has potential in classifying pulmonary oncology from free-text radiological reports according to the TNM classification system as both the T- and N-stages can be extracted with high accuracy.

Postmortem Fetal Temperature Estimation with Magnetic Resonance Imaging: Apparent Diffusion Coefficient Measurements in the Vitreous Body and Cerebrospinal Fluid.

OBJECTIVES: Magnetic resonance imaging (MRI) is increasingly used in postmortem fetal imaging. Several factors influence the quality of MRI in this setting, such as small size, autolytic and maceration changes, and temperature. Knowing the fetal temperature at the time of scanning can improve the MRI interpretation. Temperature can be calculated using diffusion-weighted imaging with measurements of the apparent diffusion coefficient (ADC) in the cerebrospinal fluid (CSF). However, this is complicated by small ventricle size and hemorrhage and, therefore, may be unreliable in postmortem imaging. The current study evaluated the feasibility and reliability of using the ADC for temperature measurements of the vitreous body compared to that of CSF. MATERIALS AND METHODS: Two lambs were scanned postmortem at five different time points over 28 hours. Furthermore, 10 stillborn fetuses were scanned once, at 4 to 62 hours after birth. The temperature was measured with a digital thermometer and calculated using the ADCs of the vitreous body (lambs and fetuses) and CSF (fetuses). RESULTS: There was an excellent correlation between measured and calculated temperatures in vitreous bodies of lambs (r = 0.997, P < 0.001) and fetuses (r = 0.970, P < 0.001). The correlation between measured and calculated temperatures in the CSF was poor (r = 0.522, P = 0.122). CONCLUSION: The calculation of the temperature based on the ADC of the vitreous body is feasible and reliable for postmortem fetal imaging.

Postmortem Diffusion-Weighted Magnetic Resonance Imaging of the Brain in Perinatal Death: An Animal Control Study to Detect the Influence of Postmortem Interval.

OBJECTIVES: Diffusion-weighted imaging may be useful as part of a postmortem magnetic resonance imaging protocol. However, apart from the effect of temperature on apparent diffusion coefficient (ADC), normal postmortem ADC changes can influence the interpretation. Therefore, this study was conducted to evaluate the correlation between normal ADC changes and postmortem intervals (PMIs) and develop a reference standard for postmortem changes after temperature correction. MATERIALS AND METHODS: Six premature lambs were scanned at different PMIs. ADC values were measured at different parenchymal locations. Correlation and linear regression between ADC values and PMI were analyzed for all locations, both uncorrected and corrected for temperature. RESULTS: All locations showed a significant negative correlation between the PMI and ADC value, with (R2 = 0.581-0.837, P < 0.001) and without (R2 = 0.183-0.555, P < 0.001-0.018) temperature correction. CONCLUSIONS: The postmortem interval is negatively correlated with ADC values in the brain. A correlation coefficient for the PMI can be calculated after temperature correction to predict ADC changes. However, further research is required to evaluate its clinical application in humans.

Treatment of decentered developmental dysplasia of the hip under the age of 1 year: an evidence-based clinical practice guideline - Part 2.

Background and purpose: Diagnostics and treatment of developmental dysplasia of the hip (DDH) are highly variable in clinical practice. To obtain more uniform and evidence-based treatment pathways, we developed the 'Dutch guideline for DDH in children < 1 year'. This study describes recommendations for unstable and decentered hips. Materials and methods: The Appraisal of Guidelines for Research and Evaluation criteria (AGREE II) were applied. A systematic literature review was performed for six predefined guideline questions. Recommendations were developed, based on literature findings, as well as harms/benefits, patient/parent preferences, and costs (GRADE). Results: The systematic literature search resulted in 843 articles and 11 were included. Final guideline recommendations are (i) Pavlik harness is the preferred first step in the treatment of (sub) luxated hips; (ii) follow-up with ultrasound at 3-4 and 6-8 weeks; (iii) if no centered and stable hip after 6-8 weeks is present, closed reduction is indicated; (iv) if reduction is restricted by limited hip abduction, adductor tenotomy is indicated; (v) in case of open reduction, the anterior, anterolateral, or medial approach is advised, with the choice based on surgical preference and experience; (vi) after reduction (closed/open), a spica cast is advised for 12 weeks, followed by an abduction device in case of residual dysplasia. Interpretation: This study presents recommendations on the treatment of decentered DDH, based on the available literature and expert consensus, as Part 2 of the first official and national evidence-based 'Guideline for DDH in children < 1 year'. Part 1 describes the guideline sections on centered DDH in a separate article.

Treatment of centered developmental dysplasia of the hip under the age of 1 year: an evidence-based clinical practice guideline - Part 1.

Despite the high incidence of developmental dysplasia of the hip (DDH), treatment is very diverse. Therefore, the Dutch Orthopedic Society developed a clinical practice guideline with recommendations for optimal and uniform treatment of DDH. This article summarizes the guideline on centered DDH (i.e. Graf types 2A-C). The guideline development followed the criteria of Appraisal of Guidelines for Research and Evaluation II. A systematic literature review was performed to identify randomized controlled trials and comparative cohort studies including children <1 year with centered DDH. Articles were included that compared (1) treatment with observation, (2) different abduction devices, (3) follow-up frequencies, and (4) discontinuation methods. Recommendations were based on Grading Recommendations Assessment, Development, and Evaluation, which included the literature, clinical experience and consensus, patient and parent comfort, and costs. Out of 430 potentially relevant articles, 5 comparative studies were included. Final guideline recommendations were (1) initially observe 3-month-old patients with centered DDH, start abduction treatment if the hip does not normalize after 6-12 weeks; (2) prescribe a Pavlik harness to children <6 months with persisting DDH on repeated ultrasonography, consider alternative abduction devices for children >6 months; (3) assess patients every 6 weeks; and (4) discontinue the abduction device when the hip has normalized or when the child is 12 months. This paper presents a summary of part 1 of the first evidence-based guideline for treatment of centered DDH in children <1 year. Part 2 presents the guideline on decentered DDH in a separate article.

Structured reporting in radiology: a systematic review to explore its potential.

OBJECTIVES: Structured reporting (SR) in radiology reporting is suggested to be a promising tool in clinical practice. In order to implement such an emerging innovation, it is necessary to verify that radiology reporting can benefit from SR. Therefore, the purpose of this systematic review is to explore the level of evidence of structured reporting in radiology. Additionally, this review provides an overview on the current status of SR in radiology. METHODS: A narrative systematic review was conducted, searching PubMed, Embase, and the Cochrane Library using the syntax 'radiol*' AND 'structur*' AND 'report*'. Structured reporting was divided in SR level 1, structured layout (use of templates and checklists), and SR level 2, structured content (a drop-down menu, point-and-click or clickable decision trees). Two reviewers screened the search results and included all quantitative experimental studies that discussed SR in radiology. A thematic analysis was performed to appraise the evidence level. RESULTS: The search resulted in 63 relevant full text articles out of a total of 8561 articles. Thematic analysis resulted in 44 SR level 1 and 19 level 2 reports. Only one paper was scored as highest level of evidence, which concerned a double cohort study with randomized trial design. CONCLUSION: The level of evidence for implementing SR in radiology is still low and outcomes should be interpreted with caution. KEY POINTS: • Structured reporting is increasingly being used in radiology, especially in abdominal and neuroradiological CT and MRI reports. • SR can be subdivided into structured layout (SR level 1) and structured content (SR level 2), in which the first is defined as being a template in which the reporter has to report; the latter is an IT-based manner in which the content of the radiology report can be inserted and displayed into the report. • Despite the extensive amount of research on the subject of structured reporting, the level of evidence is low.

Holistic processing only? The role of the right fusiform face area in radiological expertise.

Radiologists can visually detect abnormalities on radiographs within 2s, a process that resembles holistic visual processing of faces. Interestingly, there is empirical evidence using functional magnetic resonance imaging (fMRI) for the involvement of the right fusiform face area (FFA) in visual-expertise tasks such as radiological image interpretation. The speed by which stimuli (e.g., faces, abnormalities) are recognized is an important characteristic of holistic processing. However, evidence for the involvement of the right FFA in holistic processing in radiology comes mostly from short or artificial tasks in which the quick, 'holistic' mode of diagnostic processing is not contrasted with the slower 'search-to-find' mode. In our fMRI study, we hypothesized that the right FFA responds selectively to the 'holistic' mode of diagnostic processing and less so to the 'search-to-find' mode. Eleven laypeople and 17 radiologists in training diagnosed 66 radiographs in 2s each (holistic mode) and subsequently checked their diagnosis in an extended (10-s) period (search-to-find mode). During data analysis, we first identified individual regions of interest (ROIs) for the right FFA using a localizer task. Then we employed ROI-based ANOVAs and obtained tentative support for the hypothesis that the right FFA shows more activation for radiologists in training versus laypeople, in particular in the holistic mode (i.e., during 2s trials), and less so in the search-to-find mode (i.e., during 10-s trials). No significant correlation was found between diagnostic performance (diagnostic accuracy) and brain-activation level within the right FFA for both, short-presentation and long-presentation diagnostic trials. Our results provide tentative evidence from a diagnostic-reasoning task that the FFA supports the holistic processing of visual stimuli in participants' expertise domain.

T-staging pulmonary oncology from radiological reports using natural language processing: translating into a multi-language setting.

BACKGROUND: In the era of datafication, it is important that medical data are accurate and structured for multiple applications. Especially data for oncological staging need to be accurate to stage and treat a patient, as well as population-level surveillance and outcome assessment. To support data extraction from free-text radiological reports, Dutch natural language processing (NLP) algorithm was built to quantify T-stage of pulmonary tumors according to the tumor node metastasis (TNM) classification. This structuring tool was translated and validated on English radiological free-text reports. A rule-based algorithm to classify T-stage was trained and validated on, respectively, 200 and 225 English free-text radiological reports from diagnostic computed tomography (CT) obtained for staging of patients with lung cancer. The automated T-stage extracted by the algorithm from the report was compared to manual staging. A graphical user interface was built for training purposes to visualize the results of the algorithm by highlighting the extracted concepts and its modifying context. RESULTS: Accuracy of the T-stage classifier was 0.89 in the validation set, 0.84 when considering the T-substages, and 0.76 when only considering tumor size. Results were comparable with the Dutch results (respectively, 0.88, 0.89 and 0.79). Most errors were made due to ambiguity issues that could not be solved by the rule-based nature of the algorithm. CONCLUSIONS: NLP can be successfully applied for staging lung cancer from free-text radiological reports in different languages. Focused introduction of machine learning should be introduced in a hybrid approach to improve performance.

Modelling growth curves of the normal infant's mandible: 3D measurements using computed tomography.

OBJECTIVES: Data on normal mandibular development in the infant is lacking though essential to understand normal growth patterns and to discriminate abnormal growth. The aim of this study was to provide normal linear measurements of the mandible using computed tomography performed in infants from 0 to 2 years of age. MATERIAL AND METHODS: 3D voxel software was used to calculate mandibular body length, mandibular ramus length, bicondylar width, bigonial width and the gonial angle. Intra- and inter-rater reliability was assessed for these measurements. They were found to be sufficient for all distances; intra-class correlation coefficients were all above 0.9. Regression analysis for growth modelling was performed. RESULTS: In this multi-centre retrospective study, 109 CT scans were found eligible that were performed for various reasons (e.g. trauma, craniosynostosis, craniofacial abscesses). Craniosynostosis patients had larger mandibular measurements compared to non-craniosynostosis patients and were therefore excluded. Fifty-one CT scans were analysed. CONCLUSIONS: Analysis showed that the mandible increases more in size vertically (the mandibular ramus) than horizontally (the mandibular body). Most of the mandibular growth occurs in the first 6 months. CLINICAL RELEVANCE: These growth models provide insight into normal mandibular development in the first 2 years of life. This reference data facilitates discrimination between normal and abnormal mandibular growth.

Natural Language Processing in Dutch Free Text Radiology Reports: Challenges in a Small Language Area Staging Pulmonary Oncology.

Reports are the standard way of communication between the radiologist and the referring clinician. Efforts are made to improve this communication by, for instance, introducing standardization and structured reporting. Natural Language Processing (NLP) is another promising tool which can improve and enhance the radiological report by processing free text. NLP as such adds structure to the report and exposes the information, which in turn can be used for further analysis. This paper describes pre-processing and processing steps and highlights important challenges to overcome in order to successfully implement a free text mining algorithm using NLP tools and machine learning in a small language area, like Dutch. A rule-based algorithm was constructed to classify T-stage of pulmonary oncology from the original free text radiological report, based on the items tumor size, presence and involvement according to the 8th TNM classification system. PyContextNLP, spaCy and regular expressions were used as tools to extract the correct information and process the free text. Overall accuracy of the algorithm for evaluating T-stage was 0,83 in the training set and 0,87 in the validation set, which shows that the approach in this pilot study is promising. Future research with larger datasets and external validation is needed to be able to introduce more machine learning approaches and perhaps to reduce required input efforts of domain-specific knowledge. However, a hybrid NLP approach will probably achieve the best results.

Structured CT reporting improves accuracy in diagnosing internal herniation after laparoscopic Roux-en-Y gastric bypass.

OBJECTIVES: To confirm that structured reporting of CT scans using ten signs in clinical practice leads to a better accuracy in diagnosing internal herniation (IH) after gastric bypass surgery, compared with free-text reporting. METHODS: In this prospective study, CT scans between June 1, 2017, and December 1, 2018, were included from a cohort of 2606 patients who had undergone laparoscopic gastric bypass surgery between January 1, 2011, and January 1, 2018. The CT scans were made for a suspicion of IH and structured reports were made using a standardised template with ten signs: (1) swirl sign, (2) small-bowel obstruction, (3) clustered loops, (4) mushroom sign, (5) hurricane eye sign, (6) small bowel behind the superior mesenteric artery, (7) right-sided anastomosis, (8) enlarged nodes, (9) venous congestion, and (10) mesenteric oedema. Furthermore, an overall impression of IH likelihood was given using a 5-point Likert scale. CT scans performed in 2011 until 2017, without structured reporting, were included for comparison. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy were calculated using two-way contingency tables; the chi-square test was used for calculating p value. Reoperation and 3-month follow-up were used as reference. RESULTS: A total of 174 CT scans with structured reporting and 289 CT scans without structured reporting were included. Sensitivity was 81.3% (95% CI, 67.7-94.8%) and 79.5% (95% CI, 67.6-91.5%), respectively (p = 0.854); specificity was 95.8% (95% CI, 92.5-99.1%) and 88.6% (95% CI, 84.6-92.6%), respectively (p = 0.016); PPV was 81.3% (95% CI, 67.7-94.8%) and 55.6% (95% CI, 43.3-67.8%), respectively (p = 0.014); NPV was 95.8% (95% CI, 92.5-99.1%) and 96.0% (95% CI, 93.5-98.6%), respectively (p = 0.909); and accuracy was 93.1% (95% CI, 88.0-96.2%) and 87.2% (95% CI, 82.7-90.7%), respectively (p = 0.045). CONCLUSION: Structured reporting for the diagnosis of internal herniation after gastric bypass surgery improves accuracy and can be implemented in clinical practice with good results. KEY POINTS: • Ten signs are used to aid CT diagnosis of internal herniation after gastric bypass surgery. • Structured reporting increases specificity and positive predictive value and thereby prevents unnecessary reoperations in patients without internal herniation. • Structured reporting by means of a standardised template can help less experienced readers.

Redefining the structure of structured reporting in radiology.

Structured reporting is advocated as a means of improving reporting in radiology to the ultimate benefit of both radiological and clinical practice. Several large initiatives are currently evaluating its potential. However, with numerous characterizations of the term in circulation, "structured reporting" has become ambiguous and is often confused with "standardization," which may hamper proper evaluation and implementation in clinical practice. This paper provides an overview of interpretations of structured reporting and proposes a clear definition that differentiates structured reporting from standardization. Only a clear uniform definition facilitates evidence-based implementation, enables evaluation of its separate components, and supports (meta-)analyses of literature reports.

Chest X-ray evaluation training: impact of normal and abnormal image ratio and instructional sequence.

CONTEXT: Medical image perception training generally focuses on abnormalities, whereas normal images are more prevalent in medical practice. Furthermore, instructional sequences that let students practice prior to expert instruction (inductive) may lead to improved performance compared with methods that give students expert instruction before practice (deductive). This study investigates the effects of the proportion of normal images and practice-instruction order on learning to interpret medical images. It is hypothesised that manipulation of the proportion of normal images will lead to a sensitivity-specificity trade-off and that students in practice-first (inductive) conditons need more time per practice case but will correctly identify more test cases. METHODS: Third-year medical students (n = 103) learned radiograph interpretation by practising cases with, respectively, 30% or 70% normal radiographs prior to expert instruction (practice-first order) or after expert instruction (instruction-first order). After training, students performed a test (60% normal) and sensitivity (% of correctly identified abnormal radiographs), specificity (% of correctly identified normal radiographs), diagnostic performance (% of correct diagnoses) and case duration were measured. RESULTS: The conditions with 30% of normal images scored higher on sensitivity but the conditions with 70% of normal images scored higher on specificity, indicating a sensitivity and specificity trade-off. Those who participated in inductive conditions took less time per practice case but more per test case. They had similar test sensitivity, but scored lower on test specificity. CONCLUSIONS: The proportion of normal images impacted the sensitivity-specificity trade-off. This trade-off should be an important consideration for the alignment of training with future practice. Furthermore, the deductive conditions unexpectedly scored higher on specificity when participants took less time per case. An inductive approach did not lead to higher diagnostic performance, possibly because participants might already have relevant prior knowledge. Deductive approaches are therefore advised for the training of advanced learners.

Does the Use of a Checklist Help Medical Students in the Detection of Abnormalities on a Chest Radiograph?

The interpretation of chest radiographs is a complex task that is prone to diagnostic error, especially for medical students. The aim of this study is to investigate the extent to which medical students benefit from the use of a checklist regarding the detection of abnormalities on a chest radiograph. We developed a checklist based on literature and interviews with experienced thorax radiologists. Forty medical students in the clinical phase assessed 18 chest radiographs during a computer test, either with (n = 20) or without (n = 20) the checklist. We measured performance and asked participants for feedback using a survey. Participants that used a checklist detected more abnormalities on images with multiple abnormalities (M = 50.1%) than participants that could not use a checklist (M = 41.9%), p = 0.04. The post-experimental survey shows that on average, participants considered the checklist helpful (M = 3.25 on a five-point scale), but also time consuming (M = 3.30 on a five-point scale). In conclusion, a checklist can help medical students to detect abnormalities in chest radiographs. Moreover, students tend to appreciate the use of a checklist as a helpful tool during the interpretation of a chest radiograph. Therefore, a checklist is a potentially important tool to improve radiology education in the medical curriculum.

What We Do and Do Not Know about Teaching Medical Image Interpretation.

Educators in medical image interpretation have difficulty finding scientific evidence as to how they should design their instruction. We review and comment on 81 papers that investigated instructional design in medical image interpretation. We distinguish between studies that evaluated complete offline courses and curricula, studies that evaluated e-learning modules, and studies that evaluated specific educational interventions. Twenty-three percent of all studies evaluated the implementation of complete courses or curricula, and 44% of the studies evaluated the implementation of e-learning modules. We argue that these studies have encouraging results but provide little information for educators: too many differences exist between conditions to unambiguously attribute the learning effects to specific instructional techniques. Moreover, concepts are not uniformly defined and methodological weaknesses further limit the usefulness of evidence provided by these studies. Thirty-two percent of the studies evaluated a specific interventional technique. We discuss three theoretical frameworks that informed these studies: diagnostic reasoning, cognitive schemas and study strategies. Research on diagnostic reasoning suggests teaching students to start with non-analytic reasoning and subsequently applying analytic reasoning, but little is known on how to train non-analytic reasoning. Research on cognitive schemas investigated activities that help the development of appropriate cognitive schemas. Finally, research on study strategies supports the effectiveness of practice testing, but more study strategies could be applicable to learning medical image interpretation. Our commentary highlights the value of evaluating specific instructional techniques, but further evidence is required to optimally inform educators in medical image interpretation.

Teaching Systematic Viewing to Final-Year Medical Students Improves Systematicity but Not Coverage or Detection of Radiologic Abnormalities.

PURPOSE: Systematic viewing of images is widely advocated in radiology; it is expected to lead to complete coverage of images and consequently more detection of abnormalities. Evidence on the efficacy of teaching systematic viewing to students is conflicting. The aim of this study was to investigate the effects of teaching systematic viewing to final-year medical students on systematicity of viewing behavior, coverage of the image, and detection. METHODS: Final-year medical students (n = 60) viewed 10 chest radiographs in a first series before training and another 10 radiographs in a second series after training. Between series, students were taught basic chest radiographic viewing, in either a systematic or a nonsystematic manner. With eye tracking, systematicity (Levenshtein distances), coverage (percentage of image viewed), and detection (sensitivity and specificity) were measured. RESULTS: In a mixed two-by-two design, significantly higher sensitivity was found after training compared with before training (F1,55 = 6.68, P = .012, ηp2 = .11) but no significant effect for type of training (F1,55 = 1.24, P = .30) and no significant interaction effect (F1,55 = 0.12, P = .73). Thus, training in systematic viewing was not superior to training in nonsystematic viewing. A significant interaction of training type and time of viewing was found on systematicity (F1,49 = 20.0, P < .01, ηp2 = .29) in favor of the systematic viewing group. No significant interaction was found for coverage (F1,49 = 0.43, P = .51) or specificity (F1,55 = .124, P = .73). CONCLUSIONS: Both training types showed similar increases in sensitivity. Therefore, it might be advisable to pay less attention to systematic viewing and more attention to content, such as the radiologic appearances of diseases.

Systematic viewing in radiology: seeing more, missing less?

To prevent radiologists from overlooking lesions, radiology textbooks recommend "systematic viewing," a technique whereby anatomical areas are inspected in a fixed order. This would ensure complete inspection (full coverage) of the image and, in turn, improve diagnostic performance. To test this assumption, two experiments were performed. Both experiments investigated the relationship between systematic viewing, coverage, and diagnostic performance. Additionally, the first investigated whether systematic viewing increases with expertise; the second investigated whether novices benefit from full-coverage or systematic viewing training. In Experiment 1, 11 students, ten residents, and nine radiologists inspected five chest radiographs. Experiment 2 had 75 students undergo a training in either systematic, full-coverage (without being systematic) or non-systematic viewing. Eye movements and diagnostic performance were measured throughout both experiments. In Experiment 1, no significant correlations were found between systematic viewing and coverage, r = -.10, p = .62, and coverage and performance, r = -.06, p = .74. Experts were significantly more systematic than students F2,25 = 4.35, p = .02. In Experiment 2, significant correlations were found between systematic viewing and coverage, r = -.35, p < .01, but not between coverage and performance, r = .13, p = .31. Participants in the full-coverage training performed worse compared with both other groups, which did not differ between them, F2,71 = 3.95, p = .02. In conclusion, the data question the assumption that systematic viewing leads to increased coverage, and, consequently, to improved performance. Experts inspected cases more systematically, but students did not benefit from systematic viewing training.