Author Guide for Original Research
Thank you for considering the AJR for your research. This page provides guidance regarding key aspects of manuscript preparation that the editorial board will consider when evaluating your work. Please also refer to the AJR Guidelines for Authors for more information regarding manuscript types and formatting.
General guidance regarding retrospective reader-based studies
- Retrospective reader-based diagnostic performance studies should include independent readers. In general, consensus readings are discouraged as such readings do not mirror the independent readings that radiologists typically provide in clinical practice. It is more helpful to demonstrate the range of sensitivities and specificities attained by independent readers. A larger number of readers is encouraged to help adequately convey how diagnostic performance varies across radiologists. A larger number of readers also increases the authoritativeness of the work and will enhance the study’s overall impact.
- Retrospective reader-based studies associating a panel of imaging findings with a clinical endpoint should generally include a multivariable analysis to identify the strongest independent predictors. A lengthy list of imaging findings is of little practical use for reading images. Typically only 2 or 3 of these imaging findings drive diagnostic performance, while the remaining features provide confounding information and may be confusing or distracting for readers to apply. A regression or other model should be incorporated into the study analysis to inform the reader of the few features on which to rely in clinical practice.
- The title should provide an objective overview of the work that was performed, indicating the focus of the study's methods and results.
- The title should convey unique aspects of the study design (e.g., a multicenter study, based on a national survey).
- The Abstract should be able to stand on its own without the rest of the manuscript. For example, all Abstract results should have corresponding methods within the Abstract, and the Abstract conclusion should be fully supported by the Abstract results without also depending on additional results found only within the main manuscript.
- The Abstract results should provide numeric results and not just qualitative summaries of the data. Provide specific data in addition to p values.
- The Abstract conclusion should match the conclusion of the main manuscript.
- The Introduction should be focused on the specific research question and not serve as a broad review of the topic at hand. The Introduction will define an existing gap in knowledge and communicate how the authors intend to address that gap. For example, if the study aim is to compare the diagnostic performance of different DWI b values for a particular clinical condition, then the Introduction should be focused on existing knowledge regarding b-value optimization for the condition, including where there is uncertainty in the existing literature. In this same example, the Introduction should not provide a lengthy discussion about the clinical condition, nor even about the role of DWI in general for that condition, but instead should focus on issues around the specific question relating to b-value optimization.
- The Introduction should make clear the importance of the provided research question.
- Consistent with the above point, avoid extensive background on the topic, as may be encountered in a review article or textbook chapter, that is beyond the scope of the investigation. If the Introduction describes existing literature on the topic, then it should be clear how the present study is different from that literature.
- The Introduction should conclude with a clear statement of the study aim that closely matches the aim provided in the Abstract. All of the core elements within the study aim should be introduced in the Introduction prior to the aim statement itself. For example, if the study aim is to compare the PPV of T1- and T2-weighted MRI sequences for a given condition, then the Introduction should earlier make mention of PPV as well as of both T1- and T2-weighted sequences. In this regard, the purpose statement should pass a “cover test,” whereby the reader could anticipate this purpose by reading the remainder of the Introduction preceding it.
- This section is usually stratified into subsections (e.g., Patients, Image Acquisition, Image Interpretation, Reference Standard, and Statistical Analysis).
- In the first paragraph, affirm HIPAA compliance for studies performed in the U.S. as well as the study’s IRB status (e.g., IRB approval, IRB waiver, IRB oversight not required). If you assert that a study did not require IRB oversight, include a reason. For IRB-approved studies, indicate whether verbal or written informed consent was obtained from patients or if informed consent was waived.
- Clearly indicate the initial search criteria as well as the subsequent study inclusion and exclusion criteria. Provide the number of patients initially identified and the number of patients excluded for each reason that is provided. The selection process should be presented in a logical sequence whereby each step in the sequential flow represents a natural subset of remaining eligible patients up to that point. For example, if initially stating that only patients who underwent contrast-enhanced CT were selected, then the flow should not subsequently indicate exclusion of patients who did not undergo CT or who were unable to receive IV contrast media. Make clear whether the study was retrospective or prospective as well as whether or not patients were identified consecutively. Indicate the calendar dates encompassing the patient sample. Prepare a Figure 1 flowchart, ensuring that the patient selection process matches between this chart and the article text (i.e., presents exclusions that are listed in the same order and phrased in a comparable manner).
- Clearly indicate who conducted which steps of the study. Indicate if there was a study coordinator who searched institutional databases or the electronic health record, identified the cohort, or prepared image sets for retrospective review, etc.
- Be detailed when describing readers’ image evaluations. Indicate whether the readings were performed independently or in consensus and, if in consensus, how the consensus was achieved in the event of discrepancies. Indicate the readers’ backgrounds and years of posttraining experience. Indicate the details of any training session prior to the readings. For qualitative imaging features assessed by the readers, indicate the criteria for each possible option that the readers could assign for the feature. For quantitative imaging features, indicate how the readers performed the measurements. Indicate details of reader blinding (e.g., to patients’ histories, other diagnoses, earlier imaging examinations).
- Clearly indicate the reference standard. For diagnoses determined based on imaging or clinical follow-up, be explicit regarding the follow-up criteria.
- Provide methods for any post hoc subjective assessments performed by the readers in addition to their primary image analysis. For example, if after the formal reader evaluations and initial statistical analysis, a subjective assessment was performed to explore the causes of incorrect readings, then describe in the Methods section how this post hoc assessment was performed.
- For AI/ML studies of in-house tools that are not available to the public, provide a link to a repository (e.g., Github) with the code for the given tool.
- If there are earlier publications that may be similar to the present study, or if there are earlier studies with overlapping patient cohorts, then cite these earlier studies. Provide the nature of the similarity or the number of overlapping patients, and explain what is unique about the present study.
- This section is usually stratified into subsections based on related components of the analysis.
- All results should directly correspond with a step that was mentioned in the Methods section. This includes subjective, qualitative, or post hoc assessments performed after the initial reader evaluations (e.g., to explore causes of reader inaccuracies).
- Report all data of a particular type (e.g., all percentages or all measurements of a given quantitative imaging metric) to the same number of decimal places throughout the Results, which should match the number of decimal places used in the Tables and the Abstract.
- Provide numerators and denominators for all percentages, whether in the text or tables.
- Report p values ≥ .01 to 2 decimal places and p values < .01 to 3 decimal places. However, report p values in the range of .045 to .049 to 3 decimal places rather than rounding to p = .05. Report a maximum p value of p > .99 rather than p = 1.00. Report very small p values as p < .001 rather than reporting the exact p value to a larger number of decimal places. When reporting a single p value (aside from p < .001), report the exact p value rather than reporting the single p value as an inequality relative to a threshold (e.g., p < .01 or p > .05). Because p values are always less than 1, do not place a zero before the decimal point; this saves space in the text and tables. Also, use p rather than p value when the term appears in parentheses or in tables.
- When reporting p values, also report the numeric values for the actual data being compared by the p value. The p values themselves do not represent underlying data but rather indicate the confidence that a given finding is statistically significant. Likewise, avoid broad statements that two groups were “different” as this does not indicate the directionality or magnitude of the difference.
- Provide direct results and not judgments or interpretation of the results. For example, avoid statements such as “feature X was an excellent predictor of outcome Y.” In the absence of explicit criteria for excellent performance, this would not represent a direct observation but rather an interpretation of the data; others could consider the same results to not represent excellent performance.
- The first paragraph of the Discussion should provide a plainly stated summary of the study’s most important findings. Do not begin the manuscript with a repeat of the Introduction or with a broad overview of the topic.
- The Discussion should avoid a detailed repetition of numeric findings from the Results.
- The Discussion should also include a comparison of the findings with existing literature, explaining any discrepancies.
- The Discussion should remain focused on the data at hand, avoiding lengthy ancillary commentary that could have been written without having conducted the given study. The Discussion should not serve as a platform for the authors to advance their own views on the topic beyond what is supported by the provided data. For example, if the study aim is to compare two MRI sequences for a given diagnosis, and all of the provided data relate to those two sequences, then the Discussion should avoid a lengthy discussion of why MRI is preferred over other imaging modalities for the given diagnosis.
- The Discussion should not provide new information regarding the study sample that did not already appear earlier in the article, even if qualitative, subjective, or post hoc in nature. Such observations should be provided in the Results section, supported by corresponding material in the Methods section.
- The Discussion should make explicit the study’s impact, indicating how the findings will influence radiologic practice. Address why readers should care about the study’s findings.
- Avoid overstating the importance of the work or recommending widespread changes in clinical practice based on initial or small pilot studies. As appropriate, indicate the need for larger studies or other further validation.
- In general, the Discussion should include a Limitations paragraph as the second to last paragraph.
- The Discussion should conclude with a short paragraph highlighting the main finding and implication. This should mirror the Abstract conclusion.