The use of cross-sectional imaging techniques is rapidly increasing in most parts of the world [10]. However, the value of these techniques is limited without thorough, accurate image interpretation and clear communication of imaging findings. Miscommunication between radiologists and referring clinicians may give rise to adverse clinical outcomes as well as increased costs. In a recent survey of opinions about radiology reports, only 50% of the responding clinicians and 37% of the responding radiologists agreed that “The language and style of radiology reports are mostly clear” [3]. Ninety-two percent of the responding referring physicians indicated that “It is the responsibility of the radiologist to adapt his or her style and choice of words” to make the report clear and comprehensible [3].

One way to improve communication is through the use of standardized reporting “templates,” which not only require the entry of specific information in a predefined order but also typically require or promote the use of standardized terminology. Standardized reporting schemes have been found to be clearer than and preferable to free-form unstructured reports. For example, Schwartz et al. [11] found that radiologists and referring physicians gave significantly higher satisfaction ratings for structured reports than for free-form reports of oncologic body CT examinations. Barbosa et al. [12] showed that structured reports of thyroid ultrasound examinations led to increased standardization of descriptors of thyroid findings and were preferred by 75% of the radiologists and 80% of the referring endocrinologists surveyed. Ghoshhajra et al. [13] showed that the agreement of radiologists and referring physicians regarding the number of vessels with significant stenosis on coronary CT angiography was improved by the use of structured reporting software that prompted radiologists to specify which vessels showed significant stenosis (κ = 0.31 vs 0.52). Furthermore, Karim et al. [14] showed that the average time needed for reporting imaging of aortic aneurysms was significantly reduced by the use of a structured reporting system that included, among other features, drop-down menus that facilitated predefined data entries and kept free text to a minimum. In a previously mentioned survey about radiology reports [3], 85% of the responding clinicians indicated that they preferred structured radiology reports and 67% indicated that they would prefer a standard lexicon of radiologic terms.

In our study, we found that implementation of a 5-point standardized lexicon to express diagnostic certainty led to a substantial reduction in the number of different expressions used by radiologists to indicate levels of certainty regarding the presence of ECE on prostate MRI. In other words, it led to greater standardization of reporting and therefore, as suggested by the prior studies discussed, likely reduced the chances of misunderstandings. With an AUC of 0.852, the accuracy of the certainty lexicon for the diagnosis of ECE on standard-of-care MRI was comparable with the accuracy found for the detection of ECE by MRI in the literature (reported AUCs, 0.62–0.87) [15–17]. Furthermore, during the lexicon period, the level of certainty for ECE stated in the MRI reports was related to the prevalence of ECE on histopathology.

Taken together, these findings suggest that referring physicians can trust that the probability of a specific finding expressed with the lexicon is reasonably accurate. A predefined lexicon of diagnostic certainty also enables the evaluation of diagnostic accuracy from routine clinical practice reports, which is not possible with free-form reports. This concept is similar to that which underpins the value of the BI-RADS lexicon in the field of mammography. Lehman et al. [18] found that after the implementation of that lexicon for screening mammography, sensitivity (0.50 vs 0.87), specificity (0.77 vs 0.89), and positive predictive value (0.43 vs 0.78) for the diagnosis of breast cancer by mammography significantly improved. In addition, Berg et al. [19] showed that training of physicians interpreting mammography in the use of the BI-RADS lexicon improved the biopsy rate for malignant lesions from 73% to 88%.

In liver MRI, a standardized 5-point certainty lexicon (liver imaging reporting and data system [LI-RADS]) has been introduced to assign liver lesions into one of five categories of estimated likelihood of hepatocellular carcinoma (1, < 5%; 2, 5–20%; 3, 21–70%; 4, 71–95%; 5, > 95%). Petruzzi et al. [20] evaluated the LI-RADS system in 100 abdominal MRI studies of patients under surveillance for hepatocellular carcinoma. The AUC for the diagnosis of hepato-cellular carcinoma was reported to be 0.949. Similarly, another structured reporting system for hepatocellular carcinoma (organ procurement and transplant network) introduced a high level of specificity that would have influenced patient management in 17% of 129 patients who underwent liver transplantation for hepatocellular carcinoma [21].

In the field of cardiac imaging, standardized structured reporting systems have been developed and elaborated over the past four decades. Internationally accepted guidelines now exist that recommend how to report coronary angiography and cardiac CT [22–24]. This standardization makes it possible to directly compare results between institutions and facilitates scientific evaluation of these imaging methods.

In the field of prostate cancer imaging, one study found that when a detailed standardized reporting scheme was used to identify the locations of suspicious MRI findings in patients with prior negative prostate biopsies and those on active surveillance, systematic transrectal ultrasound–guided biopsy supplemented by MRI-targeted biopsy yielded a high rate of cancer detection (72%) [25]. Our results are in line with these prior results and favor the application of a standardized certainty lexicon in clinical reports of routine prostate MRI.

One limitation of our study is that the radiologists whose reports were included in the study did not always adhere to the proposed terms of the certainty lexicon. In 14.7% of the reports issued during the lexicon period, the radiologists described the tumor's relation to the prostate capsule and did not explicitly state their level of confidence for the presence of ECE. A possible explanation is that this study covered a period relatively early after the implementation of the lexicon, and the implementation of a certainty lexicon is a process that requires time for continuous process improvement and acceptance. Continuous education is essential to increase radiologists’ awareness of the importance of uniform terminology and improve communication with both patients and referring physicians. The term “less likely,” which indicates a diagnostic certainty of about 25%, was never used in the lexicon period of our study. In reexamining this finding, we realized that the use of the term “less likely” is cancer site–specific. Because uniformity of terminology across all cancer sites is important, a cancer site–specific certainty lexicon would defeat its own purpose.

Another limitation of our study is its retrospective design. Although the MRI reports were issued during routine clinical practice and therefore reflect prospective “real-world” conditions, the validation of the certainty lexicon was done retrospectively. Thus, the findings—especially those for the prelexicon period when no predefined terminology existed—may have been affected by interpretation bias on the part of the reader who retrospectively reviewed and interpreted the original reports.

A possible limitation may also be the 180-day interval between MRI and prostatectomy. However, prostate cancer is considered a slow growing tumor, and such a long interval is frequently reported and accepted in the literature on preprostatectomy MRI [26–28]. Another limitation of this study is that we did not account for the possible impact of changes in technology over time. Furthermore, the effect of a learning curve was not assessed. Although unlikely, it is possible that the radiologists could have improved their diagnostic skills over time and therefore used fewer terms even if the lexicon had not been implemented. In addition, a limitation of the certainty lexicon is that it comes with predefined thresholds. Thus, a radiologist who is, for example, 60% sure is forced to use “possibly (50%)” or “suspicious/probably (75%)” instead of expressing the exact grade of certainty.

In conclusion, a 5-point standardized lexicon of diagnostic certainty has reasonable accuracy for the diagnosis of ECE of prostate cancer on standard-of-care prostate MRI, and its implementation substantially reduces the number of expressions used by radiologists to indicate their levels of diagnostic certainty. Therefore, the use of such a certainty lexicon can provide clinicians with reliable information for clinical decision making and reduce the chances of miscommunication of imaging findings.

We thank Ada Muellner for editing the manuscript.