Despite the affinity of radiologists and radiology research experts towards image-based data, natural language is ubiquitous in the workflow of medical imaging. The diagnostic report is a written document consisting of the radiologist's interpretation of all findings detected on an imaging study. Its primary function is to communicate findings and recommendations to the referring provider and – increasingly – the patient Therefore, many radiologists exercise high levels of discretion in their word choices to ensure that the medical knowledge contained in the image interpretation is translated precisely onto the text. Nevertheless, most diagnostic reports build their foundation on freely dictated text in complete sentences, using the radiologist's own word choices. These sentences are examples of natural language, and natural language sometimes gives rise to ambiguity. For instance, a finding may be described as an abnormality, a lesion, a density, or a focus.
Additionally, the natural language also expresses an imperfect representation of otherwise quantitative assessments such as a radiologist's diagnostic certainty. In one study, researchers found that “wide variability [exists] among radiologist’ preference for phrases used to convey diagnostic certainty”; for instance, readers have rated the phrase “suggestive of” as between “low likelihood” to “very high likelihood” Despite the wide variability, word choices matter in clinical care.