In 2013, three investigators at the Visual Attention Lab in Harvard published the results of an experiment that they entitled, catchingly, ‘The Invisible Gorilla Strikes Again’. In this study, 24 radiologists were tasked with the detection of pulmonary nodules in a set of five chest CT scans, a routine component of a chest radiologist’s working life. In the final CT scan, the researchers had inserted the image of a 29 x 50 mm gorilla into the lung window images. The size of this gorilla was approximately 48 times the size of the average lung nodule, also present on the images. After the radiologists had finished their task, they were asked specifically if they had noticed the gorilla on the final CT. Of the 24 radiologists, 20 had failed to notice it. Of those 20 radiologists, eye tracking software showed that 12 of them had looked directly at the gorilla.

This startling finding was an example of inattention bias, one of the many biases that can impair a radiologist’s ability to detect significant abnormalities in imaging. It reflects the vulnerability of humans to a range of psychological pitfalls that can impair image interpretation.

Clinical Vignette

Categorisation of radiology errors

Radiology errors are broadly categorised into perceptual errors and cognitive errors.

Perceptual errors – where a lesion is missed – are far more common, and potential causes include: lesions being poorly conspicuous (either inherent to the lesion or due to suboptimal technical factors); overlapping structures and blind spots; incomplete visual search (due to inexperience, overly rapid interpretation speed, or satisfaction of search bias); and inattention bias (where an abnormality is overlooked because attention is being focused elsewhere, as was the case in the vignette presented).

Cognitive errors occur when an abnormality is detected but not interpreted correctly. This can occur in trainees simply due to inexperience but can also occur in expert radiologists due to a range of psychological factors. One example might be misinterpreting a necrotic mass as an abscess based on prior reports (an alliterative cognitive error) or because of undue influence by the way a case is presented by the clinician (framing). Another example of a cognitive error is over-calling benign lesions (e.g. misinterpreting benign subpleural lymph nodes as possible pulmonary metastases, or mistaking IV contrast mixing in the pulmonary arteries as a pulmonary embolism). This can lead to unnecessary further imaging or incorrect patient management.

A 2023 review by the State Claims Agency of radiology claims over a five-year period, showed that missed diagnosis accounted for 66% of claims against radiology services. Missed cancers were the second most common cause of missed diagnosis (after missed fractures), with 50% of those due to missed lung cancer.

Unfortunately for radiologists, the evidence of a small, missed lung cancer is always easier to see in hindsight, preserved as it will be on a chest x-ray or CT scan, often with annotated arrows pointing at it and with accompanying magnified snapshots.

Managing the risk

Even with our awareness of the wide range of perceptual and cognitive errors that can occur in routine daily practice, it is difficult for the radiologist concerned to understand why a particular error occurred. However, awareness of such interpretation errors can help mitigate their occurrence.

Perceptual errors can be reduced by attention to technical and ambient factors that might affect the conspicuousness of a lesion (image processing and display parameters, room lighting, frequent interruptions); workload and speed of reading; insistence on accurate and full clinical details in referral letters; and awareness of blind spots.

In chest CT, common blind spots include the perihilar and paramediastinal regions, particularly for the detection of small pulmonary nodules that might be mistaken for blood vessels; the airways (trachea and bronchi); the pleura (chest wall, mediastinal, diaphragmatic and fissural); the heart (both vascular abnormalities and intracardiac masses and thrombi); the aorta and pulmonary arteries (e.g. unexpected pulmonary emboli can occur in up to 4% of oncology outpatients); and extrathoracic locations (breast nodules, supraclavicular and axillary lymph nodes, bone lesions, abnormalities in the upper abdomen).

Special attention should be paid to the bones to rule out unexpected metastases and rib and vertebral body fractures. A specific ‘bone search’ that includes evaluation of the ribs and multiplanar reformats of the spine is advisable in every chest CT read.

Unfortunately, radiologist error is unavoidable, with an estimated 4-15% of diagnoses being missed, incorrect or delayed.[1] In addition to the potentially significant harm and suffering for patients, the consequences for radiologists include not just the risk of clinical negligence claims, but adverse emotional and psychological effects.

Artificial Intelligence (AI) applications are increasingly being used in radiology and might reduce both the human element and the error element of radiologist interpretation. In the meantime, it is important for radiologists to be aware of the range of factors that can lead to error and to do their best to minimise them.

References

1. Commonly Missed Findings on Chest Radiographs: Causes and Consequences. Gefter WB, Post BA, Hatabu H. Chest. 2023 Mar;163(3):650-661