As with abdominopelvic and chest imaging, vascular discrepancies formed a significant contribution to total errors (Figure 2).  This is not surprising given that most CT or MRI exams are not optimised to detect vascular anomalies.  However, carotid arterial dissections and large aneurysms can be seen on both CT and MRI without contrast, as can venous sinus thrombosis.[4], [5] One possible source of underlying error may be the “edge of film” phenomenon, with superior sagittal sinus thrombosis frequently only seen on the top slices of the axial images, and internal carotid or vertebral dissection only being visible in the bottom few slices.  Another likely reason for the number of vascular discrepancies is that the vascular tree is often only scrutinised when a specific diagnosis is queried. This is supported by a study showing that detection of ICA dissection improved from 23% to 77% when arterial review became incorporated in routine review on standard non-angiographic MRI sequences, even in inexperienced viewers.[6] 

Unsurprisingly, peripheral grey matter lesions accounted for a high number of discrepancies given the complex and convoluted course of the grey matter (Figure 3). One study showed an increase in sensitivity from 57% to 71% for the detection of stroke on CT using a level centred at 32 Hounsfield units (HU) with a width of 8 HU[7]. Other authors have also suggested the benefit of reviewing CT on a “stroke window” of 40 HU as the level centre with a width of 40 HU for a multitude of pathologies affecting both grey and white matter[8]. (Figure 4)  On a similar theme, bone review also benefits from appropriate windowing and in the context of trauma, separate bone reconstructions using a high spatial frequency reconstruction algorithm are useful for subtle fracture detection (Figure 5)[9]. 

Misclassifications in the frontotemporal parenchyma surrounding the Sylvian fissure were noted by the authors to be so common that we felt this warranted separation into its own group (Figure 6).  The difficulty of diagnosis in this region cannot be overstated and is largely a result of the complex multiplanar anatomy further complicated by the number of pathologies that frequently occur here in their earliest form, such as the subtle insular ribbon sign, early oedema or the loss of the Sylvian fissure indicating subarachnoid haemorrhage.  The use of MPR would render the “edge of film” misses null and void as the edge of a series on one plane often becomes the centre of the series on another plane.  Similar benefits should be seen in the parafalcine region, the final region of common observational error (Figure 7).  This results from the close approximation of cerebral hemisphere, falx cerebri, corpus callosum and perifalcine vessels.  From our experience, the discrepancies were more easily appreciable on coronal or sagittal reformats than on the original axial images.

Conclusion

Radiological errors are common and by recognising the potential source of errors, strategies can be formulated to try and minimise them.  With over 50% of all perceptual errors occurring in just five anatomical sites- vasculature; peripheral cerebral grey matter ; bone; parafalcine; and the frontotemporal lobes surrounding the Sylvian fissure, there is an avenue for focused review at the end of reporting to improve the accuracy of radiological reports.