ABSTRACT
The process of examining human performance in the interpretation of medical images has a very long history and researchers some 60 years ago found that there was a significant amount of inter-observer as well as intra-observer disagreement when examining X-ray images (Yerushalmy, Harkness, Cape & Kennedy, 1950; Garland, 1949). It would be envisaged that one would find a decreasing trend of errors reported in the literature emanating from the following half century of different types of experimental investigation into radiologists’performance in examining a large range of different types of medical image and the numerous publications. Unfortunately, however, these all find that no matter how expert the observer – s/he still makes errors (Scott, Gale & Hill, 2008). This unfortunately gives rise to litigious actions from patients which provides a steady source of income to some lawyers (Berlin, 1996). Making errors, by implication, means that errorless performance can be achieved
and much of the published research has gone into trying to understand why errors occur and then devise ways in which these could at least be minimised. Until fairly recently all medical images were composed of X-ray films which were examined on some form of illuminator in a darkened room. Chest X-rays used to be the most common image class studied. Moveable blinds on these viewing boxes and multiviewers (which allowed for many images to be examined simultaneously) permitted the removal of extraneous light emanating from the viewing box which otherwise caused veiling glare. Many investigations studied how chest X-ray images were
examined, chiefly because of their inherent size these were difficult to visually inspect. Eye movement studies demonstrated repeatedly that such images were examined very quickly (within 10-20 seconds) and that the observer concentrated only upon certain areas of the image. Radiological experience and expertise caused such eye fixations to be located at high probability areas of the image for finding an abnormality. Instructions or information given to the observer prior to examining the image (e.g. ‘this person has had a road traffic accident’) resulted in the observer having an increased likelihood of detecting some abnormality related to the instructions (e.g. a broken rib) but decreased the likelihood of detecting other abnormalities which were visibly present. An oft reported finding which was particularly troubling was that fairly obvious (conspicuous) abnormalities could be easily missed. Satisfaction of search, where once some abnormal features has been detected then the image inspection is terminated so leading to other abnormal features being missed. A theoretical model, based on experimental and cognitive psychology research
findings, was developed (Kundel, Nodine & Carmody, 1978; Gale, 1993) that argued that medical image inspection comprised three key processes: visual search, detection and interpretation. Simply put – saccadic eye movements serve to shift visual attention over the image causing some image areas to be seen in foveal vision where these areas are processed in detail. Ideally this process gives rise to such image features being detected which areNow that the cognitively amassed together and recognised as some image related attribute which may indicate an abnormality. Errors can occur in one ormore of these processes at the same time but importantly it is possible to tease out for an individual in a particular situation just what is causing their errors (e.g. faulting scanning, poor knowledge of abnormality appearance). This then means that training can be developed which instead of simply being generic in nature and being offered to all observers can be very targeted to the specific needs of each individual. The model has stood the test of time chiefly because it is both detailed enough to be practically useful and general enough to be applicable.
