Effects of Prevalence Expectation on Visual Search Behaviour

Effects of Prevalence Expectation on Visual Search Behaviour Page | 1 Introduction The objective of this research is to investigate the effects of prevalence expectation on visual search behaviour and cognitive function in radio-diagnosis of the adult chest radiograph and its impact upon medical expert witness testimony in malpractice litigation. Research into radiological error has consistently demonstrated disparity in radiological performance. Most studies of radiology error report significant rates of intra and inter observer variability. The pertinent question is why do radiologists make mistakes? â€Å"Conventionally, radiologists produce diagnoses on the basis of a combination of their training, experience, and individual judgment. Radiologists perceive and recognise image patterns and associate or infer a diagnosis consistent with those patterns.† [1] Accurate results depend upon the radiologist’s ability to recognise a lesion, ignore irrelevant details, and retrieve pertinent memories in order to accurately interpret an image. The radiologists primary task is to accurately identify a range of anatomical structures and pathological findings on medical images. The diagnostic process in radiology is not well understood. Basically, radiologists perceive and recognise image patterns and then associate or infer a diagnosis consistent with such patterns. Appearances on a radiograph are silhouettes of normal and abnormal anatomy. Each shadow represents a projection of layers of detail on a two dimensional surface from a three-dimensional object. The successful detect ion of the target and the time required to seek it out depends upon a number of external factors Much information needs to be processed by the radiologist during the interpretation of such shadows. The disease pattern, The clinical and demographic information relating to the patient The differential diagnoses pertaining to such a pattern. The experience of the radiologist A general aspect of everyday life is looking for a particular target amongst an assortment of other (distracting) items. Visual Search is one task that is performed routinely from radiological diagnosis to finding a definitive research text in a library. Visual search is an essential element in the cognitive process and is the interaction between the visual system, its target and the reader’s subsequent decision. Each experiment was conducted in a controlled laboratory environment, but were designed to simulate, as near as possible, the clinical arena. Expectation is explored by manipulating the information given to radiologists between general clinical information and very specific clinical information, thereby encouraging each radiologist to formulate an individual expectation of abnormal prevalence in the images presented to them. We then measured any subsequent change in behaviour as the subjects’ biases shifted. It was hypothesised thata radiologist would vary their decisions at the cost of more or less false alarms. This is termed criterion shift. The amount by which a radiologist considers false alarms to achieve a higher rate of detection is termed the bias. Bias represents an approach to lessen the consequences of a missed target. What is visual search? In a visual search task, subjects look for a target item among a number of distracting items. [1] After breaking down an image into its distinct components, the visual system directs attention to unusual areas for further analysis. This act of looking for and selecting an anomalous feature on a radiographic image is termed visual search, and is the task performed by radiologists daily. At a very basic level, there are generally considered two types of visual search termed pre-attentive and attentive. A great deal of research has been conducted regarding the differences between these two types of visual search and the factors that affect performance during each [2, 3]. Pre-attentive search has been given a number of labels including efficient search, parallel search, easy or effortless search [1] and automatic detection [4]. In this type of search, the targets are anticipated to contain features which are processed pre-attentively [5], basically drawing attention to themselves. In this type of search, the target appears immediately to the observer, requiring little search effort. For example when searching for the letter X on a page of letter Os or for a red target in a mass of blue targets. Attentive search also referred to as inefficient search [1] serial search [4] and controlled search [2] refers to a complicated search in which attention must be given to targets much less obvious. These components are influenced respectively by the characteristics of the environment (e.g., saliency of targets and distracters) and learned search strategies. The distinction between these two types of search is consistent with the view that search is driven by both bottom up and top down processes [6] wherein bottom up processes drive attention due to salient features in an environment or target features and top down processes drive attention through the function of search strategies such as the direction of attention to locations of high priority. Experienced radiologists develop a mental global impression of a standard radiograph [7, 8] sometimes called a target template [9] and in some instances use pre-attentive search as the abnormality is almost instantaneously visible. Conversel y they may use the second approach in more difficult cases. Recognising how these factors influence target detection helps to understand real-world search tasks and cognitive psychology. Together, the results can help to show how different influences affect visual search in the performance of real-life search tasks as in medical radiology, and airport baggage security. Fiore et al [10]) define threat detection in an airport screening task as the ability to rapidly recognise targets in the environment and interpret the meaning and importance of these cues. Nodine et al [11] break the radiological interpretation task down into three elements, describing the task as consisting of a search for, the recognition of an abnormality and the decision made regarding the abnormality. Many occupations depend on the speedy and effective execution of a visual search. Surf life savers in Australia are trained to search the ocean for rips, sharks and for swimmers in difficulty. Diamond cutters need to be able to evaluate the cutting potential of a rough diamond and need to be able to look at a diamond crystal and determine whether it will yield a large enough gemstone to make a profit.The quality and price will vary greatly based on the cut quality [12]. Airport baggage security screeners are trained using Threat Image Projection (TIP) technology to detect potential threat items within an X-ray image that may contain harmless clutter. There are various features which might hamper a successful visual search, including item superimposition, different viewpoints, rotation of the item and general image complexity. A number of other demands may also influence search, including psychosocial pressure (a busy airport might provoke a too rapid search), expectation (a threat it em is a rare occurrence), boredom and the number of different targets added to the list of dangerous or suspicious items. Errors in these areas are potentially life-threatening, expensive or both. In the medical arena, diagnosis in cytology and radiology are two areas that are dependent upon visual search. Clinical understanding and proficiency in visual pattern recognition serve as the basis for diagnosis by radiologists and pathologists [13]. Error in Radiology Imaging departments must offer a top quality radiological service with as little risk to the patient as is possible. It is therefore incumbent upon radiology department to address any problematic areas and try to reduce the causes of error. The accuracy of the radiological report is one element of the patient’s care. In radio-diagnosis, errors are either of perception or cognition [14]. Perceptual errors occur when targets (tumours, infection) are not visualised. A cognitive error occurs when an abnormality is seen but the radiologist draws the wrong conclusions due to unsound diagnostic reasoning. Diagnostic errors are classified as either false positive (FP) or false negative (FN). A false positive decision is due to the interpretation of a presumed target as pathological when it is in fact normal. A false-negative is an abnormality that is present but is not perceived. FN errors errors are five times more likely than FP errors [15], whilst perceptual errors are four times more frequent than decision making errors [14]. Radio-diagnosis not an exact science. Approximately 4% of Radiological interpretations contain errors, fortunately, most of these errors are clinically insignificant, or if serious errors are found, they are promptly corrected causing no harm to patients [16]. Identifying and locating items can be challenging especially when one is uncertain of where, or even what, to look for. Difficulties are more likely to occur when observers must locate the target’s position rather than simply detect the target’s presence [17]. Numerous studies have been conducted to understand error in radio-diagnosis. This issue has been recognised for a number of years. In the 1940s, Garland [18] found that 10-20% of chest radiographs of patients with suspected tuberculosis (TB) were read differently by different observers and a study by Quekel et al [19] observed that 19% of lung cancers presenting as a nodule on chest radiographs were missed. Another study [20] identified major diagnostic variation between three experienced radiologists interpreting medical images of patients in an emergency department. In this study the level of diagnostic agreement between the radiologists varied according to the anatomical area examined. Levels of agreement were abdomen (51%), chest (61%) and musculoskeletal (74%). There are a number of influences that affect the perception and diagnosis of chest lesions. Of these influences, prevalence expectation has had very little study. Target prevalence and prevalence expectation Prevalence expectation or bias occurs when expectations about an outcome influences a subject’s behaviour. In radiology this can be a factor during diagnosis. For statistical reasons, laboratory studies of visual search in typically include targets on 50% of all trials and participants often recognise that any given trial has an equal chance of having a target or not. However, real-life searches are seldom so balanced. In airport security, baggage screeners view numerous x-ray images of suitcases, but the incidence of a dangerous item happens infrequently. Consequently, one issue in visual search is exploring how a difference in target prevalence modifies searcher expectation and subsequent detection performance. This issue has been raised in multiple domains, including vigilance studies, radio-diagnostic perception, and cognitive psychology. In routine radiological examinations, the occurrence of abnormalities is generally low. This frequency fluctuates depending on the demographics of the population and the anatomical area being examined. However, in all cases, a mis-diagnosis might result in serious consequences. Research within radiology has therefore attempted to show whether low target prevalence is responsible for diagnostic error. Whilst a number of studies, within the medical and non-medical domains, have investigated whether the number of targets present (prevalence) can affect performance. [21, 22, 23, 24], there has been almost no research undertaken regarding the effect of prevalence expectation.