Semiotica 154 (2005), 1-10.
After defining the concept of error as it applies to Roentgen semiotics, we present a typology of the sources of semiotic error in clinical practice. It is a typology based on clinical observation, with the Peircean Categories and the triadic structure of Roentgen signs as organizing principles. This is followed by a review of a general psychological typology of the sources of human error (after James Reason). We conclude with the demonstration of a Category preserving correspondence between psychologic and semiotic sources of error. Thus, independently derived and empirically based typologies of psychologic and semiotic error are shown to have the same categorical structure ie. the same basis in thought.
In this study, error is understood to be the failure of an activity to achieve its intended outcome (cf. Reason 1990:9). We shall examine the types of semiotic error that occur in the reading of Roentgen images (cf. Cantor 2000, 2002a, 2000b, 2003 on Roentgen semiotics). A clinical typology of error in diagnostic radiology was first attempted by Smith (1965, 1967).
Roentgen sign reading is an activity that involves direction of attention, perception, and memory. The intended outcome of this activity is a valid interpretation. To be valid, an interpretation must be based upon some prior knowledge of the physical event that determines the perception involved in the reading process (cf. Peirce 1909:494 on 'collateral observation'). Such prior knowledge may be acquired by personal experience or learned from the experience of others. It is stored in long-term memory and retrieved in the process of image reading. In this way, the triadic relation between a perception (the representamen), a retrieved memory (the object-memory) and a mediation between them (the interpretant) constitutes a Peircean sign. This type of sign will be called a mental Roentgen sign, since the mind of the image reader is a common ground for its correlates. In contrast, a Peircean sign in which the representamen is a percept in a Roentgen image and the object is an anatomic event in a patient may be termed a material Roentgen sign, since the material world is the ground for two of its relates (cf. Cantor 2002a, 2002b). The purpose of Roentgen diagnosis is the detection, localization and identification of abnormal events in a patient through the interpretation of Roentgen signs (Cantor 2002a). In this process, only true signs serve the purpose of diagnosis. However, false signs are common by-products of image reading, where a sign is said to be false if its interpretant is false. In this study, the term semiotic error will denote the unintended creation of false mental signs in the process of image reading. Clearly, each correlate of a mental sign may be true or false since it is grounded in the mind of an image reader. A representamen of a mental Roentgen sign is false if a false perception is present or a true perception is not present. An object-memory is false if there is a false memory of true prior knowledge or a true memory of false prior knowledge. An interpretant is false if it mediates between a false representamen and a true object-memory, a true representamen and a false object-memory or a false representamen and a false object-memory. In what follows, we present a typology of false mental signs produced in the reading of Roentgen images that is based on Peircean category theory.
The truth-value of a triadic sign is the truth-value of its interpretant (cf. Cantor 2004 on truth values). The truth value of an interpretant may be either provisional or conventional. Provisional truth-values depend on the state of mind of the image reader, since perception, memory and interpretation are all subject to human error. Conventional truth-values are independent of the state of mind of any single image reader and are based on consensus in a global community of image readers. Provisional truth-values assigned by a conventionally trained reader are usually consistent with conventional truth-values of the global community of readers. The concept of semiotic error involves a disparity between provisional and conventional truth values. Hence, a sign that is provisionally true for one reader may be conventionally false in the global community and a sign that is provisionally false may be conventionally true. We shall say that a mental Roentgen sign is false if its interpretant is conventionally false. In the sections that follow, we shall examine the sources of false perceptions and false object-memories that produce the false interpretants of false Roentgen signs.
False representamina are due to perceptual error. We say that a representamen is false if it does not correspond to an event in the material world. Perceptual errors may involve the presence of a false perception (a false positive finding) or the absence of a true perception (a false negative finding). In other words, a representamen is false if a finding perceived in an image does not correspond to an abnormality in the patient (termed an 'artifact') or if an abnormality in the patient is not perceived in an image (termed a 'miss'). A false perception may originate in the visual system or in the mind of the image reader. Such representamina are optical illusions. Optical illusions encountered in the reading of Roentgen images include Mach bands and subjective contours (cf. Jaffe 1984). Mach bands appear as dark lines that conform to high contrast edges e.g. edges formed by the superposition of bones or at the cardiopulmonary interface. Subjective contours are produced in the mind of the image reader by a projectional confluence of opacities e.g. in an image of the lungs. The reader's best defense against optical illusions is an awareness of the conditions that produce them. The absence of a true perception may be due to a failure of visual detection or a failure of visual search.
Failure of visual detection
Failure to perceive an abnormal contrast separation effect at an interface will result in failure to detect the presence of an abnormal interface or the absence of a normal interface (cf. Cantor 2000 on the 'separation' signs). Clearly, conditions that produce insufficient normal contrast separation will also obscure an abnormal contrast separation. Insufficient contrast separation at an interface may result from underexposure or overexposure of a Roentgen image. Techniques are readily available that, within limits, counteract the effects of extremes of exposure. In the case of overexposure, contrast separation may be restored by transillumination of the image with a bright light. In the case of underexposure, contrast separation may be restored when the image is viewed with peripheral masking. Both techniques are most effective in conjunction with low intensity ambient lighting. Failure to perceive a line or an edge in a Roentgen image may also be an effect of visual noise or superposition. These effects may be avoided by repeat imaging with elimination of the source of visual noise, repeat imaging with a different projection or cross-sectional imaging.
Failure of visual search
The reading of a Roentgen image begins with a visual search. The term visual search denotes a visual sampling process that may be consciously or unconsciously directed (cf. Tuddenham 1962, Kundel 2000). The purpose of visual search in diagnostic radiology is the detection of any abnormality that is represented in a Roentgen image. A visual search that does not fulfill its purpose will be called ineffective. Visual search may be ineffective for several reasons. These are referred to as limited search, biased search, interrupted search, accelerated search and satisfaction of search.
Limited search. Clearly, any visual search should be exhaustive. Otherwise, an abnormality in a part of an image that is not sampled would be 'missed'. Hence, one reason for an ineffective visual search is limitation of scope. A common example of limitation of scope is the restriction of search to the central region of an image with exclusion of the periphery. Implicit in any visual search is a choice of viewing scale. An image may be minified or magnified (i.e. viewed at different scales) by adjustment of the viewing distance. Minification facilitates the detection of abnormal asymmetries and magnification increases the conspicuity of small-scale abnormalities. Without an appropriate choice of viewing scale, such abnormalities are easily missed. Hence, an exhaustive visual search must include viewing of the image at different scales. Omission of an appropriate choice of viewing scale will produce an ineffective search due to limitation of scale.
Biased search. A visual search is biased if its scope is limited by prior knowledge. Specifically, an undisciplined image reader may anticipate abnormalities in one anatomic region and limit the scope of visual search accordingly. Hence, when presented with a chest film the reader may search the heart and lungs and ignore the surrounding soft tissues and bones. With an abdomen film, the reader may search the intestinal gas pattern and soft tissues and skip the bones. With images of large joints such as the shoulder, hip, knee etc. the reader may search the articular structures and pay insufficient attention to nonarticular structures. Hence, an ineffective visual search may be due to a regional bias. In the reading of Roentgen images, collateral knowledge in the form of clinical history is often essential for a valid interpretation of findings that are morphologically nonspecific. However, an undisciplined reader may anticipate abnormalities on the basis of clinical history and limit the scope of visual search accordingly. Since clinical history is frequently misleading, visual search may be ineffective due to an anamnestic bias. In order to avoid a biased search, an image reader must be aware of the sources of bias.
Interrupted search. In principle, the scanpath employed in visual search should continous as well as exhaustive. However, personal preoccupations or environmental distractions may interrupt the search process. Interruptions that produce gaps in the scanpath produce a discontinuous search pattern. Interruptions followed by premature termination of search result in a truncated search pattern. Hence, an ineffective visual search may be due to interruption of search. To avoid ineffective visual search, any interruption must be followed by a complete repetition of the original search.
Accelerated search. An ineffective visual search may also result from acceleration of search. An accelerated visual search may produce skips in the scanpath. This may occur when real or perceived time-constraints are imposed on the image reader. A real time-constraint may be felt when, by interruption, the reader is made aware of another task demanding prompt attention. A perceived time-constraint may be felt when an impatient clinician, by interruption, requests a 'quick' reading. Under these conditions, in order to avoid an ineffective visual search, the reader must make a conscious effort to resist the temptation to accelerate the reading process.
Satisfaction of search. In satisfaction of search, the detection of one or more abnormalities is followed by a premature termination of search. It is as if the image reader's need to perform a visual search is satisfied by the successful detection of even one abnormality (cf. Tuddenham 1962). In order to avoid this type of error, the image reader must be aware of the risk of satisfaction in detection and make a conscious effort to complete any visual search, regardless of its yield.
In practice, an ineffective visual search may result from more than one of the above conditions.
A false object-memory may be a false memory of true prior knowledge or a true memory of false prior knowledge. A false memory of true prior knowledge may be due to incomplete retrieval or embellishment of true prior knowledge. False prior knowledge may be knowledge that was false when acquired, falsely inferred or simply imagined. The best defense against false object-memories is continuing medical education. Furthermore, memory may be refreshed during image reading by the use of reference books or consultation with colleagues. It is possible that personal interpretations, accumulating in the mind of the reader over years of practice, may give rise to habitual errors that are difficult to eliminate. The first line of defense against such systematic errors is to acknowledge that they may exist.
A false interpretant may mediate between a true representamen and a false object-memory, a false representamen and a true object memory or a false representamen and a false object-memory. False interpretants may result from denial or predilection. The risk for interpretant error by denial increases when there is a personal relationship between the image reader and the patient. The risk for interpretant error by predilection decreases if the interpretant is thought of as a graded list of statements (i.e. a differential diagnosis) rather than a single statement.
In this study, semiotic error is understood to be the production of false mental signs in the process of Roentgen sign reading. Hence, a categorical typology of false mental signs will serve as the foundation for a categorical typology of semiotic error. In a Peircean sign, the representamen is a representation of the object. This relation of representation implies that the representamen 'stands for' or may 'take the place of' the object in thought. For this to be possible, the representamen must share some attribute with its object. This condition of commonality is the basis for our categorical typology of false mental signs. After Peirce, there are three categories of commonality. First Category commonality is similarity of appearance (resemblance). Second Category commonality is contiguity in time and place (collocation). Third Category commonality is conventionality, interpreted as agreement in a community of minds. In the Peircean typology of signs, a First Category sign is termed an icon, a Second Category sign is termed an index and a Third Category sign is termed a symbol. Hence, false mental signs may be false icons, false indexes or false symbols. This categorical typology of false mental signs determines a categorical typology of semiotic error.
False icons
In a false icon, there is a false resemblance between representamen and object-memory. This may occur when the representamen is true and the object-memory false or when the representamen is false and the object-memory true. There may be false positive or false negative icons.
False positive icons. A false positive icon is an icon that is falsely interpreted as a presence. When the representamen is true and the object-memory is false, a false positive icon is a simulator (cf. Cantor 2003 for a discussion of fracture simulators). When the representamen is false and the object-memory is true, a false positive icon is an optical illusion.
False negative icons. A false negative icon is an icon that is falsely interpreted as an absence. When the representamen is true and the object-memory is false, a false negative icon is due to a failure of memory. When the representamen is false and the object-memory is true, a false negative icon is due to a failure of detection or ineffective visual search.
False indexes
In a false index, there is a false collocation of representamen and object-memory. When the representamen is true and the object-memory is false, false collocation is a result of Roentgen projection or superposition. When the representamen is false and the object-memory is true, false collocation is due to ineffective exposure of the image producing false separation signs at boundaries (cf. Cantor 2000 on 'separation' signs).
False symbols
In a false symbol, there is a false conventional relation between representamen and object-memory. In Roentgen diagnosis, the most common conventional relations are frequency of association and causation. Such relations are false if either the representamen or the object-memory is false.
This completes the categorical typology of false mental signs. In the next section, we shall demonstrate a correspondence between a general psychological typology of human error and our categorical typology of semiotic error.
In general, the concept of error entails intentional activity i.e. human activity that is directed by the mind towards an end (Reason 1990:5). By definition, human error is the failure of intentional activity to achieve its intended outcome, independent of chance (Reason 1990: 9,17). In a general classification of human error which he called a Generic Error-Modeling System (GEMS), James Reason distinguishes between error types and error forms. Error types are based on performance levels (after Rasmussen) and error forms are based on relation levels (Reason 1990; 61-63).
Error types
There are three fundamental error types. These are termed 'skill-based', 'knowledge-based' and 'rule-based' errors. Skill-based errors (termed 'slips') are due to failure of an intentional sensorimotor activity. These are execution errors (Reason 1990: 53, 56). Knowledge-based errors (termed 'lapses') are due to failure of knowledge storage or retrieval. These are errors of memory (Reason 1990: 43, 86). Rule-based errors (termed 'mistakes') are due to failure of a plan to achieve its intended outcome. These are problem-solving errors (Reason 1990: 8, 63).
We may now apply this classification by performance level to the specific intentional activity of Roentgen sign reading. The intention of a Roentgen sign reader is the detection and recognition of all findings in an image that represent true abnormal events in a patient. A false interpretation is a discrepancy between the intention of the reader and the outcome of the reading. Hence, for the activity of Roentgen sign reading, skill-based error is due to a failure of perception, knowledge based error is due to a failure of memory and rule-based error is due to a failure of interpretation. In semiotic terminology, a skill-based error is a false representamen, a knowledge-based error is a false object-memory and a rule-based error is a false interpretant.
Error forms
Problem solving is a mental activity that combines knowledge retrieval and conscious inference (Reason 1990: 113, 117). Knowledge retrieval is an automatic activity that involves the transfer of knowledge from memory to consciousness in response to retrieval cues (Reason 1990: 97). There are three forms of relations between retrieval cues and retrieved memories that result in knowledge retrieval. These forms may be thought of as activators of knowledge retrieval that are independent of intention. They are characterized by shared attributes, shared context and frequency of encounter (Reason 1990: 100-101). In Reason's terminology, knowledge retrieval based on shared attributes is called 'similarity-matching' and knowledge retrieval based on frequency of encounter is called 'frequency-gambling'. Hence, problem-solving errors may be due to false 'similarity-matching' or false 'frequency-gambling'. We may now apply this classification by relation form to Roentgen sign reading. Clearly, false 'similarity-matching' corresponds to a false resemblance between representamen and object-memory. Since frequency of encounter is a custom or convention of a community (Reason 1990; 98, 104), false 'frequency-gambling' is a false convention. Up to this point, the correspondence between psychological and semiotic models is incomplete since there are three types of false mediation in the semiotic model and only two forms in the psychological model. However, shared context is a third activator of knowledge retrieval that was mentioned by (Reason (1990: 100-101) but not used as a basis for a named retrieval relation. For convenience, a retrieval relation based on shared context will be termed 'context-localizing'. If we interpret a shared context in Roentgen sign reading to mean spatiotemporal proximity, then false 'context-localizing' corresponds to false collocation. Figure 1 represents a category-preserving correspondence between a general psychological model of intentional activity and our semiotic model of Roentgen sign reading. For the specific activity of Roentgen sign reading, the semiotic model may be thought of as a representation of the psychological model in terms of universal categories of thought i.e. the Peircean categories. Placing the word 'false' in front of each term in the diagram yields a category preserving correspondence between a psychological typology of the sources of human error and our categorical typology of the sources of semiotic error.
Based on the concept of a mental Peircean sign, we have constructed a categorical typology of the semiotic errors that occur in the reading of Roentgen images. We have seen that a psychological typology of the sources of human error is homomorphic to our categorical typology of the sources of semiotic error. This finding is yet another manifestation of the universality of the Peircean categories.