1 |
Introduction |
1.1
Leg Ulcers: Causes and ConcernsThe skin is a complex organ of the human body and its property of self-regeneration and its contribution to thermoregulation (temperature control of the body) are often taken for granted. The skin is a first line of defence against infection carried by myriad agents with which it is in daily contact. It prevents foreign bodies from penetrating the underlying tissues and organs and protects against damage caused by harsh physical contact. Wounds of the skin are caused by many different factors both internal and external to the human frame. Common injuries caused by external factors include cutting, abrasion, bruising and burning. Injuries that tear or remove portions of the skin are classified as
acute wounds and the natural remedial action of healthy skin is to close the wound. In contrast, the occurrence of a leg ulcer is a symptom of an underlying problem with the skin and its blood supply and drainage systems. Leg ulcers may be triggered by external agents: even something normally considered harmless, such as a sharp blow against an unyielding object, can precipitate the onset of a leg ulcer. They may also develop without external stimuli due to the severity of the underlying condition of the skin. Since the major factor in leg ulcers is a degeneration of the dermis itself, they are classified as chronic wounds and are typified by a much extended healing time when compared to acute wounds.All wounds that heal are healed by the skin's own in-built mechanisms, without which no healing could occur. In the case of
acute wounds, healing can be rapidly aided by primary intention, generally the physical closure of the wound by sutures. Such action relies upon the mechanical strength of the adjacent skin tissue to resist being torn by the suture material. A leg ulcer may be considered to be a symptom of the breakdown of the skin's ability to adequately heal itself. The poor condition of the skin in the vicinity of an ulcer means that closure of the wound by primary intention is not possible and, in any event, this would only be treatment of the symptom rather than treatment of the underlying cause of the ulcer. Therefore wound healing must be aided by the external influence of medical treatment and supervision. This practice is termed healing by secondary intention.Healing of a wound is a relatively slow process despite the attentions of the best understanding, although it should be noted that without appropriate medical treatment a leg ulcer will almost certainly never heal. The prevalence of leg ulcers in the UK alone has been estimated to affect 1-2% of the total population and estimates of the total cost of treating leg ulcers have recently exceeded £300 million
(Davis et al., 1992). A typical treatment regime may be based on a combination of the following interventions:
In time, a healing wound will tend to reduce in size, whereas the size of a deteriorating wound will increase. Wounds may take several months to heal fully, and even then they may recur because of persistence of the underlying conditions in the dermis which first caused the ulcer. Other wounds may exist for more than a year and sometimes never fully heal and it is not unusual for an ulcer to be deemed ‘indolent’, indicating a stagnation of the healing process. Therefore, since changes in wound size are an indicator of the direction of the healing process, measurement of a wound’s physical dimensions is an important tool in the management and monitoring of the healing process. To determine whether a wound is changing in size, the physician must compare recent wound measurements with the current measurement and needs to be satisfied that the measured differences are indeed a manifestation of wound healing (or conversely, deterioration) and not due to experimental error. Clearly, if there is no detectable difference, then either the measurement precision is poor leading to low confidence in measured difference or the wound is not responding to treatment and is thus indolent. The measurement precision therefore influences the likelihood of detecting changes in wound size. Generally, a patient’s wound will be inspected on a weekly or fortnightly basis when the existing dressing will be replaced, affording the opportunity to measure the wound with little interference. Failure to reliably detect a significant size change will lead to a delay of one or two weeks if one considers that the course of action may be altered partly as a consequence of measurement results.
This project arose as an enhancement of the MAVIS instrument
(Plassmann, 1992, Jones and Plassmann, 1995). MAVIS stands for Measurement of Area and Volume InStrument, and, as its name implies, this instrument has the capability of measuring the surface area of a wound and of estimating the wound volume defined as the volume of that space enclosed between the wounded surface and the reconstructed original healthy skin surface over the wound site. Measurement of both surface area and volume requires 3-D information in the form of a depth-map of the wound surface which the instrument obtains by projecting structured light on the wound and subsequently processing the resulting captured image. One step in this image processing sequence is the definition of the boundary of the wound which is required for measurement of the wound’s perimeter, area and volume. Currently, the boundary definition is performed by manually delineating the wound using a standard computer mouse to guide an on-screen cursor.
1.3 Practical Wound Measurement
Several methods for measuring various dimensions of a wound exist, ranging from an estimate of area based upon measurements of linear dimensions made with a modified ruler
(Kundin, 1989) to video-capture and three-dimensional imaging techniques (Plassmann, 1995, Plassmann and Jones, 1998). Imaging methods make use of video capture which involves more expensive equipment but has the advantage of not making contact with the wound, i.e. they are ‘non-invasive’, thus alleviating the possibility of imparting infection or of injuring delicate tissue. Both of these events will prolong occurrence of the wound, perhaps by many weeks, so it is in the best interest of the patient that such events are avoided. This thesis is concerned with the application of computer-based imaging to the wound measurement problem. In this context, the most rudimentary procedure is to perform a manual tracing (delineation) of the boundary of a wound image by using either a track-ball or mouse. Once the boundary has been defined, the area and circumference can be measured in terms of pixel areas and lengths respectively. These measurements may then be converted to physical units by application of suitably derived pixel scaling factors, notwithstanding the many sources of error and distortion introduced by the imaging and video capture processes.Manual delineation is subject to several criticisms that bear upon its expected performance as a measurement process and these are directly linked to the human factor. Due to the nature of hospital work, a successful measurement process needs to be fast in operation, at least in terms of capturing necessary data required for the measurement, and easy to use. Automation of the wound measurement process by replacing the task of manual delineation with a wound measurement algorithm would place less of a demand upon the physician. Removing the delineation process and replacing it with a computer-based algorithm removes the subjective element intrinsic to human judgment. Firstly, it is considered that the opinion as to the extent of a wound may well vary among manual delineators introducing the possibility of measurement bias. Secondly, the process of delineation is subject to the available level of human dexterity, which is likely to be variable, both for the same person under variations of physical and mental state and from person-to-person, thus producing differing levels of variability. The visually ambiguous nature of a leg ulcer is the cause of bias in area measurements and lack of clarity or contrast at the wound boundary adversely affects area measurement precision.
The growth of digital image processing combined with the relative power of the personal computer and the availability of video capture equipment at a reasonable cost has enabled imaging techniques to be extended into many spheres of medical science. It is thus feasible to propose that digital imaging techniques can be used in the process of measurement of wounds. There are many published algorithms that may be applied to the task of segmenting an image and thus attempting to measure the area of a wound. However, it is considered that measurement of wounds by manual delineation is not a trivial task, this being witnessed by the variability of measurements produced by this method. Simple imaging tasks can be achieved by digital imaging methods without the aid of higher-level intervention. However, it is reasonable to consider that an imaging task that is interpreted ambiguously by the human visual system cannot currently be replaced by an autonomous system that is constrained to rely upon the same visual data. Thus an imaging task that is guided by human judgment but which can improve measurement precision is a feasible proposal. To overcome the limitations of the manual delineation process and thus improve upon the precision of wound area measurements, and indirectly, volume measurements, it is proposed that active contour models or ‘snakes’
(Kass et al., 1987) be used to refine the manual measurement process. Much work involving the development, modification and application of active contour models to imaging specific problems has since ensued. Snakes solve the problem of continuous edge finding in noisy and variable contrast images and, given an initialisation provided by a higher-level process, allow the introduction of a form of globality to the solved edge detection problem. Taking these factors into account, the principal objectives of this project can be stated:(a) Quantification of variability: is the area measurement precision improved, or under what circumstances is the precision improved?
(b) Agreement between manual delineator performance and algorithm performance.
This thesis describes the development of four ‘flavours’ of active contour model algorithms adapting and applying the developments reported in the widespread literature available. The properties of wound images are presented and discussed in Chapter 2 with regard to their bearing upon the application of many image segmentation and edge-detection algorithms reported in the literature, including algorithms applied to other medical imaging problems. It is shown that these algorithms are not easily applied to the wound area measurement problem and that the widely varied properties of different wound images is a great obstacle to the application of many well understood and successfully applied algorithms to this task. Active contour models are chosen as the most applicable solution for several reasons:
This thesis claims the following as its original contribution to knowledge:
1.6 Definition of Measurement Terms
This section is included to avoid any confusion or ambiguity involving the interpretation of measurement results reported in this thesis. Errors of measurement may be classified as either random or systematic. All experimental error in this work is expressed in terms of precision (as a measure of random error) and bias (as a measure of systematic error). Furthermore, these quantities will usually be expressed with respect to some reference value, e.g. the ‘true’ area of the wound, giving measurement results of precision and bias in dimensionless per-unit or percentage terms.
Precision
Precision is the basic level of variation in a series of repeated measurements. In absolute terms, it shall be defined for a single set of measurements as the standard deviation of the set of measurements. In this thesis calculation of the precision of either manual or algorithmic delineation measurements of wound area is made across a range of wounds. In order to compare the precision of measurements from different wounds it is necessary to express precision as a fraction of the ‘true’ value, the best estimate of which is regarded as the mean value of the set of measurements. This assumes, however, that a mean value is an unbiased estimate of the true area of a wound. A statement of precision does not indicate whether a bias exists or not. Additionally, the true level of precision is independent of the number of measurements made, meaning that the term ‘delineator precision’ refers to the expected variability in a single measurement of the area of a wound produced by manual delineation. The definition for fractional precision used in this work, i.e. standard deviation of a random variable divided by the mean value of the variable, may be identified as the
coefficient of variation (Topping, 1972).Bias
The term bias refers to the difference between the expected value of a wound area estimate and the true area of that wound. This implies that the true area of the wound is known or that a high-accuracy estimate is available. The appraisal of manual delineation performance in Chapter 4 will show that the acquisition of such an estimate is confounded by the fact that different delineators produce differing average area measurements, i.e. their measurements are mutually biased. Mutual bias is thus the measure of the difference between the average values of two sets of measurements, and applies only when the two averages are known to be, or are considered to be, significantly different. This indicates a difference of opinion between two delineators, so that certain parts of the image considered wound tissue by one delineator are not considered so by another delineator and vice-versa.
Accuracy
The term ‘accuracy’ is often used in the context of estimation. As stated above, measurement precision is independent of the number of repeated measurements made of any particular quantity. When many such measurements are made, the resulting estimate is usually expressed as the arithmetic mean measurement and the standard error of this mean result is quoted as the ‘accuracy’. Given the basic level of precision of any piece of measurement equipment, the accuracy or expected error of an estimate made by such equipment may be improved by increasing the number of readings taken. However, standard error does not account for any bias involved in a measurement process, so it cannot be considered a measure of the total expected error unless it is known that no measurement bias exists.
The term accuracy will thus be used as a general expression of error when it is appropriate to refer to both types of error in a single phrase. An example of this occurs in Chapter 5 where varying a parameter affects both bias and precision. A graph of bias v. precision is plotted and the path or locus formed as the parameter is varied is referred to as the ‘accuracy locus’.
1 Introduction |
||