A Retrospective Evaluation of Digital Wound Imaging to Predict Response to Hyperbaric Oxygen Treatment

   Knowledge regarding the biochemical reasons for wound healing failure has grown significantly in recent years; however, transforming this knowledge into clinically useful diagnostic tools and treatments has been slow to occur.

For example, studies have shown that growth factor deficits,¹ increased activity of metalloproteases,^2-4 reduced nitric oxide production,⁵ and low availability of oxygen in the wound bed⁶ are factors associated with nonhealing. Systemic dysfunction caused by advanced age,⁷ renal failure,⁸ and chronic diseases such as AIDS⁹ also adversely affect healing. However, laboratory tests to determine specific wound deficiencies, enabling rational selection of treatment, are not available. Instead, treatment decisions are made using subjective clinical assessment and the particular bias the caregiver may have regarding available treatment options. Clinical practice studies show that wound treatment is inconsistent and frequently not based on a rational analysis of the factors contributing to a chronic wound.¹⁰ Additionally, in some cases, treatments that are empirically sound may actually be deleterious to wound healing.¹¹ Despite recent advances in knowledge, the prognosis for many patients suffering from chronic, nonhealing wounds remains poor. For example, approximately 45% of patients with diabetes and chronic nonhealing wounds of the lower extremities will eventually require amputation.¹²

   As new, effective, and in all likelihood, expensive treatments become available, correct initial treatment selection and dynamic modification of regimens, based on wound response to treatment, must be applied to improve patient outcome and reduce cost. Biochemical assessment of the wound appears to be a rational approach to treatment selection and monitoring but is not yet standardized or validated and the cost is high. As an alternative, wound morphometry using digital wound images may be an effective means of evaluating wound response to treatment in real time. This will enable a dynamic approach, where treatments can be evaluated and subsequently changed if wound-healing response proves negligible.

   Wound areas or volumes have been used to evaluate the effects of treatment. Several measures have been proposed. These include wound area measurement by tracing the wound outline on an acetate placed over the wound,¹³ measuring the volume of saline required to fill the wound,¹⁴ digital image capture with color-based image analysis,¹⁵ and biochemical markers of wound treatment response.¹⁶ However, all these methods have unique limitations. Casts, tracings, and saline filling are invasive and time consuming, while many image analysis strategies require specialized equipment, training, and extensive set-up in order to optimize image quality. Biochemical measurements are expensive and the applicability to many types of wounds and treatments is as yet unproven.

   To test the hypothesis that digital images of wounds obtained with conventional digital cameras are of sufficient quality to make a meaningful assessment of treatment response, a digital archive of wound images was evaluated retrospectively in order to evaluate the healing characteristics of patients receiving hyperbaric oxygen treatment (HBOT) - specifically, that digital images taken during the first 3 weeks of treatment and other patient data (ie, age, smoking history, diabetes diagnosis, blood glucose, and serum creatinine) will predict the eventual response to HBOT.

Methods

   Patient selection for retrospective analysis of wound images. Patients with nonhealing wounds of the lower extremities who had failed previous therapy and who met the criteria for HBOT were evaluated. These included patients treated in an HBOT facility from December 2, 1997 to May 15, 2002. Patients were excluded if fewer than three images were available for evaluation or if their wound area could not be measured using digital photography (eg, wounds between toes, deep wounds, and wounds that spread around an extremity). This retrospective analysis of patient data was conducted with the consent of the Brooks City-Base Institutional Review Board.

   Inclusion/Exclusion criteria for HBOT. The criteria for HBOT were: 1) periwound transcutaneous oximetry (TcpO₂) was greater than 40 mm Hg during breathing of 100% oxygen (O₂) under normobaric conditions, 2) the patient could clear sinuses during compression, and 3) the patient was not claustrophobic. Patients were selected for analysis only if the objective of treatment was wound healing. In some cases, patients received HBOT for the purpose of demarking viable tissue before amputation. These patients were not included in this study.

   Wound healing therapy. Hyperbaric oxygen therapy consisted of breathing 100% O₂ for 90 minutes with 5 minute intervals of room air every 30 minutes at a pressure of 2.4 atmospheres absolute, given once every week day. In addition to HBOT, patients also received wound care including dressing changes and surgical debridement. Topical medications were applied to the wound bed during wound care in virtually all cases. These products included the following debridement agents: Collagenase, papain-urea-chlorophyllin copper complex sodium (Panafil®, Healthpoint, Fort Worth, Tex.), hydrogel wound dressing (Curasol®, Healthpoint, Fort Worth, Tex.), papain-urea debriding ointment (Accuzyme®, Healthpoint, Fort Worth, Tex.); and moistening agents: hydrogel wound dressing (Hydrogel and NuGel®, Ethicon Inc., Piscataway, NJ). Other products used include growth factors (rhPDGF), desquamation products (retinoic acid analogs), antibiotics (bacitracin), antiseptics (boric acid), topical antimicrobial and absorbent antimicrobials (Iodoflex™, Healthpoint, Fort Worth, Tex.), and flexible hydroactive dressings (Duoderm®, Conva Tec, Princeton, NJ). The clinician discontinued treatment when it was determined that HBOT was no longer producing a significant improvement in healing. The decision was subjective and based on the caregiver's clinical experience with wound healing and HBOT.

   Transcutaneous oximetry. Transcutaneous oximetry values were obtained on all patients before initiating their hyperbaric oxygen treatment series. Several TcpO₂ units (Radiometer Model TCM3, Copenhagen, Denmark) were used. These values were obtained in a standard fashion after the skin sites were prepared by shaving, cleansing, and dabbing with adhesive tape. The monitor leads were attached after the ionic TcpO₂ solution was placed in the membrane/ring electrode. The chest was used as the site for control values and the other values were obtained from skin near the wound.

   Digital photography. Digital images were obtained from patients before the first HBOT treatment and every week thereafter. A ruler was placed near the wound during image capture so each image evaluated had a uniform external standard as a reference for area determination (see Figures 1 and 2). A Nikon 990 Coolpix digital camera with 2M-pixels per image and a flash was used for image capture. The camera was placed at an angle perpendicular to the wound and at a distance of one-half of a meter to enhance contrast and resolution of the wound. Medical technologists trained in the use of the camera captured the images.

   Wound area measurement and validation of method. Image analysis was performed using Zeiss Image software (Carl Zeiss, Minneapolis, MN) on a PC with 800 GHz Pentium II processor with 256M of RAM. The following steps were performed during image analysis: 1) the wound perimeter was traced using a mouse directed cursor, 2) a 2-cm distance was measured on the ruler on the image field, 3) a computer application was run as part of the Zeiss Image software to calibrate area using the ruler as a standard, and 4) the wound area calculated by the software was reported.

   A validation study was performed by placing white paper circles of 2.7 cm², 4.9 cm², and 14.5 cm² area respectively, on the lower leg of a subject and then measuring the area using the procedures used for the evaluation of patient wounds as described.

   The response variable of interest is the normalized wound area percent (WA%). It is calculated by using the following formula,
   WA%= 100*(WA_{week x}/WA_{week 0})
where WA_{week 0} is the wound area in cm² before initiating treatment and WA_{week x} was the area on week x of the treatment.

   Statistical analysis of healing patterns. Robust and minimally responsive groups of patients were identified by visual examination of the clusters of the profile plots of percent normalized wound healing areas for the first 5 weeks of HBOT for all 29 patients reviewed. The assignment to groups, made by visual inspection, was validated by F-test. Statistical analysis was performed using normalized wound areas obtained from the first 3 weeks of HBOT and included age, gender, smoking history, diagnosis of diabetes, blood glucose, serum creatinine, blood urea nitrogen (BUN), wound TcpO₂ on room air and oxygen breathing, and initial wound area. This analysis was performed on 17 of 29 patients because complete clinical laboratory data were not available on all patients. The analysis of data was done by a general mixed effects linear model, MIXED procedure of SAS (SAS Institute, Inc., Cary, North Carolina).¹⁷ Relationships are deemed to be statistically significant if P <0.05.

   Determining the predictive value of early wound measurements. For the classification rule, a linear discriminant function similar to that reported to diagnose hepatitis B infections¹⁸ was used. The rule was validated by the leave-one-out cross-validation. Briefly, the discriminant rule is defined using data from all patients except one that is "left out." The left-out patient is then assigned to a healing group by application of the rule described by all the other patients. This process is repeated until all patients are assigned. The error rate, or the percentage of patients incorrectly assigned to minimal or robust healing, is reported.

Results

   Validation of experimental method using model wounds. The validation study showed that the method was precise and accurate, although a bias towards slightly over-estimating areas of smaller wounds was noted. The model wound of 14.5 cm² was overestimated by 0.8% with a standard deviation of 2.0%. This difference was not statistically significant (P = 0.2587). Smaller wounds with a known area of 2.7 cm² and 4.9 cm² were over-estimated by 6.3% and 11.0% (standard deviations of 1.8% and 3.6%, respectively). The overestimation in these cases was found to be statistically significant (P <0.0001). This finding is expected because the outside wound margin is traced with the cursor leading to a systematic overestimation of the area.

   Assignment of patients to robust and minimal healing groups by visual inspection of normalized healing profiles. Profiles of normalized wound area for each patient during the first 5 weeks of treatment were plotted (see Figure 3). Visual inspection resulted in two groups: minimal and robust responders to HBOT. The assignment of patients to these groups was validated by statistical analysis (P = 0.0001). Robust responders were characterized by a continuous, sustained reduction in normalized wound area during the first 5 weeks of HBOT. Review of patient records showed that nonhealing wounds were attributed to the following underlying conditions: diabetes (41.4%), peripheral vascular disease (34.6%), unknown causes (10.3%), malnutrition (6.9%), venous stasis (3.4%), and a spider bite (3.4%).

   Relationship of patient factors and healing response. Minimally responsive patients showed a delayed response, with significant wound area reduction not occurring until the second week of HBOT. Robust and minimal response groups were similar with respect to gender (71% and 83% male) and positive smoking history (67% and 70%, respectively). Frequency of reported diabetes was higher in the minimal response group, (58% versus 28% in the robust); however, this difference was not statistically significant. The mean values and standard deviations for parameters from 17 patients is shown in Table 1. Figure 4 shows the relationship between periwound TcpO₂ during air and oxygen breathing. Patients with higher TcpO₂'s during oxygen and air breathing do better than patients with lower values.

   Statistical analysis was performed to determine how demographic factors and clinical laboratory values are related to normalized wound areas during the first 3 weeks of treatment. A subset of 17 patients, for which complete laboratory values were available, was used for this purpose. The analysis enables the identification of factors (eg, age) that have a statistically significant effect on normalized wound areas measured during the first 3 weeks of HBOT. However, the method does not enable precise definition of levels that are associated with robust or minimal healing. For example, using this method, the authors found that age has a significant impact on reduction of normalized wound area made during the first 3 weeks of HBOT (P = 0.0072). However, this does not necessarily mean that older patients heal more rapidly, but rather that age affects normalized wound area in a complex manner. Similarly, a statistically significant relationship was present between age and week of treatment (P = 0.0075). Stated another way, this implies that the rate of healing or the change in normalized wound area during HBOT is affected by age. Serum glucose affected normalized wound area in the two groups differently (P = 0.0012). Serum creatinine also had an effect on wound healing (P = 0.0127). The results (see Table 1) and the findings obtained by statistical analysis were reconciled by considering that the healing response was complex and that more than two healing groups, minimal or robust, might be present in the HBOT-treated population. More data are required for a more complete assessment of interaction between clinical lab values, wound measurements, and healing response.

   Evaluating the ability of early wound images to predict eventual response to HBOT. The authors hypothesized that normalized wound areas, measured during the 3 weeks of HBOT following the first week of treatment, would classify a patient into one of the two groups mentioned above. Images from the first 3 weeks lead to correct classification of all minimally responsive patients. However, two out of 16 robust responders were incorrectly classified as minimal responders, resulting in a total error of misclassification of 6.9%.

Discussion

   The current report demonstrates that wound area measurements, obtained from digital images, using inexpensive, commercially available digital cameras are useful in evaluating response to HBOT. Normalized wound areas obtained during the first 3 weeks of HBOT predict, with 100% precision, if a patient will be minimally responsive to a full (6 weeks or longer) course of HBOT. This observation is similar to that of Phillips et al¹⁹ that showed that during treatment of venous foot ulcers, reduction in wound area of at least 40% during the first 3 weeks of treatment predicted complete healing in 70% of cases. van Rijswijk and Polansky²⁰ showed that during treatment of Stage III and Stage IV pressure ulcers with a hydrocolloid wound dressing, wound areas measured in the first 2 weeks predict eventual response to a longer course of treatment. Briefly, patients who showed a 39% or greater reduction in wound area, as measured by acetate tracing of the wound, during the first 2 weeks, showed a more expeditious healing response than other cohorts. When taken together, these results suggest a finding similar to that reported here - that early wound healing measurements may be useful in predicting treatment response of many types of wounds to different types of treatments. Clearly, prospective studies that include larger numbers of patients than that used here, and a wider scope of clinical laboratory data, are expected to greatly improve predictive power. Such an approach may have great impact on improving clinical outcomes and reducing the cost of treatment by reducing the number of futile treatments.

   The main current findings show that wound area measurements enable comparison of treatment effects in a manner that is both objective and quantifiable. Treatment decisions - namely, changes in therapy - can be made early in the course of therapy using this approach. For example, if a patient were classified as minimally responsive to HBOT, it would be clear that the HBOT is less than completely effective and alternatives should be considered. When digital image data are coupled with demographic and clinical laboratory values, significant improvements in the quality of wound care can be achieved more rapidly than is currently possible. Moreover, this approach enables comparisons between wound care clinics. The use of objective measures will facilitate meaningful dialogue between caregivers and ultimately lead to improved wound care, better patient outcomes, and reduced treatment cost.

   The proposed approach requires refinements because in two out of 16 cases, patients eventually showing a robust response to HBOT were incorrectly classified as minimally responsive. Decision rules should err on the side of delivering treatment, at least up to some point. As more data are added, robust and conservative decision rules enabling difficult decisions (eg, discontinuing treatment) can be formulated. Clearly, a balance must be struck between the level of patient care and benefit to be gained by treatment. Some wound treatments, including HBOT, are expensive and in some cases futile.

   A rational, objective approach to wound care, using wound area measurement as a treatment guide, may reduce the number of HBOT treatments given, improve outcome, and reduce overall wound care cost. However, several notable limitations to digital imaging must be noted. Useful digital information cannot be obtained from all wounds. Wounds between toes, for example, cannot be evaluated by this method. Evaluation of diffuse breakdown of skin across large areas such as in the pretibial area is also problematic. However, for the patient data examined here, greater than 80% of patients with slowly healing wounds in the lower extremities can be evaluated in a meaningful way with simple digital imaging.

   Another limitation is the lack of information regarding changes in the depth of the wound over time. In some cases, significant reduction in wound volume may occur with little change in area. Although not measured in this study, the majority of wounds assessed were relatively shallow; thus, volume reduction was expected to be highly correlated with area reduction. However, additional work is required to validate this assumption. In a study to compare different methods of measuring wound areas, Plassmann et al²¹ showed that several methods of volume using saline, materials for making dental impressions, and computerized video analysis all provide reasonable precision and accuracy, and can be used in the clinic at a reasonable cost with minimal training. A strategy involving video capture of wound images and computation of wound volumes seems to be the most reasonable alternative in terms of cost and ease of use. Investigators performing prospective clinical studies should carefully consider which technologies are available and which methods have been validated for wound assessment.

Conclusion

   Wound healing area measurement can be a useful tool in monitoring the effectiveness of treatment and perhaps a means of rationally determining when a treatment should be modified. However, this approach is empirical and provides no information that justifies use of any particular treatment. In order to make the best decision about treatment selection, a far more comprehensive assessment of the underlying biology of a problem wound in a particular patient is required. A more complete assessment may include information regarding the wound microenvironment - in particular, the availability of nutrients and oxygen; immune system activity; enzymatic activity; and systemic considerations including patient genotype and the severity and length of chronic illness. At this time, truly comprehensive wound assessment is beyond the reach of current technology, at least at a reasonable cost. However, when a comprehensive assessment becomes available, treatments will be selected for their ability to correct the most important deficiencies.

   Is this approach feasible? Cancer treatment is currently undergoing a revolution brought on by the technological innovation that has permitted the survey of thousands of genes and expressed proteins simultaneously at low cost. Although the utility of this approach in treatment selection will take several years to validate, an individual's gene^19,22 and protein²³ profile is highly predictive of cancer treatment response and outcome. Using wound area measurements is a low-cost and low technology approach to a vexing problem in wound care: evaluation of treatment response. The approach used here is only an interim solution until that time when comprehensive biochemical wound assessment becomes available. However, a systematic approach to treatment evaluation will still provide significant advantages over current methods that are vulnerable to caregiver bias and to subjective measures of treatment response.

1. Cooper DM, Yu EZ, Hennessey P, Ko F, Robson MC. Determination of endogenous cytokines in chronic wounds. Ann Surg. 1994;219(6):688-691; discussion 691-692.

2. Trengove NJ, Stacey MC, MacAuley S, et al. Analysis of the acute and chronic wound environments: the role of proteases and their inhibitors. Wound Repair Regen. 1999;7(6):442-452.

3. Ashcroft GS, Horan MA, Herrick SE, Tarnuzzer RW, Schultz GS, Ferguson MW. Age-related differences in the temporal and spatial regulation of matrix metalloproteinases (MMPs) in normal skin and acute cutaneous wounds of healthy humans. Cell Tissue Res. 1997;290(3):581-591.

4. Vaalamo M, Mattila L, Johansson N, Kariniemi AL, Karjalainen-Lindsberg ML, Kahari VM, et al. Distinct populations of stromal cells express collagenase-3 (MMP-13) and collagenase-1 (MMP-1) in chronic ulcers but not in normally healing wounds. J Invest Dermatol. 1997;109(1):96-101.

5. Witte MB, Barbul A. Role of nitric oxide in wound repair. Am J Surg. 2002;183(4):406-412.

6. Davidson JD, Mustoe TA. Oxygen in wound healing: more than a nutrient. Wound Repair Regen. 2001;9(3):175-177.

7. Ashcroft GS, Horan MA, Ferguson MW. The effects of ageing on cutaneous wound healing in mammals. J Anat. 1995;187(Part 1):1-26.

8. Attinger CE, Ducic I, Neville RF, Abbruzzese MR, Gomes M, Sidway AN. The relative roles of aggressive wound care versus revascularization in salvage of the threatened lower extremity in the renal failure diabetic patient. Plast Reconstr Surg. 2002;109(4):1281-1290; discussion 1291-1292.

9. Burns J, Pieper B. HIV/AIDS: impact on healing. Ostomy/Wound Management. 2000;46(3):30-40.

10. Plassmann P, Peters JM. Recording wound care effectiveness. Journal of Tissue Viability. 2002;12(1):24-28.

11. Ennis WJ, Meneses P. Clinical evaluation: outcomes, benchmarking, introspection, and quality improvement. Ostomy/Wound Management. 1996;42(10A Suppl):40S-47S.

12. Dormandy J, Heeck L, Vig S. Major amputations: clinical patterns and predictors. Semin Vasc Surg. 1999;12(2):154-161.

13. Majeske C. Reliability of wound surface area measurements. Phys Ther. 1992;72(2):138-141.

14. Berg W, Traneroth C, Gunnarsson A, Lossing C. A method for measuring pressure sores. Lancet. 1990;335(8703):1445-1446.

15. Plassmann P, Jones TD. MAVIS: a non-invasive instrument to measure area and volume of wounds. Measurement of area and volume instrument system. Med Eng Phys. 1998;20(5):332-338.

16. Dodson W, Kalns J, Scruggs J, Kiel J, Wolf G. 3-Nitrotyrosine predicts healing in chronic diabetic foot wounds treated with hyperbaric oxygen. WOUNDS. 1999;11(6):129-136.

17. Littell RC, Milliken GA, Stroup WW, Wolfingen RD. SAS System for Mixed Models. Cary, NC; 1999.

18. Neiss, Ohlen. Statistics Entscheidungshilfe zur Fruherkennung der chronischen Hepatities B. Tagungsberichte, III. Symposium on operations. Research. 1978;35:391-403.

19. Phillips TJ, Machado F, Trout R, Porter J, Olin J, Falanga V. Prognostic indicators in venous ulcers. J Am Acad Dermatol. 2000;43(4):627-630.

20. van Rijswijk L, Polansky M. Predictors of time to healing deep pressure ulcers. Ostomy/Wound Management. 1994;40(8):40-51.

21. Plassman P, Melhiush J, Harding K. Methods of measuring wound size: a comparative study. Ostomy/Wound Management. 1994;40(7):50-60.

22. Torhorst J, Bucher C, Kononen J, et al. Tissue microarrays for rapid linking of molecular changes to clinical endpoints. Am J Pathol. 2001;159(6):2249-2256.

23. van ‘t Veer LJ, Dai H, van de Vijver MJ, et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature. 2002;415(6871):530-536.