An Accurate and Reliable Method of Thermal Data Analysis in Thermal Imaging of the Anterior Knee for Use in Cryotherapy Research

Methods 

The study received approval from Bolton local research ethics committee and the Faculty of Health Ethics Committee, University of Central Lancashire (UCLAN); it also conformed to the 1964 Declaration of Helsinki. Nine patients were recruited from the waiting list of the Bolton Primary Care Trust and the Anterior Knee Pain Clinic at Royal Bolton Hospital, Lancashire, England. All patients provided written informed consent. As part of the standardized data collection procedure,20, 23, 25, 26 patients were asked before data were collected to adhere to the following; no alcohol consumption 24 hours before their appointment, no smoking, consumption of caffeine, no participation in physical activity 2 hours before their appointment, and no use of ointments or heat rubs on the day of their appointment.

Exposed lower limbs were given a 15-minute acclimatization period, away from heat, sunlight, and drafts to allow stabilization to ambient room temperature.20, 23, 25, 26 With the patient in a stable and relaxed position, with knee angle standardized to 45°, small thermally inert markers were placed on known anatomic landmarks as detailed in table 1.

Table 1. Placement of Markers

	Marker Location	Marker Placement
	Superior	Base of patella
	Medial	Medial border of patella tendon margin at tibiofemoral joint line level
	Lateral	Lateral border of patella tendon margin at tibiofemoral joint line level
	Inferior	Tibial tubercle

Thermal imaging data of the knee were collected using a thermal camera.a The camera was mounted on a tripod, the height of which was determined by the camera angle, and the lens to knee (object) distance standardized to 45° and 0.8m, respectively. The angle of 45° was to ensure that the camera lens was parallel with the anterior surface of the patella. The distance of 0.8m was based on our earlier work27 using a reference object with a known area of 0.5m2. An image of the reference object was taken with the camera at distances of 0.8, 1.0, 1.2, and 2.0m. The greater distance of 2.0m gave a pixel resolution of just 12 pixels/m2, whereas at 0.8m the camera gave a much better resolution of 90 pixels/m2. Technical specificationa of the camera outlines a minimum operating distance of 0.3m. It was not possible in this study to bring the camera closer than 0.8m to the knee because of the patients’ positioning.

We took a baseline thermal image of the knee before the thermal washout. The temperature recorded in this image was considered the normal resting temperature for each patient.

Thermal Data Analysis 

Quantification of the thermal images was facilitated by Bespoke computer software,c designed and written in-house.

Patella region identification 

After the thermal imaging data were collected, a region similar in geometry and position to that of the patella was defined based on the location of the 4 markers positioned by the clinician (fig 1). A Matlab programd was written to digitize the marker positions and subsequently calculate the mean patellar region. The thermal imaging data were plotted as a 3-dimensional plot then orientated in a coronal plane view (see fig 1). The computer program then delivered instructions in a text format describing the order in which the markers should be digitized. On completion of the digitizing process, virtual markers were offset from the actual markers as a percentage of the vector solution, between the proximal and distal markers to define the y axes and between the medial and lateral to define the x axes. The virtual markers were defined as: 15% distal from the proximal marker, 30% proximal from the distal markers, 15% lateral from the medial marker, and 15% medial from the lateral marker. Following this, the image was cropped to a rectangular shape based on the minimum y and minimum x marker coordinates, the dimensions of which were defined by the axis lengths (x, y). An ellipse was defined, with x and y coordinates defined by the axis within the cropped image and z coordinates defined as the minimum and maximum on the z scale (the temperature scale). The z axis allowed for extraction of the temperature gradient of the patella region. The average temperature of this region was then defined as the mean value along the z axis within the ellipse.

View Large Image Download to PowerPoint

Fig 1. The visual output of the thermal data when processed in the custom written program, the crosshair is the digitizing tool. The coronal plane view of the anterior knee.

Interrater reliability 

We wrote a modified version of the Matlab program within which a single set of data was looped through 10 trials providing instructions for the digitizing process: each coordinate digitized was exported to Excel.e Ten volunteers (5 men, 5 women) from the UCLAN staff were recruited as raters to follow the instructions provided within the program and to digitize the image. Before the test, the raters were told what the marker appearance was and how to digitize it; no additional training was provided. From these data interrater reliability was assessed with 1-way analysis of variance (ANOVA) (P≤.05); from this we calculated the intraclass correlation coefficient (ICC), based on the standard error of the mean as defined by Snedecor and Cochran.31

Results 

We calculated the mean x and y coordinates of the proximal, distal, medial, and lateral markers. The individual rater’s coordinates were then plotted about the ensemble mean of the raters, the standard deviation about the mean did not exceed 1 pixel (fig 2).

View Large Image Download to PowerPoint

Fig 2. (A) Proximal marker; (B) distal marker; (C) medial marker; and (D) lateral marker. The dispersion of each rater’s (+) mean digitized point about the ensemble mean from the group of raters (□).

The 1-way ANOVA identified significant differences between the raters at the proximal x, distal x, and medial x markers. A Tukey test found that the significant effect was caused by a single rater incorrectly digitizing these markers. The 95% confidence intervals (CIs) show close upper and lower bounds, not greater than 1 pixel in magnitude, which supports the pairwise comparison results where the significance was due to a single rater. The ICC produced coefficients within acceptable bounds, ranging from .82 to .97 (table 2).

Table 2. Interrater Reliability Measures in Terms of a 1-Way ANOVA, ICC, and Confidence Intervals

	Coordinates	1-Way ANOVA P Value	ICC	95% CI⁎
	Proximalx	.001	.87	62.861−63.296
	Proximaly	.153	.98	117.208−117.605
	Distalx	.000	.86	51.909−52.227
	Distaly	.466	.99	218.597−218.886
	Medialx	.025	.95	74.994−75.280
	Medialy	.047	.96	190.925−191.251
	Lateralx	.054	.96	39.976−40.296
	Lateraly	.010	.93	190.014−190.307

⁎ Measured in pixels, based on the resolution of 90pixel/m2.

Discussion 

Cryotherapy remains the most commonly used modality in the acute management of musculoskeletal injuries.1, 32 Optimum treatment modality and application time is yet to be established. The thermal imaging camera is proposed as a useful research tool to advance understanding and develop sound clinical guidelines in cryotherapy. Currently, an accurate, reliable method of thermal imaging data extraction and analysis is lacking.

An AMS was developed for thermal imaging data analysis of the knee and interrater reliability of the system was tested. Reliability is defined as the extent of reproducibility or consistency of values measured under identical conditions.33 The AMS defines an anatomic frame over the anterior knee based on specific bony anatomic landmarks about the knee.

Previously, Devereaux et al15 analyzed thermal imaging through visual evaluation and comparison of pathologic and nonpathologic areas. This method lacks quantification, and is subject to the opinion of the clinician assessing the image, and therefore has poor interrater reliability.

Davidson and Bass13 employed the use of a thermal profile line, which consists of 2 transverse placements; through the mid patella and along the lower border of the patella and longitudinally through the long sagittal diameter of the patella. The thermal profile line reflects a temperature profile of the line along which it is placed. Although this adds value to the thermal image, the lack of specificity prevents consistency of placement for repeated measurement, and does not account for variation in size and anatomy.

Siegel et al16 analyzed 4 temperature reference sites, based on anatomic landmarks: the suprapatella, infrapatella, and mid region of medial and lateral menisci of the patellofemoral joint—stating that the use of these reference sites allowed for comparison of thermal change on repeated measurement within each knee. Although, in principle this method utilizes anatomic landmarks, location of the landmarks takes place on the thermal image, consequently limiting reliability of placement. This limitation is also true of the thermal profile line advocated by Davidson and Bass.13

Clinically, anatomic landmarks are identified through palpation—visualization alone is not sufficiently reliable. Thermal imaging does not provide the clarity necessary to reliably identify these points. The AMS addresses this problem by using thermally inert markers on the knee, providing a standardized patient specific method that accommodates for individual anatomy. The system is inherently self-normalizing. Figures 3A and 3B highlight the ease of landmark location on the thermal image when using the AMS, when compared with no markers. Skin markers are clear and successfully define an anatomic frame, through a range of temperatures (see fig 3).

View Large Image Download to PowerPoint

Fig 3. Thermal image of anterior knee (A) with no marker system and (B) with AMS.

Our results show the ICC produced high coefficients, ranging from .82 to .97 (see table 2). This indicates that interrater reliability is adequate, and highlights the consistency of the analysis between raters (clinicians). Although the digitizing process is a potential source of error, the results show that the error rate is low.

The measure used to quantify the repeatability of the algorithm was based on the dispersion of the digitized points about a mean. This showed high repeatability, however, it did not actually account for variance in temperature. This can be accounted for when the error about the mean digitized point is calculated as a percentage error of the vertical and horizontal pixel distances between the markers. The results show the horizontal error to be 1.3% and the vertical error to be .75%. This can be carried forward to the temperature calculation, where there could be errors of these magnitudes. However, percentage errors of this magnitude could not be deemed to be significant.

The overall reliability of the AMS is acceptable and the magnitude of potential error is nonsignificant.

Conclusions 

The AMS provides a standardized, reliable, self-normalizing method for thermal imaging data analysis that has been shown to be effective in a clinical setting. AMS enhances the worth of thermal imaging in cryotherapy research, and as such, will serve to inform current clinical practice about cryotherapy application.

Although the anterior knee was the focus of this study, the principles are applicable to other anatomic regions of the body. Because the system utilizes definition of a specific area for data analysis, it is imperative that at least 3 markers be used to form the system. It is preferable that they be placed on bony landmarks to ensure reliability.

The AMS will be used in future work involving analysis of skin surface temperature data, and is proposed as a standard for thermal imaging data analysis.