Proceedings 3rd Annual Conference IEEE/EMBS Oct.5-8, 001, Istanbul, TURKEY 1 of 4 NEW DIFFERENTIAL DATA ANALYSIS METHOD IN THE ACTIVE THERMOGRAPHY. J. Rumiński, M. Kaczmarek, A. Nowakowski Deartment of Medical and Ecological Electronics, Technical University of Gdańsk, Gdańsk, Poland Abstract - A new method for arametric images synthesis in active thermograhy is roosed. Basing on exerimental results the thermal model of an observed object is defined. Images of reconstructed model arameters are formed allowing clear visualization of object thermal roerties. Studies made on hantoms and in-vivo measurements are resented. Possible alications of the method are discussed. Keywords - Active thermograhy, image synthesis, arametric imaging I. INTRODUCTION During the last decades of the XX-th century different methods to enhance quality of thermal images or to create quantitative detection rules for medical diagnosis were roosed. For examle alication of microwave irradiation may imrove contrast in tumor detection [1]. Introduction of statistical descrition and differential analysis of thermal image regions [] has significant ractical value. Detecting of hysiological function changes from time sequential thermal images of skin surface is also ossible as it was roosed in [3]. In arallel to classical thermograhy new dynamic measurement techniques were develoed for nondestructive testing and evaluation (NDT/NDE) [4] indicating ossibility of medical alications [5]. Here a new method for arametric images synthesis in active thermograhy is described. Some results of hantom and in-vivo studies are resented. II. METHODOLOGY In active thermograhy external or internal excitations are alied to an object to modify it s temerature distribution. Measurements of temerature during and/or after the excitation roduce data which can be rocessed to extract information about dynamic roerties of the object. In ulsed thermograhy, used in the reorted study, a target object is excited during a given time eriod t 1 U P ( t) = A* ( Φ( t) Φ( t t1) ), (1) where: A amlitude of the excitation, Φ (t) - Heaviside, ste function. Deending on a source of excitation the object temerature increases (heating e.g. otical source) or decreases (cooling e.g. a fan or an air conditioner). As it was analyzed by Buettner [6] temerature decrease of human tissues after removal of a heating source is described by a sum of exonents. Exonential character of temerature changes is also valid for cooling object using forced convection. In our studies we focus on otical sources as well as on forced convection generators since they are chea and simle to use. Let s assume that temerature of a single oint of an object can be measured over a given time eriod t what is reresented by a set of samles S c = { s1, s, s,.., s c }, () where: c total number of samles measured after removal of a heating source or during forced convection cooling, index of a measurement oint. Alying exonential model for thermal transients of the object the estimate of a samle is given by L ˆ = 0 + *ex t i s i B Bl ; i = 1..c, (3) l= 1 τ l where: B 0 - an offset, L a number of exonents in the model (or number of layers in the equivalent electrical RC circuit), Bl and τ l a set of arameters for each exonent (layer), t i - measurement time. Parameters of the model can be determined using one of curve fitting methods, assuming given value of the fitting error ε s sˆ 1 1 < ε. (4) In the reorted studies we used the Marquardt method with the modification given by Bevington [7]. As a result of the fitting rocedure a set of model arameters is created for any of simultaneously measured object oints forming new arametric images. The number of useful arametric images deends on the object structure and measurement accuracy. In our studies one and two exonential models equivalent to one and two hysical layers are sufficient (A1= B 1, A=1/ τ, A3= B, A4=1/ τ ). Synthesis of arametric images in active thermograhy is useful only if new information is carried on by those images, or if there is an enhancement of contrast, comaring to traditional thermograms. The validity of the roosed method is tested basing on exeriments and on simulations. Simulation based verification of the roosed method has been done using Finite Element Method heat-flow modeling. As an object a uniform cylinder has been created (diameter 0cm, deth 7 cm) with embedded in the center small cylinder (diameter 4cm, deth cm). About 5000 elements were used to comose the object. The object was heated during 30 s with the radiant flux 000 (W/m). Exerimental verification of the method was erformed on hantoms and in-vivo. In the alied ulsed thermograhy the otical heating excitation was usually lasting 30s. The thermal camera, Agema Thermovision 900, was laced 55 cm far from a target object. The thermal resonse was measured after excitation of the object by a thermal ulse 1 0-7803-711-5/01$10.00 001 IEEE
Reort Documentation Page Reort Date 5OCT001 Reort Tye N/A Dates Covered (from... to) - Title and Subtitle New Differential Data Analysis Method in the Active Thermograhy Contract Number Grant Number Program Element Number Author(s) Project Number Task Number Work Unit Number Performing Organization Name(s) and Address(es) Deartment of Medical and Ecological Electronics, Technical University of Gdañsk, Gdañsk, Poland Sonsoring/Monitoring Agency Name(s) and Address(es) US Army Research, Develoment & Standardization Grou (UK) PSC 80 Box 15 FPO AE 09499-1500 Performing Organization Reort Number Sonsor/Monitor s Acronym(s) Sonsor/Monitor s Reort Number(s) Distribution/Availability Statement Aroved for ublic release, distribution unlimited Sulementary Notes Paers from the 3rd Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 5-8 October 001, held in Istanbul, Turkey. See also ADM001351 for entire conference on cd-rom., The original document contains color images. Abstract Subject Terms Reort Classification Classification of Abstract Classification of this age Limitation of Abstract UU Number of Pages 4
Proceedings 3rd Annual Conference IEEE/EMBS Oct.5-8, 001, Istanbul, TURKEY of 4 generated by a set of lams (1000W) located 55 cm far from the object surface. Thermal images were catured every 1 s during 10 seconds (10 images) after switching the heat source off. As an image is constructed by a matrix of 04x18 ixels this gives 611 measurement oints for the image. Different target objects were used in exeriments. Some reliminary results on hantoms have been ublished in [8]. Here results of extended studies on hantoms, and additionally in-vivo measurements are discussed. Phantoms were made as gelatine objects (diameter 0cm, deth 7 cm) with some embedded materials as it is resented in figure 1. values is lower than 6%. For the -exonent model high Fig. 3. Simulation results: Temerature decrease of the hantom F1 materials - 1-layer model arameters. Temerature distribution in the hantom 10 s after heating was switched off is resented. differences (5-10 times) are observed for A arameter values for each of exonents. Big differences exist for each of exonents. Such big differences are exected according to the equation of temerature decrease after removal of the heating source, resented by Buettner [6] Fig. 1. Two classes of hantoms used in studies. In the hantom F1 such materials as W1 silicon, W acryl, W3 lexiglass, W4 and W5 gelatine with different density were embedded equo-distantly to the gelatine surface. Materials in the second hantom were embedded 3 mm (W6-brass, W8-coer) and 5 mm (W7, W9-aluminum, W10-coer, W11-brass) below the hantom surface. In-vivo measurements were erformed on eight anesthetized igs according to all rocedures acceted by the local Committee of Ethics and Good Practice. u (n+ 1) ( 0, ) = + T t To Ti e /(n + 1) n= 1, (5) k π where: u = t ; (6) ρ C 4 d k thermal conductivity; ρ C -volumetric heat caacity; d deth, for boundary conditions x = d; t = 0; T > To; (7) x d; t = 0; T = To+Ti(1-x/d). Basing on this formula and on our results it can be stated that for current measurement accuracy in thermograhy the - layer model is sufficient to reresent measured data. If temerature changes introduced by an excitation source in the object are high enough then the -layer model estimation is better correlated to the measured data comaring to 1-layer model (fig 4). However, if temerature changes introduced in an object are Fig.. Pig with indicated burned fields. Symmetric injuries on the ig skin were generated using reheated burning tools of controlled temerature which were alied during a given eriod of time. Deth of resulting burns was evidenced by histoathologic evaluation as it is resented in [9]. Placement of burns is shown in Fig.. III. RESULTS Results of simulations erformed for materials as used in the hantom F1 clearly indicate differences in roosed arameters values (Fig. 3). For the 1-exonent model a high correlation between values of the arameter A calculated for the simulated and for the measured data exists. In reeatable exeriments the maximal deviation between those Fig. 4. Differences between 1-exonent and -exonent original data estimation. small the 1-layer model is suitable because the limited measurement accuracy is not allowing for roer
Proceedings 3rd Annual Conference IEEE/EMBS Oct.5-8, 001, Istanbul, TURKEY 3 of 4 determination of higher order model arameters. In figure 5 an examle of traditional (static) thermogram and the A arametric image of the hantom F1 are resented. Placements of embedded materials is indicated in fig. 1. In the A image W1, W and W3 fields are easy to be detected while in the static thermogram only W3 is observed. For the F hantom between 0.01 and 0.06. The relation between those injuries and healthy skin roerties is shown in figure 9. Evident difference between injuries and healthy skin can be used to Fig. 5. Classical thermogram (left) and the A arametric image of the F1 hantom. the -layer model gives much better results than the 1-layer model. The A4 arametric image (Fig. 6 right) resents all fields with the embedded materials while they are invisible in the thermogram. Even more, it is ossible to distinguish between materials embedded closer to the surface (W6, W8) and deeer (other fields). Fig. 7. The A arametric image of the ig No 3 reconstructed for data measured 30 minutes after burning. Below a hotograhy of measured area taken at the same time. Fig. 6. Classical thermogram (left) and the A4 arametric image of the F hantom. Very interesting results are also obtained for in-vivo measurement data. More than 150 measured sequences were rocessed to create 1-layer and -layer arametric images. However because of high fluctuations of the measured signal and small temerature changes in objects, values of the second set of arameters in the -layer model were close to zero. Concluding, the 1-layer model seems to be suitable for screening uroses as the following roerties has been observed: - injuries directly caused by the burning tool are characterized by smaller values of A1 and A arameters than for the healthy skin (e.g. a black field in fig. 7), - injuries develoed after burning (usually after 10-4 hours) are characterized by higher values of A (e.g. white areas in fig. 8), while A1 has different values (smaller and higher than for the healthy skin) deending on the animal. In the first class of injuries A arameter values are below 0.019, usually 0.009. In contrary, for the second class of injuries values of A arameters are grater than 0.09. The healthy skin is characterized by A arameter values Fig. 8. The A arametric image of the ig No 4 reconstructed for data measured 4 hours after burning. Below a hotograhy of measured area taken at the same time. Fig. 8. Characteristics of temerature changes for two classes of injuries and healthy skin.
Proceedings 3rd Annual Conference IEEE/EMBS Oct.5-8, 001, Istanbul, TURKEY 4 of 4 establish threshold values for the A arameter. It is ossible then to distinguish automatically secific tyes of tissue changes. In figure 9 an examle of the thresholding oeration (threshold=0.031) is resented, alied to the A arametric image for data collected from ig No 4, 96 hours after burning. Since A arameter values are reeatable and related to thermal roerties of the object they can be used in differential analysis within a one image and between different images reconstructed for articular objects. IV. DISCUSSION AND CONCLUSION Parametric images synthesized from a sequence of thermograms reresent thermal roerties of a tested object. Differentiation of secific materials (tissues) is ossible using comarison of corresonding τ l arameters. Phantom studies results as well as in-vivo results allow to identify such object roerties (materials) that are imossible to be indicated with traditional thermograhy. However further enhancement must be done to the measurement setu. Increase in temerature resolution of a thermal camera and in temerature contrast of an object (imrovements in the excitation system) should allow the discrimination of materials with small differences of thermal roerties (mainly conductivity). Extended medical investigations must be done to establish calibration rocedures for direct determination of the thickness of burned tissues. Other ossible alications of the resented method may concern all cases where thermal roerties of external tissues are affected, see e.g. [1] for cardiologic cases. REFERENCES Fig. 9. Binary image of the A arameter after thresholding oeration (threshold=0.031) alied for data collected from ig No 4, 96 hours after burning. Below a hotograhy of measured area taken at the same time. Additionally, interesting founding was observed for burned fields generated by high temerature burning tools alied for short time (e.g. the field No 7). In the A arameter images they were characterized by a very high value (usually -4 times greater than for the healthy skin) in comarison to other fields. Fig. 10. The offset image and the A arameter image of the ig No 4 reconstructed for data measured 48 hours after burning. [1] Thomson J.E., Simson T.L., Caulfield J.B.,: Thermograhic tumor detection enhancement using microwave heating. IEEE Trans. on Microwave theory and techniques, 6 (1978), 573-580. [] Mabuchi K., Chinzei T., Fujimasa I., Haeno S., Abe Y., Yonezawa T.: An image-rocessing rogram for the evaluation of asymmetrical thermal distributions. Proc. 19th Int. Conf. IEEE/EMBS, Chicago, USA, (1997), 75-78. [3] Fujimasa I., Chinzei T., Mabuchi K., Converting algorithms for detecting hysiological function changes from time sequential thermal images of skin surface, 1995 IEEE Engineering in Medicine and Biology, IEEE. Part vol., 1709-10. [4] Maldague X., Alications of Infrared Thermograhy in NonDestructive Evaluation, Trends in Otical Nondestructive Testing, Pramod Rastogi, (000), 591-609. [5] Rumiński J., Kaczmarek M., Nowakowski A., Hryciuk M., Differential Analysis of Medical Images in Dynamic Thermograhy. Proc. of the V Conf. on Alication of Mathematics in Biology and Medicine, Ustrzyki, Poland, (1999), 16-131 [6] Buettner K., Effects of extrem heat and cold on human skin, J. Al. Phys, 3: 691-713, 1951. [7] Bevington P.R., Robinson D.K., Data Reduction and Error Analysis for The Physical Sciences, McGraw-Hill Higher Education, 1991. [8] Rumiński J., Nowakowski A., Kaczmarek M., Modelbased arametric images in dynamic thermograhy, Polish J Med Phys & Eng; 6(3):159-164, 000. [9] Kaczmarek M., Nowakowski A., Renkielska A., Grudziński J., Stojek W., Investigation of skin burns basing on active thermograhy, see the current roceedings. [10] Nowakowski A., et al., Thermograhic and electrical measurements for cardiac surgery insection, see the current roceedings.