Real-Time Cal i bra tion of Tem per a ture Es ti mates dur ing Radio fre quen cy Ab la tion

Transcription

1 ULTRASONIC IMAGING 26, (2004) Real-Time Cal i bra tion of Tem per a ture Es ti mates dur ing Radio fre quen cy Ab la tion T. VARGHESE 1, 2 AND M.J. DANIELS 1, 3 1 Department of Medical Physics 2 Department of Biomedical Engineering 3 Department of Physics The University of Wisconsin-Madison Madison, WI tvarghese@wisc.edu Radio fre quen cy ab la tion is an in ter sti tial fo cal ab la tive ther apy that can be used in a percutaneous fash ion and per mits in situ de struc tion of hepatic tu mors. Re cur rence rates af ter rf ther apy are as high as 34-55%, due to dif fi cul ties in ac cu rately iden ti fy ing the zone of ne cro sis (ther mal le sion) be cause of the low in trin sic acous tic con trast be tween nor mal and ab lated liver tis sue. Our goal is to pro vide real-time ultrasonic tracking of temperature changes over the large range of temperatures traditionally used ( C) in rf ablation procedures using an external ultrasound transducer. Temperature estimates are ob tained us ing a cross-cor re la tion al go rithm ap plied to rf ul tra sound echo sig nal data ac quired at dis crete intervals during heating. Apparent tissue displacement estimates obtained at these discrete time-intervals are accumulated to obtain a cumulative displacement map, whose gradient provides after appropri - ate scal ing pro vides a tem per a ture map at the spec i fied elapsed ab la tion du ra tion. Tem per a ture maps are used to dis play the ini tial tem per a ture rise and to con tin u ously up date a ther mal map of the treated re - gion. In this pa per, we de velop cal i bra tion curves that re late the echo shift due to the change in the speed of sound and thermal expansion to the corresponding temperature increase on in-vi tro tis sue spec i mens. These cal i bra tion curves can then be uti lized for the real time cal i bra tion and anal y sis of tem per a ture es - ti mates ob tained from the rf echo sig nals dur ing ab la tion. Tem per a ture maps ob tained us ing the cal i bra - tion curve compare favorably to temperature estimates observed using the invasive thermosensor read ings on the ab la tion elec trode and pre vi ous re sults that uti lized a lin ear cal i bra tion fac tor. Key words: Ablation; elastography,;elasticity; imaging; radiofrequency ablation; strain; speed of sound; thermal strain; thermal expansion; temperature imaging; thermal imaging; ultrasound. IN TRO DUC TION Min i mally-in va sive ther a pies have gained in creas ing at ten tion in the last de cade as an al - ter na tive to stan dard sur gi cal ther apy. They are be ing in ves ti gated for the treat ment of many dis ease pro cesses, such as pri mary hepatocellular car ci noma (HCC), lower uri nary tract symp toms due to be nign pros tatic hy per pla sia (BPH) and small re nal-cell car ci noma (RCC). 1-7 Ther mal ther a pies such as radio fre quen cy (rf), 2, 8-13 microwave, high-in ten sity 7, fo cused ul tra sound (HIFU) or la ser en er gies 20, 21 are some of the more com monly-used modalities. Temperatures greater than 42 0 C are con sid ered po ten tially le thal, de pend ing on the du ra tion of ap pli ca tion, 22 and tem per a tures greater than 60 0 C are as so ci ated with uni form tis sue ne cro sis. 23 Ben e fits over sur gi cal re sec tion in clude the an tic i pated re duc tion in mor - bid ity and mor tal ity, low cost, suit abil ity for real time im age guid ance, abil ity to per form ab /04/5 $18.00 Copy right 2004/5 by Dynamedia, Inc. All rights of re pro duc tion in any form re served.

2 186 VARGHESE AND DANIELS la tive pro ce dures on out pa tients and the po ten tial ap pli ca bil ity to a wider spec trum of pa tients, in clud ing nonsurgical can di dates. The trend to ward min i mally-in va sive op tions in the man age ment of re nal tu mors has also prompted in ter est in ab la tion tech niques as pos si ble al ter na tives to rad i cal or par tial nephrectomy. 3, 24 The role of nephron-spar ing, tu mor-ab la tive sur gery has ex panded be yond the tra di tional cir cum stances of neoplasms in sol i tary/func tion ally-sol i tary kid neys or tu - mors present bilaterally. 3, 4 Lap aro scopic re nal cryoablation, rf, mi cro wave and HIFU ab la - 3, 24 tive ther a pies are some of the mo dal i ties used for re nal tu mors. Rf ab la tion em ploys nee dle elec trodes placed percutaneously and di rectly into re nal le sions while mi cro wave thermotherapy uses small ap pli ca tors to de liver mi cro wave en ergy to tis sues. The high risk of sur gi cal re sec tion of liver tu mors cou pled with the low num ber of re sec - tion candidates has increased the interest in minimally-invasive treatment, 2, 8-13 for both pri - mary and sec ond ary liver tu mors. Pri mary liver neoplasms and metastases in the liver rep re sent sig nif i cant sources of mor bid ity and mor tal ity in the United States. HCC is one of the most com mon solid or gan ma lig nan cies world wide with an an nual in ci dence of at least one mil lion new pa tients. 25 Ap prox i mately 150,000 new cases of co lon can cer are di ag nosed each year in the United States, of which ap prox i mately 50% will have re cur rences within the first 5 years. Of these re cur rences, 20% are pre dom i nantly lo cal ized to the liver. 26 Op ti mal ther apy of both pri mary and met a static liver dis ease is com plete re sec tion with neg a tive mar gins. If un treated, the sur vival rates of both these dis eases are dis mal. Un for tu nately, only 10-20% of pa tients with ei ther pri mary or met a static liver dis ease are ame na ble to sur gi cal ther apy. 26 Rf ab la tion is an in ter sti tial fo cal ab la tive ther apy in which an elec trode is placed into a le - sion to cause tis sue heat ing from ionic ag i ta tion, re sult ing in heat ing and cau ter i za tion of the tu mor mass and elec trode track. Rf elec trodes are of small di am e ter (gen er ally gauge) and can be used in a percutaneous fash ion, with min i mal risk of bleed ing. 13,27 Rf ab la tion have the ad van tage of tis sue co ag u la tion at the elec trode in ser tion site (re sult ing in a low rate of intra- and postprocedural bleed ing) and percutaneous ap pli ca tion via a rel a tively small electrode. 28 Mi cro wave ab la tion has many ad van tages over rf ab la tion and has been used to vary ing de grees in Asia En thu si asm for its use in the United States has been low due to the large elec trode size (14 gauge), small zone of ne cro sis (1.6 cm in di am e ter) 29 and lack of FDA ap proval of the only com mer cially-avail able de vice (Microtaze, Nip pon Shoji, Osaka, Ja pan). Four teen-gauge elec trodes would not be com pat i ble with rou tine percutaneous use in the chest, and would only be used with cau tion in the ab do men. Un like rf ab la tion, boil ing and char ring of tis sue sur round ing the an tenna do not sub stan tially af fect the de liv ery of power into tis sue. Guid ance and mon i tor ing for rf ab la tion in clin i cal se ries for hepatic tu mors have been done al most ex clu sively via con ven tional transabdominal ul tra sound. 13 Ad van tages of ul tra - sound in clude wide spread avail abil ity, real-time guid ance for elec trode place ment and ac cu - rate, con ve nient punc ture guides. How ever a sig nif i cant draw back of cur rent ul tra sound meth ods is the zone of ne cro sis pro duced dur ing rf or mi cro wave ab la tion is not eas ily vi su - al ized by transabdominal sonography be cause of the low in trin sic con trast be tween nor mal and ab lated liver and ar ti facts from gas bub ble for ma tion and the rf gen er a tor. These con di - tions re sult in un cer tainty in de ter min ing the ex tent of ab la tion dur ing an rf pro ce dure and may be a con trib ut ing fac tor to the high pos i tive re cur rence rates seen in clin i cal se ries. Lo - cal re cur rence rates as high as 34-55% have been re ported for percutaneous rf ab la tion guided by con ven tional sonography. 12, 13 This is a much higher rate than the 12% lo cal re cur - rence rate re ported for cryosurgical pa tients fol lowed over the last 5 years 30 or the 16.7% lo - cal recurrence rate reported for surgical resection. 31 The poor vi su al iza tion of the ab lated re gion has prompted cli ni cians to seek other meth ods, such as x-ray CT, to guide the ab la tion pro ce dure. 32

3 REAL-TIME CAL I BRA TION OF TEM PER A TURE ES TI MATES DUR ING RF AB LA TION 187 Noninvasive and real-time de ter mi na tion of the tem per a ture dis tri bu tion in tis sues un der - go ing ther mal ab la tion could play a sig nif i cant role in suc cess ful de liv ery of these treat - 18, 33 ments. MRI has been used for mon i tor ing the tem per a ture pro file gen er ated by HIFU. Other methods include impedance tomography, 34 microwave radiometry 35 and ul tra sound Sev eral ul tra sonic meth ods re lated to var i ous tem per a ture-in duced changes in tis sue have been pro posed to es ti mate tem per a ture. Ul tra sound pa ram e ters in clude at ten u a tion, 41 back - scattered power, 42 and speed of sound. 43 Ther mal ex pan sion with out 37 or with sound speed variations have also been mon i tored. Dif fer ent al go rithms have been used in the es ti ma tion of the tem per a ture change. Seip and Ebbini 37 es ti mated tem per a ture changes in 1-D us ing an autoregressive model, track ing fre - quency vari a tions of the echo com po nents in the spec tral do main and ex am in ing changes in the mean scat terer spac ing. Sev eral tech niques are based on track ing the echo shift in the time do main and dif fer en ti at ing the time-shift es ti mates along the ax ial di rec tion to ob tain a temperature profile Al though prom is ing, none of the ul tra sonic meth ods pro posed above 36-38, 40 has been able to mea sure tem per a ture changes larger than a few de grees, with Si - mon et al es ti mat ing the larg est tem per a ture rise from 23 to 43 0 C over 208 min utes. 36 We have pre vi ously dem on strated the abil ity con tin u ously track and pro vide tem per a ture im ages over the en tire range of tem per a tures uti lized for rf ab la tion. 39 How ever, the method de scribed by Varghese et al 39 does not pro vide real-time im ag ing of tem per a ture changes. For practical application of ultrasonic temperature imaging for thermal ablative therapies, the abil ity to pro vide real-time track ing of 2-D tem per a ture changes over the large range of tem per a tures tra di tion ally used in ab la tion ther apy ( C, lower tem per a tures are not con sid ered as they are be low the hu man body s in ter nal tem per a ture of 37 C) and over the en tire du ra tion of the ab la tion pro ce dure is es sen tial. In this pa per, we il lus trate the use of cal i bra tion curves (ob tained us ing a pri ori ex per i ments) for spe cific tis sue types that would enable the real-time calibration of ultrasonic temperature images. Our technique coupled with the de vel op ment of faster tech nol ogy would en able the gen er a tion of real-time tem per - a ture im ages to mon i tor ab la tive ther a pies. CAL I BRA TION CURVE GEN ER A TION Echo sig nal shifts dur ing ab la tive ther apy arise from both speed of sound (SOS) and tis sue ex pan sion changes due to the tem per a ture in crease. There fore, both of these ef fects have to be in cluded in the gen er a tion of a cal i bra tion curve that would re late the acous tic echo shift changes with temperature. Fig ure 1 il lus trates the rel a tive vari a tion in tis sue ex pan sion (spec i men thick ness changes) and SOS changes ver sus tem per a ture ob tained from data ac quired on liver tis sue in our lab o - ra tory. 44 The SOS and tis sue thick ness of five in de pend ent liver tis sue spec i mens were ini - tially mea sured in a wa ter bath main tained at a con stant tem per a ture of 37 C. Each of these five spec i mens were then cooked in a dif fer ent wa ter bath for a du ra tion of 20 min utes af ter the wa ter bath had been raised to tem per a tures of 50, 60, 70, 80 and 90 C, re spec tively, the cor re spond ing tem per a tures in fig ure 1. The spec i mens were then placed in the wa ter bath at 37 C and both the SOS and tis sue thick ness were mea sured quickly us ing both trans mit ted and back scat tered ul tra sound sig nals. 44 The ex per i ment was re peated 10 times us ing liver tis sue from in de pend ent an i mals to ob tain the sta tis ti cal mean and stan dard de vi a tion both the SOS and tis sue ex pan sion with re spect to the mea sure ment ac quired at 37 C. Note that the vari a tions in the SOS are less than 2% of the SOS at room tem per a ture over the en tire range of ther mal ab la tion tem per a tures. Fig ure 1 also shows the sig nif i cant con tri -

4 188 VARGHESE AND DANIELS Percent Change (Ref at 37 C) Sound Speed Thickness Temperature ( C) FIG. 1 Plots com par ing the per cent age change in spec i men thick ness and SOS with tem per a ture. bu tion of tis sue ex pan sion to changes in echo shift in duced by tem per a ture el e va tion. Tis sue thick ness var ies in a lin ear fash ion with tem per a ture over C range, as shown in figure 1, with the per cent thick ness change re duc ing for tem per a ture greater than 80 C. The SOS changes on the other hand are sig nif i cant only at the lower tem per a tures less than 60 C. How ever, SOS vari a tions at higher tem per a ture would in tro duce un cer tain ties on the or der of about 2% in the tem per a ture mea sure ment due to its de crease at higher tem per a tures. The per cent age change in the SOS there fore is in sig nif i cant when com pared to the per cent age changes due to ther mal ex pan sion in the tem per a ture range from C. The cal i bra tion curve de vel oped in this pa per com bines both these ef fects to ob tain a re la tion ship be tween the acous tic echo shift with tem per a ture in crease. The speed of sound vari a tions for four tis sue types over 10 in de pend ent an i mals mea sured by Techavipoo et al 45 are il lus trated in fig ure 2. These curves were as sem bled by de ter min - ing shifts in pulsed ul tra sound travel time through a spec i men of the tis sue type af ter the tis - sue is heated to a pre cise tem per a ture level. 45 The SOS vari a tion was es ti mated us ing a nar row band sub sti tu tion tech nique. 45 The trans mit ting trans ducer (Panametrics V309, Waltham, MA) is 13 mm in di am e ter with a 5 MHz cen ter fre quency and driven by 30 Hz du - ra tion narrowband pulses us ing a func tion gen er a tor (Wavetek model 81, San Diego, CA) and a power am pli fier (Am pli fier Re search model 75A250, Souderton, PA). The re ceiv ing trans ducer (Aerotech Delta, Krautkramer Inc., Lewistown, PA) is 8 mm in di am e ter and has a 5 MHz cen ter fre quency. The re ceived sig nals were re corded and av er aged over 50 suc ces - sive traces by a dig i tal os cil lo scope (LeCroy 9410, Chest nut Ridge, NY), trig gered by the func tion gen er a tor. In ad di tion to the echo-shift in for ma tion, the thick ness of the tis sue spec - i men heated was also mea sured care fully and uti lized to com pute the SOS val ues. The er ror bars for only the liver spec i mens are shown in fig ure 2, with the data from other tis sue hav ing er ror bars in a sim i lar range.

5 REAL-TIME CAL I BRA TION OF TEM PER A TURE ES TI MATES DUR ING RF AB LA TION Liver Muscle Kidney Prostate FIG. 2 Prop a ga tion speeds vs. tem per a ture mea sured at 5 MHz. Re sults are shown for ca nine liver, mus cle, kid - ney and prostate tissue specimens, respectively. The cal i bra tion curves uti lized in this pa per are com puted for each of the four tis sue types shown in fig ure 2, uti liz ing the same narrowband sub sti tu tion data used to com pute the SOS vari a tions in fig ure 2. The cal i bra tion curves shown in fig ure 3, on the other hand, uti lize only the raw echo-shift in for ma tion ob tained with and with out the sam ple pres ent in the acous tic path. This raw echo shift in cor po rates the ef fect of both SOS and tis sue ex pan sion changes with tem per a ture. In other words, the ad di tional echo shift due to the ther mal ex - pan sion of tis sue is al ready in cluded. The cal i bra tion curves are then ob tained by ac cu mu lat - ing time-de lay in for ma tion on tis sue spec i mens in a wa ter tank at the dif fer ent tem per a tures used in the mea sure ment. The ac cu mu lated or to tal echo shift is then nor mal ized by the spec - i men thick ness at mea sured at room tem per a ture (not at the ab lated tem per a ture) to ob tain the nor mal ized cal i bra tion curves (the nor mal ized cal i bra tion value, i.e., the to tal echo shift nor mal ized by spec i men thick ness is what is rep re sented on the y-axis) in fig ure 3. Note the ap prox i mately lin ear re la tion ship of the nor mal ized cal i bra tion value for all the tis sue spec i - mens with tem per a ture. This re sult also jus ti fies the as sump tion of a lin ear in crease in the dis place ment due to com bined vari a tions in the sound speed and tis sue ex pan sion with tem - per a ture, used pre vi ously. 39 The cal i bra tion curves il lus trated in fig ure 3 show the po ten tial of uti liz ing these curves for possible real-time calibration of temperature estimates. The algorithm used for temperature es ti ma tion in Varghese et al 39 can be uti lized since this method also re lies on the es ti ma tion of the time shifts or time de lay be tween a base line echo sig nal and one af ter a tem per a ture el e - va tion. The tem per a ture es ti ma tion al go rithm is de scribed be low. Temperature estimation algorithm A base line frame of rf echo sig nals was ac quired be fore heat ing; sub se quent frames were ac quired ev ery half-sec ond dur ing heat ing. Echo shifts, or ax ial dis place ments, be tween

6 190 VARGHESE AND DANIELS Liver Muscle Kidney Prostate FIG. 3 Cal i bra tion curves ob tained from the ac cu mu lated time shifts from the tis sue spec i mens used in the speed of sound ex per i ment in fig ure 2. suc ces sive frames were com puted by ap ply ing a nor mal ized crosscor re la tion anal y sis to cor - re spond ing beam line pairs in the rf echo-sig nal data sets. The dis place ment es ti mates were then accumulated to obtain a cumulative displacement estimate over the elapsed ablation du - ration. The cumulative displacement estimates were then differentiated in the axial direction and scaled to ob tain a tem per a ture im age over the elapsed ab la tion du ra tion. Tem per a ture es ti mates are ob tained by us ing the equa tion, T( z ) c ( 0 2 ( z ) )( ) z where ( z) z is the gra di ent of the cu mu la tive time shift or dis place ment in the win dowed echo sig nal. The term c 0 /2 rep re sents the scal ing fac tor, where is a con stant that takes into con sid er ation ther mal ex pan sion and the change of the SOS with tem per a ture and c 0 is the ini tial SOS in the me dium. Prior knowl edge of the value of this scal ing fac tor would en able real-time calibration of the temperature estimate. However, exact computation of this scaling fac tor is dif fi cult and re search ers have at tempted to ap prox i mate the con tri bu tion of this term. Pre vi ous re sults re ported by Varghese et al 39 uti lized a con stant scal ing fac tor ob tained using the cumulative displacement estimate obtained at a temperature of 100 C and then uti - lizing this scaling factor to obtain temperature estimates at different ablation durations. In this pa per, we uti lize the a pri ori data ob tained on SOS vari a tions and tis sue ex pan sion to de - rive a cal i bra tion curve for the tis sue be ing im aged to ob tain a real-time cal i bra tion or scal ing of the tem per a ture es ti mated. Es ti ma tion of the large range of tem per a tures uti lized for ab la -

7 REAL-TIME CAL I BRA TION OF TEM PER A TURE ES TI MATES DUR ING RF AB LA TION 191 tion ther a pies are ob tained by ob tain ing echo sig nal data over 0.5 s in cre ments fol lowed by the ac cu mu la tion of the time-shift sim i lar to that used to com pute the cal i bra tion curve in fig - ure 3. In the fol low ing sec tions, we will com pare tem per a ture maps ob tained uti liz ing the pre vi ously-used lin ear as sump tion 39 and the cal i bra tion curves de vel oped in this pa per. MA TE RI ALS AND METHOD Rf ablation in vi tro was performed on lobes of liver tis sue with ap prox i mate di men sions of 40 mm by 40 mm with a thick ness of 20 mm that were en cased in a gel a tin cube with di men - sions of 80 mm. Ul tra sound im ag ing to guide the rf elec trode place ment and mon i tor the tem per a ture pro file in tis sue was performed us ing an Acuson 128XP (Acuson Inc., USA) real-time scan ner us ing a 5 MHz lin ear ar ray trans ducer (40 mm). Ul tra sound rf echo sig nals are dig i tized us ing a 12-bit data ac qui si tion board (Gage Inc., Mon treal, Can ada) with a sam - pling rate of 50 MHz. A per sonal com puter con trols the op er a tion of the sys tem. A RITA model 30 electrosurgical de vice (RITA Med i cal Sys tems Inc., Moun tain View, CA) was used for in-vitro rf ab la tion pro ce dures. The rf elec trode con sists of a 15 gauge shaft through which four sharp tines, each 0.5 mm in di am e ter (25 gauge) and 3-5 cm long, can be de ployed. Fully ex tended, the tines are in an um brella con fig u ra tion, with tines at 90 in ter vals. The last 1 cm of the elec trode tip and each tine con sti tute the elec tri cally-ac tive sur face. The rf elec trode was in serted into the liver tis sue through the gel a tin phan tom at a depth of 3 cm, per pen dic u lar to the ul tra sound im age plane, and the tines de ployed within the liver. Rf ab la tion of the tar get tis sue in vi tro was per formed for 3 min utes by rais ing the tem - per a ture to 100 C at a 50 W power level. Pro cess ing for the pre lim i nary data was per formed off-line. The man u fac turer sug gested guide line for cre at ing a 3 cm ther mal le sion or re gion of ne cro sis is 10 min ute ab la tion du ra tion at a tar get tem per a ture of 100 C with the de fault con trol mode that uti lizes the av er age of the tem per a ture es ti mates mea sured at the tines for feed back con trol of the rf gen er a tor. Rf ablation in-vivo was per formed on the mus cle tis sue of a fe male pig ( 50 kg). A RITA model 1500, sim i lar to the model 30 de scribed above, was used for in-vivo rf pro ce dures. A dispersive ground ing pad was ap plied to the an i mal s flank pre op er a tively af ter shav ing body hair. Open laparotomy was per formed through a bi lat eral subcostal in ci sion. The rf elec trode was placed into the liver pa ren chyma and four ab la tions were per formed at dif fer - ent lo ca tions. Rf ab la tion was also per formed on mus cle tis sue in the an i mal thigh. The in-vivo pro ce dures were per formed sub ject to a pro to col ap proved by the Uni ver sity of Wis - consin-research Animal Resources Center. Tem per a ture was mon i tored us ing in va sive thermosensors on the tines of the rf elec trode. For each le sion, the ul tra sound trans ducer was fixed above the treated re gion and frames of rf echo data were ac quired for off-line tem per a ture im ag ing. Rf ab la tion was per formed us ing the man u fac turer sug gested guide lines of heat ing for 10 min utes af ter at tain ing the tar get temperature (100 C). The tem per a ture im ag ing was re peated at two lo ca tions in the an i mal thigh mus cle tis sue. RESULTS For the in-vitro ex per i ment on liver tis sue, tem per a ture im ages ob tained by cal cu lat ing the gradient of the cumulative displacement estimates over the respective elapsed ablation durations are pre sented in fig ure 4. The al go rithm uti lized to es ti mate the tem per a ture is de - scribed in the pre vi ous sec tion. The ul tra sound-based tem per a ture im ages shown in fig ures 4 and 5 were ob tained us ing a lin ear scal ing fac tor as de scribed by Varghese et al. 39 Ob serve the pro gres sive in crease in the tem per a ture dis tri bu tion from fig ures 4 (a) to (d), as shown by

8 192 VARGHESE AND DANIELS the color-bar val ues. We found from our ex per i ments that a tem per a ture change of 1 C in liver tis sue re sulted in a sig nal change that is equiv a lent to a strain of 2%. Note that we reach a tem per a ture of 100 C in the tem per a ture map af ter 1 min ute (Fig. 4 (c)), val i dated by in va - sive thermosensor read ings. The tem per a ture is main tained at around 100 C, ver i fied in the tem per a ture map ob tained af ter a du ra tion of 2 min utes (Fig. 4 (d)). The hotspots in the tem - per a ture im age cor re spond to re gions around the tines of the rf elec trode. One of the prob - lems with rf ab la tion is the cre ation of hot spots and cor re spond ing cold spots in tis sue dur ing the ab la tion pro ce dure. 46 Cold spots may also oc cur around re gions sur round ing large blood ves sels (due to the cool ing ef fect due to blood flow) lead ing to pos si ble in com plete ab la tion. The ab la tive ther apy re lies on con vec tive heat trans fer to ab late re gions around the cold spots. For the in-vivo ex per i ment, the tem per a ture im ages are pre sented in fig ure 5. Sim i lar to the in-vitro sit u a tion in fig ure 4, we ap pear to track the tem per a ture in crease with time. Plots of the tem per a ture val ues mea sured invasively at the tines of the rf elec trode are pre sented in fig ure 6, with the av er age tem per a ture value shown as the thick solid line. Note the close cor - respondence between the maximum temperature observed in the temperature maps at differ - ent time in stances in fig ure 5 and the in va sive tem per a ture mea sure ments in fig ure 6. For example at T= 30 s, the in va sive tem per a ture is around 60 0 C, sim i lar to the max i mum value on the color bar in fig ure 5(b). The tar get tem per a ture of C is at tained af ter 2.4 min utes of heat ing, with the tem per a ture be ing main tained con stant from this point for ward. This re - sult is closely mir rored in the tem per a ture on the color bar in fig ure 5. For re sults pre sented in fig ures 4 and 5, cal i bra tion of the tem per a ture im ages was per - formed by com put ing a sin gle scal ing fac tor as pre vi ously de scribed by Varghese et al. 39 Dis place ment es ti mates ac cu mu lated over the du ra tion from the start of the ab la tion pro ce - dure to the time when the in va sive tem per a ture on the thermosensors was 100 C was used to com pute the scal ing fac tor. This scal ing fac tor was used to scale tem per a ture es ti mates (gra - di ent of ac cu mu lated dis place ment) at dif fer ent times shown in figures 4 and 5. Cal i bra tion in this man ner as sumes a lin ear re la tion ship be tween the com bined vari a tions in SOS and tis - sue ex pan sion with tem per a ture. The re sults in fig ures 4 and 5 il lus trate the fea si bil ity for tem per a ture mon i tor ing. How - ever, the tem per a ture maps dis played in fig ures 4 and 5 must be pro duced in real-time. Real-time tem per a ture im ag ing re quires faster com pu ta tion and up dat ing of the tem per a ture im ages shown in fig ures 4 and 5, which is fa cil i tated by im prove ments in com puter tech nol - ogy and faster dig i tal sig nal pro cess ing. It also re quires the ap pro pri ate scal ing or cal i bra tion of the tem per a ture im ages to ob tain a mea sure of tem per a ture in crease or the tem per a ture dis tri bu tion in tis sue. In va sive thermosensors pro vide ac cu rate tem per a ture es ti mates at spe cific lo ca tions; how ever, they are un able to pro vide a tem per a ture dis tri bu tion. On the other hand, tem per a ture im ag ing with ul tra sound is able to pro vide a dis tri bu tion of the tem - per a ture or heat ing pat terns. There fore, an abil ity to quan tify the tem per a ture dis tri bu tion with ul tra sound may have a far-reach ing im pact on the prac tice of ab la tive ther a pies. In this pa per, we pro pose the uti li za tion of the cal i bra tion curves ob tained from prior ex per i men ta - tion in the real-time cal i bra tion of ul tra sound-based tem per a ture im ages. The cal i bra tion curves shown in fig ure 3 pro vide a re la tion ship be tween the tem per a ture in crease and a fac tor that would pro vide the cor rect tem per a ture in crease when used as a scal ing fac tor on the dis place ment es ti mates. Note that, the ac cu racy and the er ror in the tem - per a ture es ti mates due to use of the cal i bra tion curves have to be stud ied. The un cer tain ties in the tem per a ture es ti mated with ul tra sound may be large. Nev er the less, the use of the cal i - bra tion curve pro vides the abil ity to ob tain a ther mal dose dis tri bu tion in tis sue. The use of the cal i bra tion curves for the same data set used in fig ures 4 and 5 is dis played in fig ures 7 and 8.

9 REAL-TIME CAL I BRA TION OF TEM PER A TURE ES TI MATES DUR ING RF AB LA TION 193 (a) (b) (c) FIG. 4 Images of the temperature distribution in ablated liver tissue. Temperature monitoring is performed by acquiring an rf echo signal frame during half-second intervals. (a) Denotes temperature distribution after 10 seconds of ab la tion, (b) af ter 20 sec onds, (c) af ter 60 sec onds (1 min ute) and (d) af ter 120 sec onds (2 min utes). Ob serve the in crease in max i mum tem per a ture value from (a) to (d) on the color bar. The tem per a ture val ues reach the max i mum set temperature of C after 1 minute of heating (corroborated by thermocouple readings). The temperature is held con stant at 2 min utes as ob served by com par ing the color bar. (d)

10 194 VARGHESE AND DANIELS (a) (b) (c) FIG. 5 In-vivo images of the temperature distribution in ablated muscle tissue. (a) Denotes temperature distribu - tion ob tained af ter 10 sec onds, (b) af ter 30 sec onds, (c) af ter 150 sec onds (2.5 min utes) and (d) af ter 300 sec onds (5 min utes). Ob serve the in crease in the tem per a ture from (a) to (d) by val ues on the color bar start ing at an ini tial tem - perature of 45 0 C. The temperature distribution reaches the maximum set temperature of C af ter 2.4 min utes of heat ing (see thermosensor read ings in fig ure 14). The tem per a ture re mains con stant at around C at 5 min utes as seen from the color bar in (d). (d)

11 REAL-TIME CAL I BRA TION OF TEM PER A TURE ES TI MATES DUR ING RF AB LA TION FIG. 6 In va sive es ti mates of the tem per a ture mea sured at the tines of the rf elec trode. For the ab la tion pro ce dures shown in fig ures 7 and 8, the cal i bra tion curve was ap plied to the ac cu mu lated changes in the echo ar rival time ac com pa ny ing sub se quent rf heat ing. The av er age dis tance of sev eral echo shift max ima from the trans ducer was uti lized to nor mal ize the ac cu mu lated echo shifts due to tem per a ture in creases. This nor mal ized value is then com pared to the cor re spond ing value on the cal i bra tion curve (Fig. 3) and the tem per a ture is read off the cal i bra tion curve. This tem per a ture value is then used to ap pro pri ately scale the temperature estimates in the individual accumulated thermal images. The re sults shown in fig ures 7 and 8 il lus trate that the cal i bra tion curve pro vides com pa ra - ble re sults to those ob served on the in va sive thermosensors on the tines of the rf elec trode. The cal i bra tion curves for mus cle tis sue (Fig. 3) were used to an a lyze the in-vivo experiment data to pro duce real time tem per a ture maps. Figure 8 again dem on strates that by us ing the cal i bra tion curve the tem per a ture maps can be cre ated in real-time. These maps com pare fa - vor ably to those shown in fig ure 5. Fig ures 7 and 8 thus jus tify the as sump tion of a lin ear in - crease in the dis place ment due to com bined vari a tions in the SOS and tis sue ex pan sion with temperature that the calibration curve predicts. DIS CUS SION AND CON CLU SIONS One of the ma jor lim i ta tions of tem per a ture im ag ing us ing ul tra sonic meth ods is the in - abil ity to ob tain real-time con ver sion of the ac cu mu lated time-shifts into ac tual tem per a ture changes. We pro pose a new method that would en able the real-time cal i bra tion of ther mal images. In the ac tual ab la tion pro ce dures shown in fig ures 7 and 8, the cal i bra tion curve was ap plied to ac cu mu lated changes in the echo ar rival time ac com pa ny ing sub se quent rf heat - ing. The use of pre vi ously-cal cu lated cal i bra tion curves pro vides the po ten tial of ob tain ing real-time ul tra sound-based tem per a ture im ages. Al though we still have not eval u ated the

12 196 VARGHESE AND DANIELS (a) (b) (c) FIG. 7 Im ages of the tem per a ture dis tri bu tion in ab lated liver tis sue us ing the cal i bra tion curve. The re sults shown in this fig ure use the data gen er ated for fig ures 4. (d) accuracy and precision of the temperature estimates, the results indicate the feasibility of ob - tain ing rea son able tem per a ture read ings on the ther mal im age.

13 REAL-TIME CAL I BRA TION OF TEM PER A TURE ES TI MATES DUR ING RF AB LA TION 197 (a) (b) (c) FIG. 8 In-vivo im ages of the tem per a ture dis tri bu tion in ab lated mus cle tis sue us ing the cal i bra tion curve and the same data used to gen er ate fig ures 5. (d) Fig ures 7 and 8 dem on strate that the cal i brated tem per a ture im ages can be pro duced us ing calibration curves that are determined a pri ori and al low for real-time cal i bra tion of the tem -

14 198 VARGHESE AND DANIELS per a ture val ues es ti mated. The cal i bra tion curves (Fig. 3) show that the pri mary ef fect of tem per a ture in crease is an in crease in tis sue thick ness (ex pan sion) over C) and SOS vari a tions are sig nif i cantly smaller that tis sue ex pan sion in the tem per a ture range C. The re sults pre sented in this pa per also jus tify the as sump tion of a lin ear in crease in the dis - place ment due to com bined vari a tions in the sound speed and tis sue ex pan sion with tem per - a ture, that was used pre vi ously. 39 Knowl edge of the slope of these curves for a par tic u lar tis sue type may be suf fi cient to cal i brate the ac cu mu lated time shifts used for ther mal im ag - ing, as shown in fig ures 7 and 8 un der both in-vitro and in-vivo con di tions, re spec tively (us- ing same data shown in fig ures 4 and 5). Tis sue thick ness var ies in a lin ear fash ion with tem per a ture over the C temperature range and con trib utes to the lin ear ity of the cal i bra tion curve as shown in fig ure 1. The SOS changes are sig nif i cant only at lower tem per a tures and are in cor po rated into the cal i bra tion curve. The re sults shown in fig ure 1demonstrates that ac cu rate es ti ma tion of the tem per a ture us ing ul tra sonic tech niques is there fore pos si ble since the pri mary con tri bu tion to the tem - per a ture es ti ma tion is the lin ear ther mal ex pan sion as op posed to the SOS changes. How - ever, it has to be noted that SOS vari a tions at higher tem per a ture would in tro duce un cer tain ties < 2% in the tem per a ture mea sure ment due to its de crease at higher tem per a tures. The ex per i men tal data shown in fig ure 3 have to be im proved by in clud ing a larger num - ber of sam ples, pro duc ing cal i bra tion curves with smaller er ror bars for more ac cu rate tem - per a ture es ti ma tion. In ad di tion, the im pact of tis sue com po si tion on the slope of the cal i bra tion curve for the dif fer ent tis sue types has to be eval u ated. 47 For ex am ple, dif fer ent lev els of fatty in fil tra tion of the liver tis sue may change the slope of the cal i bra tion curve shown in fig ure 3. An other pos si bil ity would be the de vel op ment of these cal i bra tion curves for tis sue con stit u ents such as pro teins, wa ter, fat, etc., which may en able the gen er a tion of cal i bra tion curves pred i cated on an a pri ori knowl edge of the tis sue com po si tion, per haps ob tained us ing core bi op sies. AC KNOWL EDGE MENTS This work was sup ported in part by the Wis con sin Alumni Re search Foun da tion (WArf) and start-up grant funds from the De part ment of Med i cal Phys ics, Med i cal School and Grad - u ate School at the Uni ver sity of Wis con sin-mad i son. The au thors wish to thank Udomchai Techavipoo, Ph.D, for his help in col lect ing some of the data that was used in this pa per. REFERENCES 1. Beerlage HP, Thuroff S, Madersbacher S, et al. Cur rent sta tus of min i mally in va sive treat ment op tions for lo - calized prostate carcinoma, Eur Urol 37, 2-13 (2000). 2. Tunuguntla HS, Ev ans CP. Min i mally in va sive ther a pies for be nign pros tatic hy per pla sia, World J Urol 20, (2002). 3. Murphy DP, Gill IS. En ergy-based re nal tu mor ab la tion: a re view, Semin Urol Oncol 19, (2001). 4. Chin JL, Pautler SE. New tech nol o gies for ab la tion of small re nal tu mors: cur rent sta tus, Can J Urol 9, (2002). 5. Mir za AN, Fornage BD, Sneige N, et al. Radio fre quen cy ab la tion of solid tu mors, Cancer J 7, (2001). 6. Goldberg SN, Ga zelle GS, Mueller PR. Ther mal ab la tion ther apy for fo cal ma lig nancy: a uni fied ap proach to underlying principles, techniques, and diagnostic imaging guidance, AJR 174, (2000). 7. Hill CR, ter Haar GR. High in ten sity fo cused ul tra sound po ten tial for can cer treat ment, Br J Radiol 68, (1995).

15 REAL-TIME CAL I BRA TION OF TEM PER A TURE ES TI MATES DUR ING RF AB LA TION Chopra S, Dodd GD, 3rd, Chintapalli KN, et al. Tu mor re cur rence af ter radio fre quen cy ther mal ab la tion of hepatic tu mors: spec trum of find ings on dual-phase con trast-en hanced CT, AJR 177, (2001). 9. Gervais DA, McGovern FJ, Wood BJ, et al. Ra dio-fre quency ab la tion of re nal cell car ci noma: early clin i cal experience, Radiology 217, (2000). 10. Goldberg SN. Radio fre quen cy tu mor ab la tion: prin ci ples and tech niques, Eur J Ul tra sound 13, (2001). 11. Haemmerich D, Staelin ST, Tungjitkusolmun S, et al. Hepatic bi po lar ra dio-fre quency ab la tion be tween sep - a rated multiprong elec trodes, IEEE Trans Biomed Eng 48, (2001). 12. Rossi S, Di Stasi M, Buscarini E, et al. Percutaneous RF in ter sti tial ther mal ab la tion in the treat ment of hepatic can cer, AJR 167, (1996). 13. Solbiati L, Ierace T, Goldberg SN, et al. Percutaneous US-guided ra dio-fre quency tis sue ab la tion of liver metastases: treat ment and fol low-up in 16 pa tients, Radiology 202, (1997). 14. Lu MD, Chen JW, Xie XY, et al., Hepatocellular car ci noma: US-guided percutaneous mi cro wave co ag u la - tion ther apy, Radiology 221, (2001). 15. Hyodoh H, Hyodoh K, Takahashi K, et al. Mi cro wave co ag u la tion ther apy on hepatomas: CT and MR ap - pearance after therapy, J Magn Reson Im ag ing 8, (1998). 16. Shibata T, Iimuro Y, Yamamoto Y, et al. Small hepatocellular car ci noma: com par i son of ra dio-fre quency ab - lation and percutaneous microwave coagulation therapy, Radiology 223, (2002). 17. Chapelon JY, Ribault M, Ver nier F, et al. Treat ment of local ised pros tate can cer with transrectal high in ten - sity focused ultrasound, Eur J Ul tra sound 9, (1999). 18. McDannold N, King RL, Jolesz FA, Hynynen K. The use of quan ti ta tive tem per a ture im ages to pre dict the op ti mal power for fo cused ul tra sound sur gery: in vivo ver i fi ca tion in rab bit mus cle and brain, Med Phys 29, (2002). 19. Sanghvi NT, Fos ter RS, Bihrle R, et al. Noninvasive sur gery of pros tate tis sue by high in ten sity fo cused ul tra - sound: an up dated re port, Eur J Ul tra sound 9, (1999). 20. Heisterkamp J, van Hillegersberg R, Ijzermans JN. In ter sti tial la ser co ag u la tion for hepatic tu mours, Br J Surg 86, (1999). 21. Malone DE, Lesiuk L, Brady AP, et al. Hepatic in ter sti tial la ser photocoagulation: dem on stra tion and pos si - ble clin i cal im por tance of intravascular gas, Radiology 193, (1994). 22. Rosner GL, Clegg ST, Prescott DM, Dewhirst MW. Es ti ma tion of cell sur vival in tu mours heated to non uni - form temperature distributions, Int J Hyperthermia 12, (1996). 23. Zervas NT, Kuwayama A. Pathological characteristics of experimental thermal lesions. Comparison of in - duction heating and radiofrequency electrocoagulation, J Neurosurg 37, (1972). 24. John son DB, Nakada SY. Cryosurgery and nee dle ab la tion of re nal le sions, J Endourol 15, (2001). 25. Bridbord K. Pathogenesis and pre ven tion of hepatocellular car ci noma, Cancer Detect Prev 14, (1989). 26. Steele GJ. Colorectal Can cer Met a static to the Liver (Mosby, St. Louis, 1995). 27. Curley SA, Davidson BS, Flem ing RY, et al. Laparoscopically guided bi po lar radio fre quen cy ab la tion of ar - eas of por cine liver, Surg Endosc 11, (1997). 28. Livraghi T, Goldberg SN, Lazzaroni S, et al. Small hepatocellular car ci noma: treat ment with ra dio-fre quency ablation versus ethanol injection, Radiology 210, (1999). 29. Seki T, Wakabayashi M, Nakagawa T, et al. Ul tra son i cally guided percutaneous mi cro wave co ag u la tion ther apy for small hepatocellular car ci noma, Cancer 74, (1994). 30. Kane R, McPhee DM, Kruskal J, et al. Five year sur vival in US-guided hepatic cryosurgery (ab stract), Radi - ology 205, 201 (1997). 31. Hughes K, Scheele J, Sugarbaker PH. Sur gery for colorectal can cer met a static to the liver. Op ti miz ing the re - sults of treat ment, Surg Clin North Am 69, (1989). 32. Cha CH, Lee FT, Jr., Gurney JM, et al. CT ver sus sonography for mon i tor ing radio fre quen cy ab la tion in a porcine liver, AJR 175, (2000). 33. Hynynen K, Chung A, Fjield T, et al. Fea si bil ity of us ing ul tra sound phased ar rays for MRI mon i tored noninvasive sur gery, IEEE Trans Ultrason Ferroel Freq Cont 43, (1996).

16 200 VARGHESE AND DANIELS 34. Paulsen KD, Moskowitz MJ, Ryan TP, et al. Ini tial in vivo ex pe ri ence with EIT as a ther mal es ti ma tor dur ing hyperthermia, Int J Hyperthermia 12, (1996). 35. Meaney PM, Paulsen KD, Hartov A, Crane RK. Mi cro wave im ag ing for tis sue as sess ment: ini tial eval u a tion in multitarget tissue-equivalent phantoms, IEEE Trans Biomed Eng 43, (1996). 36. Simon C, VanBaren P, Ebbini ES. Two-dimensional temperature estimation using diagnostic ultrasound, IEEE Trans Ultrason Ferroel Freq Cont 45, (1998). 37. Seip R, Ebbini ES. Noninvasive es ti ma tion of tis sue tem per a ture re sponse to heat ing fields us ing di ag nos tic ultrasound, IEEE Trans Biomed Eng 42, (1995). 38. Maass-Moreno R, Damianou CA, Sanghvi NT. Noninvasive tem per a ture es ti ma tion in tis sue via ul tra sound echo-shifts. Part II. In vi tro study, J Acoust Soc Am 100, (1996). 39. Varghese T, Zagzebski JA, Chen Q,. Ul tra sound mon i tor ing of tem per a ture change dur ing radio fre quen cy ablation: preliminary in-vivo results, Ultrasound Med Biol 28, (2002). 40. Sun Z, Ying H. A multi-gate time-of-flight tech nique for es ti ma tion of tem per a ture dis tri bu tion in heated tis - sue: theory and computer simulation, Ultrasonics 37, (1999). 41. Ueno S, Hashimoto M, Fukukita H, Yano T. Ul tra sound ther mom e try in hyperthermia, in Proc IEEE Ultrason Symp, pp (1990). 42. Straube WL, Arthur RM. Theoretical estimation of the temperature dependence of backscattered ultrasonic power for noninvasive ther mom e try, Ultrasound Med Biol 20, (1994). 43. Nasoni RL. Tem per a ture cor rected speed of sound for use in soft tis sue im ag ing, Medical-Physics 8, (1981). 44. Techavipoo U, Varghese T, Chen Q, et al. Tem per a ture de pend ence of ul tra sonic prop a ga tion speed and at - tenuation in excised canine liver tissue measured using transmitted and reflected pulses, J Acoust Soc Am 115, (2004). 45. Techavipoo U, Varghese T, Zagzebski J, et al. Tem per a ture De pend ence of Ul tra sonic Prop a ga tion Speed and Attenuation in Canine Tissue, Ultrasonic Imaging 24, (2002). 46. Tungjitkusolmun S, Haemmerich D, et al. Mod el ing bi po lar phase-shifted multielectrode cath e ter ab la tion, IEEE Trans Biomed Eng 49, (2002). 47. O Brien WD, Jr, Erdman JW, Jr, Hebner TB. Ul tra sonic prop a ga tion prop er ties (@ 100 MHz) in ex ces sively fatty rat liver, J Acoust Soc Am 83, (1988).