Spetrosop Reonstruted Near Infrared Tomograph Imagng for Breast Caner Dagnoss A Thess submtted to the Faulty n partal fulfllment of the requrements for the degree of Dotor of Phlosophy by Troy O. MBrde Thayer Shool of Engneerng Dartmouth College Hanover, New Hampshre May 200 Examnng Commttee Ulf Österberg, Ph. D., Char Keth Paulsen, Ph. D., Member Bran Pogue, Ph. D., Member Dean of Graduate Studes 200 Trustees of Dartmouth College Mhael Patterson, Ph. D., Member Sgnature of Author
Abstrat Near-nfrared NIR lght has the potental to be used as a non-nvasve means of dagnost magng wthn the human breast. Due to the dffusve nature of lght n tssue, omputatonal model-based methods are requred for funtonal magng wthn the breast. These methods an be used to reonstrut tomograph mages of absorpton and satterng propertes of the nteror of the breast from lght measurements at the skn's surfae. The work presented here desrbes mprovements n both the data aquston system and mage reonstruton algorthm of NIR tomograph magng for breast aner dagnoss. Spefally, the desgn, onstruton, and testng of a frequeny-doman paralleldeteton NIR data aquston system s presented. Data aquston from the system requres under 30 seonds for a sngle tomograph sle at one optal wavelength wth a measurement repeatablty for a sngle phantom on average of 0.5% n AC Intensty and 0.4 degrees n phase. In addton, methods for mproved reonstruton n the areas of magng addtonal tssue hromophores and the separaton of absorpton and satterng heterogenety are desrbed. The mproved system and reonstruton methods have been tested extensvely wth omputer smulaton and tssue-smulatng 'phantoms'. In addton, fortyfve healthy volunteers, and fve women wth breast abnormaltes have been maged. Moderate-resoluton quanttatve tomograph mages of absorpton and satterng are reonstruted from data aqured at the breast tssue surfae, and the quanttatve absorpton mages are ombned at multple optal wavelengths to generate mages of oxygenated and de-oxygenated hemoglobn, lpd, and water onentraton. Ths researh has led to an mproved novel magng system whh s potentally useful n the dagnoss and deteton of breast aner.
Aknowledgements I have been truly blessed wth the fnest group of advsors ever assembled. I quote an anonymous grant revewer: "Ths s a hghly aomplshed group. Dr. Pogue... s a keenly adept member". I owe my greatest grattude to Bran Pogue, for patently sharng hs abundane of knowledge, leadng by example wth hs treless work eth, and always fndng the tme to talk wth me and read my work. I ould not have found a better researh advsor. I also am partularly ndebted to the magan Keth Paulsen, who an smultaneously teah three ourses, run four grant programs, wrte ompetng renewals, read four theses, sour fve journal artles, mentor twenty researhers, and stll fnd tme to be a generous and knd ndvdual. There are few heads of large researh programs of hs alber. My offal thess advsor, Ulf Österberg, has been extremely helpful throughout my studes. I truly appreate the many onversatons that we have had spannng general sene and theory, teahng, runnng, polts, hansaws, and even optal tomography. He has been for me what few Ph.D. students ever atually have -- a true "advsor". It s hard to magne wth ths exellent ombnaton of nternal advsors, that I would also have a great external advsor n Mhael Patterson, untl you realze that he was Bran Pogue's Ph.D. advsor. I only hope that I an mantan ths han of exellene as Bran's frst Ph.D. advsee. I am grateful for the omments and suggestons from Dr. Patterson durng hs vsts to Dartmouth and for hs thorough readng and omments on ths thess. I would lke to aknowledge lnal ollaboraton and help from Steven Poplak, Sandra Soho, Wendy Wells and Sunshne Osterman; prevous researh by Huabe Jang and Davd Rnehart; valuable student work by Clare Wllsher, Elzabeth Whte, and Sean Stauth; expermental ad from Shudong Jang and Greg Burke; tehnal expertse from Davd Stratton, and patene and onversaton from offemates, Eljah Evenstar Wallae Van Houten, Yong Yang, Dnse Wllams, Karen Lunn, and Ha Sun. Fnanal support was provded by the Natonal Insttutes of Health and the Alma Hass Mlham Fellowshp. I would also lke to aknowledge professors Martha and Holls MBrde and ther hldren Andrew and Kar for years of enouragement towards ths degree and other more broad ntangbles. Fnally, Lots Of Valuable Enouragement from my wfe Jeanne makes t all worth whle! I wll forever gve you my lots of valuable enouragement...! And to Shaylee! I dedate ths thess to the true embodment of the Ph.D., Bors Katsen, 997, and to Quon Chuk Tam, 2000, a good father and a good man.
Contents Abstrat............................................................ Aknowledgements................................................... Table of Contents.................................................... Lst of Fgures....................................................... Lst of Tables....................................................... Chapter. Overvew...............................................A. Introduton................................................B. Bakground and Sgnfane..................................C. NIR Tomograph Imagng at Dartmouth...................... Chapter 2: FEM reonstruton algorthm.......................... 2.A. Introduton............................................... 2.B. Dffuson Approxmaton.................................... 2.C. Fnte element soluton...................................... 2.C.. Forward solver..................................... 2.C.2. Levenberg-Marquardt nverse soluton.................. 39 2.C.3. Normalzaton of the senstvty matrx.................. Chapter 3: Smulaton............................................. 3.A. Introduton.............................................. 3.B. Smulaton proedure at Dartmouth............................ Chapter 4: Instrumentaton........................................ 4.A. Introduton............................................... 4.B. Manual aquston......................................... 4.C. Intal automated aquston setup............................. 4.D. Current parallel deteton system.............................. Chapter 5: Phantom testng......................................... 5.A. Introduton............................................... 5.B. Phantom studes at Dartmouth................................ 5.B.. Lqud phantoms.................................... 5.B.2. Resn sold phantoms................................. 5.B.3. Compressble sem-sold phantoms..................... 5.B.4. Gelatn phantoms.................................... Chapter 6: Calbraton and homogeneous fttng algorthm.......... 6.A. Introduton............................................. 6.B. Pratal Consderatons..................................... 90 v v xv 2 9 28 28 34 34 43 47 47 53 53 55 63 80 80 8 84 86 87 90 v
6.C. Homogeneous Fttng Algorthm............................... 6.C.. Motvaton and prevous method Method A............ 6.C.2 Current homogeneous fttng algorthm Method B....... 6.D. Results.................................................... Chapter 7: Separaton of satterng and absorbng heterogenety... 7.A. Introduton............................................... 7.B. Hstoral motvaton........................................ 7.C. Improved algorthm......................................... 05 7.D. Phantom studes........................................... Chapter 8: Heterogeneous and Contnually-varyng phantoms...... 8.A. Introduton............................................... 8.B. Contnuously varyng and rregularly shaped objets............. 8.C. Heterogeneous Bakground and Inlusons..................... Chapter 9: Funtonal Imagng.................................... 9.A. Introduton.............................................. 9.B. Theory.................................................... 9.B.. Tssue Charatersts................................ 30 9.B.2. Determnaton of [Hb-T] and SO 2.................... 9.C. Smulaton of Breast Caner Lesons.......................... 9.D. Homogeneous Measurements................................ 9.E. Heterogeneous phantom studes............................. 9.E.. Phantom wth nreasng blood onentraton........... 9.E.2. Phantom wth ontrast n [Hb-T] and SO 2............. Chapter 0: Patent Imagng...................................... 0.A. Introduton.............................................. 0.B. Intal automated system.................................... 0.B.. Patent 5: 3 m fbroadenoma....................... 0.B.2. Patent 6: 0.8 m nvasve aner..................... 0.B.3. Comparson wth early normal values................ 0.C. Current automated system.................................. 0.C.. Normal patent study............................. 0.C.2. Study of Lpd and Water ompensaton.............. Chapter : Conludng remarks...................................a. Conluson...............................................B. Future studes............................................ Appendx A: Itemzed Components Lst............................ Appendx B: Fle namng onventons.............................. Referenes......................................................... v 93 93 95 99 04 04 3 8 8 22 30 30 3 34 40 5 52 55 59 59 59 6 62 63 63 65 75 76 8 83 87
Lst of Fgures Fgure.: Shemat of the funtonal magng proess.................... Fgure.2: a Dagram of data aquston system, b photograph of fber-opt nterfae............................................................ Fgure.3: Flow hart of fnte element based mage reonstruton algorthm for a sngle wavelength................................................... Fgure.4: a Plot of NIR absorpton spetra for water, lpds, Hb-O 2 and Hb-R. Values for absorpton oeffent are dsplayed for anatomally relevant onentratons for breast tssue. b Mathematal dagram of onverson of absorpton mages to funtonal mages of breast tssue hromophores........... Fgure.5. Shemat for mage reonstruton showng requrements left and steps for testng rght mage reonstruton methods........................ Fgure.6. Smulatons showng mproved reonstruton of satterng and absorbng nlusons for nose-free data. a Orgnal smulated mage wth three embedded objets. Fgure b shows the reonstruted absorpton oeffent mage and the reonstruted satterng oeffent mage......................... Fgure.7. Smulatons of heterogeneous objet wth heterogeneous nluson wth an average 2: ontrast n absorpton and :2 absorpton n satterng. The frst two olumns are the orgnal absorpton and satterng dstrbutons pror to smulaton. The seond two olumns are the reonstruted absorpton and satterng mages. Eah suessve row represents an nreasng level of added heterogenety to the orgnal dstrbuton n the frst row. All absorpton and satterng mages are dsplayed on the dental sale......................... Fgure.8: Smulatons showng mprovement of oxygen saturaton mages wth nreasng number of wavelengths n the presene of sgnfant nose. a Orgnal smulated oxygen saturaton mage wth embedded objet wth 20% derease n oxygen saturaton, whh was used to generate smulated data wth 0% added nose. Fgure b shows the reonstruted oxygen saturaton mage usng only 2 wavelengths and the reonstruted oxygen saturaton mage usng 4 wavelengths......................................................... Fgure.9: Hemoglobn onentraton and oxygen saturaton mages of a phantom ontanng two targets -- one wth a 4: hemoglobn onentraton ontrast ompared wth the bakground and the other had a 00% derease n oxygen saturaton ontrast ompared to the bakground. The fgure a shows a shemat of the phantom, b reonstruted hemoglobn Hb onentraton mage, oxygen saturaton mage. d and e are one-dmensonal profle plots horzontal through the enter of b and respetvely. Ideal profles are plotted for omparson.......................................................... Fgure.0: Images of patent wth bengn fbroadenoma. Shown n the fgure are the a x-ray mammograph, b representaton of loaton of tumor taken from a sketh by radologst based on x-ray mammography mages and palpaton, hemoglobn onentraton mage reonstruted from near nfrared absorpton mages at 750 and 800 nm.............................................. Fgure.: Images of patent wth malgnant dutal nvasve arnoma. Fgure dsplays a x-ray mammograph, b representaton of loaton of tumor taken from a sketh by radologst based on x-ray mammography mages and palpaton, and 0 2 6 7 9 2 22 23 24 v
hemoglobn onentraton mage reonstruted from near nfrared absorpton mages at 750 and 800 nm.............................................. Fgure.2: Plot of average hemoglobn onentraton from near-nfrared mages of volunteers. Also plotted are two loalzed values for hemoglobn onentraton for a malgnant tumor and a bengn tumor.................................. Fgure.3: Plot of average hemoglobn onentraton from near-nfrared mages obtaned from a set of sxteen volunteers wth normal mammograms............ Fgure 2.: Fnte element fne mesh 425 nodes, 777 elements used n the forward solver for ths thess. The letter S and D represent soure and detetor loatons respetvely.................................................. Fgure 3.: Shemat of Matlab program to generate optal property dstrbutons and map them onto the fnte element mesh. a A 200x200 pxel mage of the dstrbuton s reated n Matlab. b The mage s blurred by an averagng flter wth sze equal to the average node spang. The mesh s overlayed on the mage and the values of the pxel nearest to eah node s wrtten to fle. Ths proess effetvely uses lnear nterpolaton to map the dstrbuton onto the nodes............................................................ Fgure 3.2: Shemat of the forward smulator. Node-based mages of a a and b s ' generated from Matlab program. Ampltude and phase measurement data from the forward smulaton for mages a and b labeled "experment" n ths graph. Plotted for referene s the measurement data for a homogeneous dstrbuton wth the same bakground propertes as a and b................. Fgure 3.3: Reovered mages from smulated measurement data plotted n Fgure 3.2. usng the FEM reonstruton program. Dsplayed are the node-based mages of a a and b a '. No nose was added to the measurement data n ths ase..... Fgure 4.: An early phantom magng setup. The ylndral phantom s n the enter of the mage. Two fber opts lead to holes drlled n the phantom, one from a laser soure and the other leadng to a PMT detetor. The nlusons n the phantom an be arefully postoned usng the setup attahed to the top of the ylnder, and phantoms were maged wth and wthout the nluson present....... Fgure 4.2: Some phantom mages obtaned from the frst expermental setup. The ontents of the phantoms are summarzed below the mages and further detals an be found n the orgnal paper by Jang et al............................. Fgure 4.3: Shemat of the frst-automated magng setup showng the onnetons to and from the laser soure and PMT on translaton stages.......... Fgure 4.4: Photograph of the automated setup. Note that for ths early setup, fbers were postoned ndvdually by loosenng of the thumb srew and manual reloaton........................................................... Fgure 4.5: Heterodynng shemat from Davd Rnehart thess, where n our ase,?=2p00 MHz s the loal osllator sgnal,?+d=2p00.00 MHz s the frequeny of the nput and deteted sgnal, and d=2p khz s the heterodyne or "beat" frequeny. A 0 khz low-pass flter n the data aquston blok removes the hgh-frequeny sgnals.............................................. Fgure 4.6: Photograph of laboratory setup, where TS laser, Argon on pump laser, and eletro-opt modulator are mounted on optal benh on the rght. The frst patent bed setup s shown at the left sde of the photograph................ Fgure 4.7: Photographs of assembled lathe-huk fber postonng system. By rotatng the metal spokes n red below the lathe huk, the fber opts are moved 25 25 26 34 48 49 50 54 55 56 57 59 60 v
radally. The photographs on the rght show the fbers at a lose poston dameter of openng = 6 m and at a ompletely open poston dameter = 0 m. A "labjak" s mounted below the lathe huk to allow for vertal postonng of the array............................................................... Fgure 4.8: a Top vew and b utaway sde vew of ntal radal postoner desgn.............................................................. Fgure 4.9: Image of Laser dode array. The dode lasers A-5 are mounted n fber launh modules. A omputer ontrolled RF swth B provdes the dode wth eletral urrent. The output of the lasers s nput nto a 6: fber opt ombner C........................................................ Fgure 4.0: Image of rak mounted system. The four labeled levels house A the omputer, B the stage ontroller and PMT voltage soure, C the fve laser dodes, RF swth, and urrent soure, D the frequeny generators............. Fgure 4.: The parallel deteton assembly. a The 6 RF mxers are mounted on a rular alumnum ds, whh s attahed to a rotary stage fxed to a square alumnum plate. A DC spltter box s also mounted on the enter of ths plate. b The 6 PMTs are mounted on the perphery of a seond rular ds. The RF spltter and soure fber are also mounted to ths ds. The fully assembled deteton array nludes the PMT ds mounted above the mxer ds. A fxed square alumnum plate for the fber opt lght delvery s mounted above the PMTs.............................................................. Fgure 4.2: The fber opt/patent nterfae for dret fber ontat wth the breast n a rular geometry. Fber opt bundles at the tssue nterfae A are mounted to sxteen radally-algned horzontal translaton stages B. The nterfae s attahed to a vertal translaton stage C. The motor ontrollers D and the deteton array E are mounted to an alumnum base below the patent nterfae. The vertal and horzontal stages are ontrolled by a sngle push button ontroller. Fgure 4.3: The ustom bult patent table. The woman les prone on the table wth her breast pendant through the openng. The fber-opt nterfae s loated below the openng. An nexpensve PC vdeo-am s plaed below and updated mages are vewed on a omputer sreen suh that the nurse an easly poston the fbers n ontat wth the breast usng the push-button ontroller................ Fgure 4.4: a AC ampltude and b phase of deteted sgnal for the same optal sgnal for all 6 PMT detetors.................................... Fgure 4:5: a Relatve AC power level for the same measured voltage 0. V at dfferent gan settngs. b Phase shft measured for the dental optal phase but at dfferent gan levels. Both measurements are reported for detetor number one.. Fgure 4:6: a AC Voltage plotted aganst relatve optal power and b measured phase shft for the same optal sgnals for eght representatve detetors D - D8 at a sngle gan settng of 0.7 V. A log sale s used for the power of the optal sgnal n plot b................................................ Fgure 4.7: Calbraton urves for PMT response at dfferent gan settngs. A fxed optal sgnal ntensty was vared wth neutral densty flters for all gan settngs. A regresson to the deteted log ampltude at eah detetor was performed versus log of nput optal power. The onstant term of the regresson s related to the lnear slope ampltude/nput power and s plotted n fgure a versus gan level for eght representatve detetors D - D8. The phase shft s approxmately onstant for deteted sgnals below saturaton level and above the nose floor. The 6 62 64 65 67 69 70 72 72 73 v
average phase shft for eght detetors D - D8 s plotted n fgure b for dfferent gan settngs......................................................... Fgure 4.8: a Plot of the deteted AC ntensty sgnal versus the nput power at the gan settng of 0.95 of the maxmum gan. The long dashed lne s the system nose level and the short dashed lne s the nose level of the eletrons. The sold lne ndates the lnear response of the detetor pror to reahng the nose floor. b Plot of phase error wth respet to nput power. The phase error dereases wth nreasng sgnal to a onstant RMS level of less than a degree at typal power levels............................................................... Fgure 4.9: a Absorpton and b satterng mage of a homogeneous phantom wth albraton and expermental data aqured by manual postonng setup. Absorpton and d satterng mage from automated postonng setup. The x and y axes of the mages are unts of mllmeters and the sale bar unts are nverse mllmeters.......................................................... Fgure 5.: Reovered values for a a and b s ' at 785 nm for Intralpd per unt onentraton for homogeneous soluton based on data aqured wth urrent deteton system and proessed wth the homogeneous fttng algorthm.......... Fgure 5.2: Reovered values for a a and b s ' at 785 nm for 2% Ink per unt onentraton. Note a nterept s 0.004 mm - nstead of expeted value for water of 0.002 mm -................................................... Fgure 5.3: Photographs of resn sold phantoms. a homogeneous phantom, b phantom wth holes drlled for addton of nlusons......................... Fgure 6.: Plot of entre 256 measurement a lnintensty and b phase shft data set. Ths data s averaged over all soures and homogeneous fttng algorthm appled to fnd best ft. Also dsplayed s lnintensty and d phase data averaged over all soures and the alulated data for the best ft homogeneous optal propertes..................................................... Fgure 6.2: Shemat desrbng two homogeneous fttng algorthms.......... Fgure 6.3: Plot of measured and alulated a log of dstane tmes AC ntensty, lnri, and b phase,?, versus soure to detetor dstane, r.................... Fgure 6.4: Plot of a lnintensty data at 6 detetor stes showng offset due to algnment dfferenes, b alulated data from fnte element soluton, and dfferene between a and b. Dfferene data s used as the 'albraton' fator............................................................... Fgure 6.5: Image after frst teraton of reonstruton of absorbng objet wth Gaussan profle wth a ntal estmate equal to average optal propertes, b 0% error n ntal estmate, and 50% error n ntal estmate. The vertal axs unts are n mm -. d, e, and f are profle plots along a vertal lne through the enter of mages a, b, and respetvely............................... Fgure 6.6: Plot demonstratng measured data wth overall offset due to long term drft of laser ntensty. The average dfferene s used to orret ths offset....... Fgure 6.7: Homogeneous fts to measured data for the two methods desrbed n Fgure 6.2: a Method A satterng oeffent data, b Method B satterng oeffent data, Method A absorpton oeffent data, d Method B absorpton oeffent data. Eah onentraton was measured three tmes at eah of three wavelengths......................................................... Fgure 6.8: Shemat of nputs to mage reonstruton algorthm from homogeneous albraton and ntal estmate alulaton. Inputs from albraton methods are dsplayed n bold boxes...................................... 74 77 78 83 84 86 92 95 97 99 00 0 02 03 x
Fgure 7.: Dagram of senstvty matrx, J T J. The parttons llustrate the satterng and absorpton seton of the matrx for NN number of nodes and NM measurement stes. ln I θ and are the dervatve of the ampltude wth D j a j respet to dffuson oeffent D and the dervatve of phase shft wth respet to absorpton oeffent a, respetvely, for measurement ste at node j......... Fgure 7.2: One teraton reonstrutons for a smulated absorbng-only nluson 2: ontrast n a. The absorpton and redued satterng oeffent mages are presented respetvely usng a,b the orgnal non-normalzed phase-andampltude method,,d the method wth normalzed J T J matrx, and e,f the urrent method where the J T J matrx s normalzed and the phase terms are weghted............................................................ Fgure 7.3: One teraton reonstrutons for a smulated nluson wth 2: ontrast n both a and s '. The absorpton and redued satterng oeffent mages are presented respetvely usng a,b the orgnal non-normalzed phaseand-ampltude method,,d the method wth normalzed J T J matrx, and e,f the urrent method where the J T J matrx s normalzed and the phase terms are weghted............................................................ Fgure 7.4: a Orgnal absorpton oeffent a and b the redued satterng oeffent s ' smulated mage wth three embedded objets. Fgure shows the reonstruted a mage and d the s ' mage................................ Fgure 7.5: An 86 mm dameter sold ylndral tssue-smulatng phantom wth three 7 mm dameter ylndral holes s shown. Absorbng-only ontrast was added to hole A, satterng-only ontrast to hole B, and absorbng and satterng ontrast to C. Bakground optal propertes were estmated to be a =0.006 mm - and s '=.0 mm -..................................................... Fgure 7.6: Reonstruted mages of a,,e a and b,d,f s ' for ontrast levels of a,b ½:,,d ½:, and e,f 2½: for both oeffents are dsplayed. In the upper entral hole C, both a and s ' are nreased together, whereas n the left hole A only a s nreased, and n the rght hole B only s ' s hanged. The x and y axes are poston n mllmeters and the unts of the sale bars are nverse mllmeters.......................................................... Fgure 7.7: Comparson of reonstruted and expeted ontrast at the enter of the reonstruted objets n Fgure 7.6. The reonstruted values for a absorpton and b satterng oeffent are plotted versus the expeted results for embedded objets wth ontrast n A absorpton-only, B satterng-only, and C absorpton and satterng. The reonstruted rosstalk n terms of absorpton ontrast for the satterng-only objet and d satterng ontrast for the absorptononly objet are also dsplayed........................................... Fgure 8.: Orgnal and reonstruted absorpton oeffent mm - mages of an objet wth a Gaussan profle and a rular step nluson. 3% nose was added to the smulated measurement data pror to mage reonstruton. a Orgnal rular step objet; b orgnal Gaussan objet; horzontal profle of a; d horzontal profle of b; e reonstruted mage of a; f reonstruted mage of b; g horzontal profle of e; h horzontal profle of f. X-Y oordnates are n unts of mm................................................................ 07 2 3 4 5 6 20 x
Fgure 8.2: Smulaton of objets wth dfferng rregularly shaped nlusons n a and s '. The orgnal a and s ' property dstrbuton s shown n a and b, wth the reonstruted mages n and d. The sale bars are n unts of mm -.......Fgure 8.3: Exsed breast tssue ontaned n a thn plast bag s plaed wthn the fber opt nterfae. The sxteen soure and detetor fbers are radally postoned n dret ontat wth the sample................................ Fgure 8.4: The frst two olumns are the orgnal absorpton and satterng dstrbutons pror to smulaton. The seond two olumns are the reonstruted absorpton and satterng mages. Eah suessve row represents an nreasng level of added heterogenety to the orgnal dstrbuton n the frst row. All absorpton and satterng mages are dsplayed on the dental sale............ Fgure 8.5: a Reonstruted absorpton and b satterng mages of embedded pork heterogenety wthn exsed breast tssue. and d are the profle graphs through a and b respetvely. e-h are dental mages for the ase where the sample was rotated 90 degrees ounterlokwse............................ Fgure 9.: Plot of near-nfrared absorpton spetra for water, lpds, oxygenated Hb-O 2, and de-oxygenated hemoglobnhb-r. Values taken from Wray et al. for Hb-O 2 and Hb-R, from Hale and Querry for water, and from Quaresma et al. for lpds............................................................... Fgure 9.2: a Absorpton oeffent mm - mage at 750 nm of a smulated normal tomograph seton of breast tssue ontanng a anerous leson. b Vertal profle graph through the enter of a. Hemoglobn onentraton M mage and d hemoglobn oxygen saturaton % mage of the smulated breast and anerous leson. e and f are vertal profle graphs through the enter of mages and d respetvely. X-Y oordnates are n unts of mm............ Fgure 9.3: Results for 0% nose mage reonstrutons of a smulated breast aner leson wthn a tomograph ross seton of normal breast tssue. The orgnal mages and values are shown n Fgure bakground of 0 M hemoglobn onentraton and 70% oxygen saturaton ontanng a rng of 40 M hemoglobn onentraton wth 50% oxygen saturaton. a Hemoglobn onentraton mage and b hemoglobn oxygen saturaton mage for the regresson usng two wavelengths. and d are vertal profle graphs through the enter of a and b respetvely. e and f are the hemoglobn onentraton and oxygen saturaton mages usng three wavelengths. g and h are profle graphs through the enter of e and f., j, k, and l are the mages and profle graphs for the regresson usng four wavelengths...................... Fgure 9.4: Comparson of measured values and those predted by Staveren et al. for the redued satterng oeffent s of a phantom ontanng 0.5% Intralpd n water. Expermental estmates are reahed by fttng the data from a homogeneous phantom to the fnte element model usng Method A desrbed n Chapter 6........................................................... Fgure 9.5: Plot of expermentally measured absorpton oeffent versus onentraton of blood for three wavelengths: ahb-r and bhb-o 2. Measurements were reorded from a homogeneous satterng medum of 0.5% Intralpd n a 72 mm dameter ylndral phantom. Homogeneous fttng Method A was used for ths data................................................ Fgure 9.6: Comparson of absorpton oeffent alulated from expeted values for blood and water and measured values: a 0 M de-oxygenated hemoglobn n 0.5% Intralpd and water soluton; b 0 M oxygenated hemoglobn n 0.5% x 2 24 25 27 33 36 39 42 44
Intralpd and water soluton. Homogeneous fttng Method A was used for ths data............................................................... Fgure 9.7: Comparson of absorpton oeffent values from Hale and Querry wth the absorpton oeffent of water measured n a plast beaker versus a thnwalled balloon. Note that the offset s not apparent n the thn-walled balloon ase. Homogeneous fttng Method A was used for ths data........................ Fgure 9.8: Comparson of absorpton oeffent alulated from expeted values for blood and water and measured values: a 24 M de-oxygenated hemoglobn and b 24 M oxygenated hemoglobn n 0.5% Intralpd and water soluton. Homogeneous fttng Method B was used for ths data........................ Fgure 9.9: NIR spetra and graphal results of least squares fts to determne hemoglobn onentraton and hemoglobn oxygen saturaton level for a 33 year old woman, and b 62 year old woman.................................... Fgure 9.0: Colleton of absorpton oeffent mages for a 25mm rular nluson n a 72 mm rular phantom. The bakground s 0.5% blood n 0.5% Intralpd and water. The nluson n the frst row has no blood. The nluson n the seond row ontans % blood, the thrd row.5% blood and the fourth row 2% blood. Column a s measured at 750 nm, olumn b at 800 nm, and olumn 830 nm............................................................. Fgure 9.: Comparson of absorpton oeffent taken from the enter of the nluson n the mages n Fgure 9.0 wth absorpton oeffent values alulated from expeted values for blood and water.................................. Fgure 9.2: Absorpton oeffent a, hemoglobn onentraton [Hb-T], and hemoglobn oxygen saturaton SO 2 mages of an nreased [Hb-T] nluson left off-enter and a de-oxygenated blood nluson rght off-enter n a oxygenated blood bakground. a Depton of orgnal phantom, b 750 nm a mm - mage, 802 nm a mage, d [Hb-T] M mage, e SO 2 mage. f and g are onedmensonal transets through mages d and e respetvely. Ideal profles are plotted for omparson................................................. Fgure 0.: Reonstruted mages for a woman wth a bengn fbroadenoma Patent 5. Dsplayed are the absorpton oeffent a mages sale bar unt of [mm - ] at a 750 nm and b 800 nm and the hemoglobn onentraton [Hb-T] mage sale bar unt of [M]. The x and y axes of the mages are poston n unts of mm. Also dsplayed are the values for a vertal transet through the mddle of the [Hb-T] mage. y-axs s [M] and x-axs s vertal poston n mm......... Fgure 0.2: Reonstruted mages for the ontralateral mammographally normal breast for patent 5. Dsplayed are a the hemoglobn onentraton [Hb-T] mage and b vertal transet through the mddle of the [Hb-T] mage. sale bars and axes are dental to Fgure 0. and d................................. Fgure 0.3: Reonstruted mages for a woman wth an nvasve arnoma Patent 6. Dsplayed are the absorpton oeffent a mages sale bar unt of [mm - ] at a 750 nm and b 800 nm and the hemoglobn onentraton [Hb-T] mage sale bar unt of [M]. The x and y axes of the mages are poston n unts of mm. Also dsplayed s the values for a vertal transet at x = 42 mm of the [Hb- T] mage. y-axs s [M] and x-axs s vertal poston n mm................ Fgure 0.4: Reonstruted mages for the ontralateral mammographally normal breast for patent 6. Dsplayed are a the hemoglobn onentraton [Hb-T] mage and b vertal transet through the mddle of the [Hb-T] mage. sale bars and axes are dental to Fgure 0.3 and d................................. 45 46 47 50 53 54 56 60 60 6 62 x
Fgure 0.5: Plot of average hemoglobn onentraton from near-nfrared mages of volunteers. Also plotted are two loalzed values for hemoglobn onentraton for a malgnant tumor and a bengn tumor.................................. Fgure 0.6: Plot of average a hemoglobn onentraton, b oxygen saturaton, absorpton oeffent at 785 nm, and d redued satterng oeffent at 785 nm versus age from near-nfrared mages obtaned from a set of sxteen volunteers wth normal mammograms. Error bars represent dfferene n mean value between the two breasts...................................... Fgure 0.7: Plot of average hemoglobn onentraton versus a body mass ndex kg/m 2 and b radograph densty from near-nfrared mages obtaned from a set of sxteen volunteers wth normal mammograms. The error bars represent the standard devaton between multple measurements, suh as left and rght breast or multple planes wthn a sngle breast. In b, the error bars are based on values from multple subjets.................................................. Fgure 0.8: Tomograph NIR mages of female breast wth 3 m leson loated at 3 o'lok presented n the manner used by our lnal database. a. - d. and. - l. are a and s ' mages n graysale unts of /mm at the four wavelengths ndated. e. and g. are Hb-T and SO 2 mages determned usng method as desrbed n the text assumpton of 30% and 60% bulk water and lpd ontent. The x and y axes of the mages are poston n unts of mm. The average values and standard devaton of all pxels are ndated below eah mage. f. and h. are horzontal transets through the enter of mages e. and g..................... Fgure 0.9: Images of a. water, b. lpds,. Hb-T and e. SO 2 are reovered usng a least squares ft to determne all four parameters method 2 from the a mages presented n Fgure 0.8. The x and y axes of the mages are poston n unts of mm. d. and f. are horzontal transets through the enter of mages. and e....... Fgure 0.0: Images of a. satterng power b. water,. lpds, d. Hb-T and f. SO 2 are reovered usng method 3. The satterng power mage s alulated from a ft to ln s ' for the sequene of mages n Fgure 0.8 versus lnλ. The water and lpd mages are then determned from equatons 2.a. and 2.b. The x and y axes of the mages are poston n unts of mm. e. and g. are horzontal transets through the enter of mages d. and f............................................ Fgure 0.: Tomograph NIR mages of the ontralateral normal female breast. a. - b. and. - d. are a and s ' mages n graysale unts of /mm at the two wavelengths ndated. 785 and 826 nm mages are omtted. e. and g. are Hb-T and SO 2 mages determned usng method as desrbed n the text assumpton of 30% and 60% bulk water and lpd ontent. The x and y axes of the mages are poston n unts of mm. f. and h. are horzontal transets through the enter of mages e. and g....................................................... 63 64 65 70 7 72 73 x
Lst of Tables Table.: Results for average hemoglobn onentraton of the bakground and lesons for two women maged usng the expermental data aquston set-up..... Table 5.: Repe for resn phantom mxng proedure developed by Elzabeth Whte.............................................................. Table 5.2: Repe for slone soft phantom Shudong Jang mxng proedure... Table 5.3: Repe for gelatn phantom Dun L/Todd Kerner/Elzabeth Whte proedure........................................................... Table 6.: Comparson of molar extnton oeffent for oxygenated hemoglobn measured by Wray et al. wth results from use of homogeneous fttng algorthm Method B. based on data from phantoms nreasng hemoglobn onentraton Fgure 6.7.d........................................................ Table 8.: The normalzed standard devaton dvded by the mean of the spatal propertes of the mages n Fgure 0.4 are tabulated. These values are ompared wth the reonstruted ontrast of the embedded objet for dfferent levels of added heterogenety. The normalzed standard devaton s the same untless sale as the reonstruted ontrast, thus, for a devaton of 0.20, an objet wth.2 or 0.8 ontrast would be at the level of plus or mnus one standard devaton........... Table 8.2: Values from homogeneous ft ompared wth reovered entral values of embedded pork. Also, the normalzed standard devaton desrbed n Table 8. of the entre mage s reported....................................... Table 9.: Comparson of the expermental determnaton of the molar absorpton oeffent slope of a lnear regresson to data wth the Wray et al. and Hale and Querry values: ahb-r and bhb-o 2..................................... Table 9.2: Results of least squares fts to alulate Hb-O 2 and Hb-R ontent of known 0 M hemoglobn onentraton, 0.5% Intralpd and water homogeneous phantom............................................................ Table 9.3: Numeral results of least squares fts to determne hemoglobn onentraton and hemoglobn oxygen saturaton level for a 33 year old woman, and b 62 year old woman. Water and lpd onentratons were onstraned to the approxmate values dsplayed........................................... Table 9.4: Lnear regresson results from data plotted n Fgure 9. A slope of ndates a : orrespondene between the absorpton oeffent values from the enter of the nluson n the mages n Fgure 9.0 and the absorpton oeffent results alulated from expeted values for hemoglobn and water............... Table 0.: Results for average hemoglobn onentraton of the bakground and peak onentraton for patents 5 and 6.................................... Table 0.2: Values for SO 2 and Hb-T for dfferent methods of ompensatng for water and lpd ontent. Method, the assumpton of a fxed bulk water and lpd ontent, orresponds to Fgure 0.8. Method 2 Fgure 0.9 nvolves a leastsquares ft to all four hromophores. Method 3 Fgure 0.0 uses the satterng power mage to mage water and lpds. Method 4 assumes no lpd and water ontent............................................................. 25 85 87 88 03 24 26 43 46 49 54 62 72 xv
Chapter : Overvew.A. Introduton Ths thess desrbes the development and mprovement of both the data aquston system and mage reonstruton algorthm for near-nfrared NIR tomograph magng for breast aner dagnoss. Spefally, the data aquston system was mproved n both speed and auray through the mplementaton of parallel aquston, and several avenues were explored for mprovng reonstruton n the areas of magng tssue hromophores and separaton of absorpton and satterng heterogenety. The mproved system and reonstruton methods were tested extensvely n omputer smulatons and tssuesmulatng 'phantoms', as well as healthy volunteers and some breast aner patents. The thess s organzed nto hapters: Overvew 2 Reonstruton algorthm: The dervaton and formulaton of the model and reonstruton program are presented. 3 Computer smulatons: The methods and tools used for omputer smulaton of NIR tomograph magng are summarzed. 4 Imagng system: The entre magng system nludng data aquston system, fber-opt nterfae, laser soures, and omputer ontrols s presented. The desgn, onstruton, and testng of a sxteen hannel parallel deteton system s detaled. Some hstoral data are provded on past data aquston systems. 5 Phantom studes: The materals and methods for onstrutng and testng wth tssue-smulatng phantoms are presented. 6 Calbraton and homogeneous fttng algorthm: The albraton of the DAQ system s desrbed. Of partular mportane to albraton, the homogeneous fttng algorthm s detaled.
7 Separaton of satterng and absorpton heterogenety: Methods and results for mprovng separaton of absorbng and satterng heterogenety are presented and dsussed. 8 Heterogeneous Phantoms and Contnually-varyng objets: Typal phantom and smulaton studes are performed wth step objets n a homogeneous bakground. Both normal and anerous breast tssue s heterogeneous. Smulaton and phantom studes are presented whh nvolve heterogeneous bakground and objets wthout step hanges n optal propertes. 9 Imagng of tssue hromophores: Usng omputer smulatons and NIR measurement of phantoms, least squares fttng tehnques and the use of multple optal wavelengths are explored as a means to extrat spatal maps of oxygenated Hb-O 2 and de-oxygenated hemoglobn Hb-R as well as water and lpds. 0 Patent Imagng: Data has been aqured from both healthy human volunteers and patents wth known breast malgnanes usng the parallel data aquston system. These data are reonstruted n attempt to better separate and understand the role of satterng, absorpton, and tssue hromophore heterogenety wthn the healthy human breast and breast aner tumors. Conlusons/ Future Dretons The work presented here bulds on mportant researh n NIR magng and photon mgraton at other nsttutons and prevously at Dartmouth College. The followng seton attempts to provde the hstory of researh n the feld and provde the reader wth a better dea of the orgn of the researh desrbed heren..b. Bakground and Sgnfane.B.. Bakground of Optal Imagng 2
As early as 929, Cutler et al. reported utlzng a lght for magng through the breast. Ths orgnal optal projeton magng, or transllumnaton, was arred out n a dark room by vsually lookng for shadows ast by large breast tumors. Improvements n transllumnaton followed, nludng the use of two wavelength regons as well as olor flm and vdeo ameras for deteton. For ths ontnuous-wave projeton magng, senstvty has been found to be smlar to x-ray mammography for large tumors, but senstvty for small tumors under m 2 and spefty for all tumor szes s substantally lower 3. Several developments have led to renewed nterest n optal magng of the breast, spefally, n vvo NIR spetrosopy, photon mgraton theory, mprovements n lght generaton and deteton, and vnumeral modelng tehnques oupled wth the avalablty of nreasng omputng power. These spef areas are dsussed below for ther relevane to ths thess. The understandng that hemoglobn onentraton and oxygen saturaton an be reovered from non-nvasve NIR spetrosop measurements through several entmeters of tssue arose over the past entury. As early as the turn of the entury, measurements of hemoglobn dervatves at two wavelengths were dsussed as a method for assessng oxygenaton, 4,5 however, lnal and ommeral oxmeters were not developed untl the 930s 6,7 and 970s 8 respetvely. A study by Jobss 9 reported n 977 usng NIR wavelengths to determne onentratons of oxygenated and de-oxygenated hemoglobn s onsdered a pvotal demonstraton of the potental of NIR spetrosop measurements for tssue volumes 5. Hemoglobn ontrast n the breast between anerous and normal tssue was estmated to be 4: by Profo et al. 0 usng exsed tssue and performng vessel ounts. Further haraterzaton of ontrast n the NIR was shown by Peters et al. and Troy et al. 2 usng optal measurements of exsed tssue. In the late 990's, ths hemoglobn ontrast 3
has been onfrmed from n vvo NIR measurements 3-5 and estmated to be wthn the range of 2: to 3:. The understandng of lght propagaton n tssue, spefally the development of photon mgraton theory and the dffuson approxmaton to lght travel n a hghly satterng medum has generated an entrely new feld of study. The applaton of the dffuson approxmaton appled to laser pulses n hghly satterng meda s assoated wth researh by Ishmaru n 978. 6 Whle dffuson theory s useful for modelng CW lght propagaton, the use of tme-resolved measurements provdes nformaton on the nreased pathlength of lght n tssue due to multple satterng events, thus mprovng the ablty to separately reover absorpton and satterng propertes. The use of tme-doman measurements -- reordng the pont spread of a poseond pulse of lght after passng through a tssue volume -- was ntated n several groups nludng measurements by Chane et al. 7,8 Applaton of analyt solutons from dffuson theory were frst presented by Patterson et al. 9 allowng quanttatve reovery of optal propertes from tme-resolved measurements. Frequeny-doman measurements, where the phase shft of a snusodal wave passng through a tssue volume s reorded, provde a more nexpensve and perhaps stable manner of aqurng tme-based nformaton. Fshkn et al. 20 and Patterson et al. 2 developed smlar analyt solutons n the frequeny-doman and performed some of the frst frequenydoman measurements to non-nvasvely reover optal propertes. The development of fber opts, lasers, and photomultpler tube PMT detetors has been essental n the ablty to buld nexpensve buld NIR data aquston systems. Clad fber opts were developed n the 950s, wth the frst low-loss fbers 7 db/km reported by Cornng n 970. 22 Commeral use of fber opts has taken off only durng the past 20 years n onert wth the wdespread avalablty of semondutor lasers. Sne the development of the frst laser n 960 by Theodore Maman, the feld of lasers n 4
medne has taken off rapdly. Wth the avalablty of nexpensve and stable semondutor lasers at many wavelengths throughout the NIR, CW and frequeny-doman setups an be run off smple wall outlets for power. Tme-doman systems requre modeloked lasers whh surfaed n the 970s and tend to be more expensve and senstve to moton. 23,24 PMTs were developed n the late 930s and have sne been the standard for applatons requrng both senstvty and speed of response. 25 Whle sold-state detetors, suh as harged-oupled deves CCD and photododes PD, were developed later and tend to be more stable for CW and low frequeny applatons, nether has yet to rval the PMT n senstvty at hgh frequeny. v Exponental nreases n omputatonal power have led to an exploson of the applaton of numeral methods to nverse problems. Fnte element FEM based algorthms appled to optal magng whh would have been unsolvable only a few deades ago due to memory and speed onstrants, an urrently be solved n two-dmensons n a matter of mnutes. Even wth the hgh power omputng avalable at the turn of the 2st entury, however, full 3-dmesonal solutons stll an requre many hours. Many of the ntal applatons of the dffuson approxmaton to lght travel n tssue were based on analyt solutons usng smplfyng assumptons, suh as sem-nfnte or perturbaton theores, 9-2 to better estmate absorpton and satterng propertes from pont measurements. For more general magng, numeral methods whh an aurately model the lght propagaton for arbtrary geometres and property dstrbutons have been appled. Snger et al. 26 desrbed model-based mage reonstruton methods as appled to objets that "dffuse radaton" n 990. Arrdge et al. 27,28 appled fnte element FEM modelng to lght propagaton n tssue based on the tme-resolved dffuson approxmaton. Inverse solutons whh mnmze the dfferene between alulated and measured data suh as Newton mnmzaton or onjugate gradent methods have been appled n both tme-doman 29 and 5
frequeny-doman 30,3 to reover mages of absorpton and satterng based on smulated measurements at the perphery of a hghly satterng objet. The ombnaton of all of these developments has led to the urrent potental of NIR spetrosop magng for quanttatvely reoverng spatally-resolved maps of funtonal parameters suh as hemoglobn. Dffuse optal magng s beng examned for applatons suh as breast magng, 4,32-36 neo-natal bran montorng, 37 and mappng bran funton. 38 Ths thess onentrates on breast magng applatons, but many of the developments desrbed apply n general to the feld of dffuse optal magng..b.2. Health-related sgnfane Quanttatve measurements of hemoglobn related parameters n human tssue have been demonstrated wth non-nvasve tehnques usng near nfrared lght. 3,4,38-42 Wth the use of model-based omputatonal methods near nfrared dffuse optal tomography an generate spatally-resolved absolute mages of hemoglobn onentraton and oxygen saturaton. 5,35 These quanttatve mages are obtaned non-nvasvely and at low ost. These hemoglobn based mages may be useful for both breast aner dagnoss and deteton. Loalzed nreases n blood vessel growth angogeness and therefore blood volume are observed at the loaton of anerous tumors; ths nrease has been estmated to orrespond to a two to four fold nrease n hemoglobn onentraton wthn breast aner tumors relatve to healthy tssue. 0,42,43 Also, t has been reported that a loalzed.4 to 4.4 derease n oxygen partal pressure s present wthn breast aners. 44,45 Ths dereased oxygen pressure may orrespond to a measurable derease n hemoglobn oxygen saturaton. These two physologal fators suggest that there s onsderable potental for quanttatve hemoglobn magng as a method of haraterzng breast tumors non-nvasvely. 6
The urrent method of breast sreenng, x-ray mammography, s wdely aepted and prated. X-ray mammography s hgh resoluton and very good at detetng abnormaltes n older women. In fat, t s estmated that 85%-90% of breast aner tumors are deteted n ths populaton. 46 However, x-ray mammography has a low postve predtve value -- about 60% to 85% of bopses are performed on bengn tumors. 47 X-ray methods have lower senstvty for dense breast tssue whle also havng rsk n ndung aners and there usage for sreenng for women under 50 years of age wthout a hstory of breast aner has been a subjet of ontroversy. 48 Ultrasound magng s urrently a useful adjunt to dstngush ysts from other masses, however, there s no non-nvasve magng modalty for relably dstngushng bengn and malgnant tumors. 49 Near nfrared magng has the potental to derease the number of unneessary bopses, by dstngushng tumors based on hemoglobn related parameters, as well as to mprove early deteton and dagnoss of breast dsease n younger women. Further, there may be a potental nhe n followng up women's breast aners to trak the response to treatment, or n stagng the tumor development n order to hoose the optmal treatment. Ths exploratory study s desgned to buld and refne a system whh an be used to address some of these potental lnal applatons..b.3. Quanttatve Hemoglobn Imagng De-oxygenated and oxygenated blood spefally hemoglobn, water, and lpds have dfferent absorpton spetra n the vsble and near-nfrared, allowng separaton of these omponents by measurements at multple wavelengths. 50-52 Tssue s hghly absorbng at vsble wavelengths, however, a low absorpton wndow n the near-nfrared, spefally 700-900 nm, allows for measurements through many entmeters of tssue. Imagng breast tumors wth lght has been performed as far bak as 929. The dffulty, however, s that tssue s hghly satterng at vsble and near-nfrared wavelengths. Therefore, an mage 7