- Email: [email protected]
Autocovariance based analysis of functional MRI data
640
Abstracts
moL PSYCIUATRY
1996:39:500-666
477. AUTOMATED MEASUREMENT OF SUBCORTICAL BRAIN MR STRUCTURES IN SCHIZOPHRENIA D.V. losifescu I, M.E. Shenton l , R. Kikinis 2 , J. DengJer2, S.K. Warfield 2, & R.W. McCarley I IClinical Neuro~cience Division. Laboralory of Neuroscience. Brocklon VA Medical Center, Department of Psychiatry; 2S urgical Planning Laboratory, MRI Division, Dllpanment of Radiology. Harvard Mctlic:lI School The possibility of detecting very small volume and shape abnomt.alities on MR biJin images is currently burdened by the laborious task of manually tr.lcing bmln struClW'e contours. Ela.~tic malching. an nutomated MR image processing technique. provides not only an exception:!l increuse in speed (hundreds of an:nocnic regions can be measured in a short period of lime) but also completc reproducibility. We used elaslic matching to warp an MR brain alia... onto 28 new MR brain images from 14 schizophrenia patienls and 14 110nn:ll controls, m:ltchcd for age, gender and ham.lt:Jness. We uppiled two successive image processing algorithms for each new brain image. First. :I rigid body transformalion removed tr.1nslational. rotational nnd scale differences between the atlas and the new image. Second. clasI:c matchinG corrected Ihe local shape differences by m:llehing the edges of the white and gray maltcr surf:lccs. This n:sulled in uutomalic identificalion of all the alias anatomical structures on Ihe new brain image. To a.c;sess accumey in measurcmenl, we compared the thal:lmic and the busnl ganglia volumes, measured with clastic malching. with the same volumes traced m:munl1y on the MR images. The accuracy of the match. measured as the percent difference between matched and true volumes. avcr.lged 98% for gray and white malter volumes. 94% for thalamus and 91 % for ba...al ganglia. Using ela.c;lic mmching. we found un 11 % increase in MR volume for the bil~i:ll ganglin in schizophrenics but no difference in thalamic MR volume compared with nornla) controls. which reproduces previous results in our lubor.Jtory.
478. CLUSTER ANALYSIS OF MRI REGIONAL BRAIN VOLUMES IN SCHIZOPHRENIA D.H. Mathalon, L. Marsh, K.O. Lim, & A. Pfefferbaum Depurtmcnl of (l.~ychiatry. Stanford University School of Medicine and VA Palo Aho Health Cure System, Palo Alto, CA 94304 We report an approach to identify biologically homogeneous subgroups of schizophrenia lhat may reflect distinct pathophysiological entities nnd account for lhe clinical heterogeneity. We oblained rcgionnl bmln volumes with MRI in 78 men with DSM-IIIR schizophrenill lind performed two clust!!r analyses (Ward's method). In Ihe first analysis. 78 schizophrenics were clustered into two subgroups based 011 the extent of deficits in six axi:1I MRI cortical gray mailer regions (prefrontal. frontal. frontul-tempoml, Icmpoml·parielnl, p:ltiClal·occipital). The hij;h-d.:::fict! cluster (n=46) consistently had more s.cvcrc gruy mailer deficits than the low-deficit cluster (n=32) across nil of the cortical regions (nIl p<.OOO2), nnd also showed worse BPRS scores on somatic concern. emotional withdrawal, and IIncoopcralivcnes5, Agr. and clinical c:ours~ did not distinguish lhese c1ustm. In the second analysis, a subset or 39 schizophre!1ics wilh coronal and llxial MRls were clustered based on the
axial regions and coronal volumcs of left and right temporal lobe gmy mailer. hippocampus tissue. lemporal hom and third ventricle. Before clustering, the regions were reduced to 6 orthogonal faclors based 011 a principal components analysis: Venlocular System, Temporal Horns, Temporal Lobe Gray, Hippocampus, Frontal-Temporal-Parictal Gray, and Prefrontal.Occipila! Gmy. The 39 schizophrenics were clustelcd into two subgroups based on dislingu~hable pallcms llnd extent of volume abnonnalhics. The high conical-low temporal cluster (n""21) showed significantly grealer deficits in the two COl1ical Gray faclors but smaller deficilS in the Temporal Oray and Tempoml Hom factoff, liS well as being signilicantly older and showing worse BPRS scores on somatic concern, mannerisms llnd posluring. 1lu;~ low cortical·high tempornl cluster (n"" 18) showedlhe opposile pattern, These clusters did not differ in the Ventricular System or the Hippocampas. These results provide preliminary evidence for neuroanatomical subgroups in schizophrenia based on severit}' and pnllcm of regional involvement. Supported by MH 30854. Department of Veterans Affairs
479. AUTOCOVARIANCE BASED ANALYSIS OF FUNCTIONAL MRI DATA
L.e. Maas·,2.3, B.deB. Frederick l ,:!, D.A. Yurgelun-Todd 1,2, & P.F. Renshaw 1•2 'Brain Imaging Center, Mclean Hospit.at. Belmont. MA; 2Consolidatcd Department of Psychiatry. Harvard Medical School, Boston, MA; JHarvanl University· Massnchusetts Instilute of Technology Division of Health Scicnces ami Technology, Cambridge. MA Functional magntltic resonance imaging (fMRI) is nn important 1001 for the lloninvasivc mapping of bmht llctivity. In blood oxygenalion level dependent (BOLD) IMRI, post-processing of tMRI data sets is required to Identify activated regions wilhout thc need for Cx.0benous contrast ngllnts or radiotroc:c~. Most currcnlllpplicalions employ difference Hest. cOlTClalion coefficient. or Fourier methods to generate activation mllps. However, the sensitivity of difference and correlation coefficient methods is compromised by spatially varying hemouynilmic clclays :md riSe or fnll limes associated with activation. and both methods require a priori assumptions about these characteristics, Furthennore. quantification of :lctivation energy by Fourier methods is difficult due to the windowing effects of finite data Sel length and the harmonics introduced by non-sinusoidaillclivation proliles. We have developed 11 method for the detcction and quantification of activation in BOLD fMRI data sets based on pixel-wise autocovariance analysis of the time series data. For a periodic experimental stimulus, any activation profile with the same periodicily as the slimulus can be detected and quantified without a priori knowledge of its shape or hemodynamic delays. The SCIlsitivity of Ihe ACV lechniquc is not reduced by spatilll variation of activation characteristics. and energy quantification is not complic:lted by signal shape or sampling effects. Additionally. a measure of the activation contrnsHonoise ratio can be extracted, and the statistical significance of llctivation Clln be assessed, To demonstrnte the advantages of this ACV ba.c;cd method, two seLe; of data wcre collected in a human pholic stimulation study with alternating Iiglll and dark intervals lasting 5 or 10 seconds each. As the stimulus nnd rest intervals were shol1ened, 11 reduced number of activated pixels wns detected with both correlation cocfncient (125 pixels for lOs vs. 85 pixels for 55) nnd t·test (127 vs. 90) rncthods, whereas ACV analysis, whose sClIsilivily increases with the number of experimental cycles studied, delected additional activated pixels (124 vs. 135). Additionally, the three comparison methods yielded lower esli· miltes of activation magnitUde at individUll1 pbtcls than ACV energyba!\cd qutll1tificalion, As the limitations described above for Ihese
Abstmcts
comparison methods tend to lead to underestimation, this finding suggests that ACV analysis may more accur:ucty qu:mtitutc activation.
480. DECOUPLED ROTATIONAL AND TRANSLATIONAL IMAGE REGISTKATION FOR FUNCTIONAL MRI DATA SETS L.C. Maas 1•2 •3, a.deB. Frederick l •2, & P.F. Renshaw l •2 IBrain Imaging Center. McLean Hospital, Belmont. MA; 'Consolidated Department of Psychiatry. HllrVard Medical School, Boston, MA: 3Harvanl University - Massachuselts Inslitute of Technology Division of HCllhh Sciences and Technology, Cambridge. MA Functional magnctic resonance imaging (fMRl) is an impommt tool for lhe noninvasive mapping of human brain activity, as tlctivlltion maps can be geneI'llted rapidly, without the use of radiotrllcers. Howevcr, accurate posl-processing of these data sets is required to generate usefuillctivation maps, In echo planar IMRI, fmmc-lo-fmme subject motion during the acquisition of the image sct C3n IC3d to significant artifacts during post-processing. Motion correl3ted with the presentation of experimental stimuli or the performance ofmsks is nn especially important problem. as it may lead to false activation artifacts. While a variety of imllge registl'lltion algorithms exist for the correction of frame·lo·frame motion, the large number of images in typical lMRI dilta scts make mcthods requiring the identificlllion of landmarks or surfllces imprnctical. In methods where an error measure is iteratively minimized. careflll choices of inhilll parometers arc required for convergence. Bnd convergence time Cllnnot be predicted. Typically, these methods simultaneously estimate rotational nnd translational motion, increasing the complexity of the estimation. We have developed an in-plane frame-to·frame inmge registration algorithm suited to the difficultics of large fMRI data sets. By decoupling the estimation of rotational and translational motion. fast non-iterativc methods can be u!'>ed to determine corrcction panlmcters sequentiillly. The dccoupling is effected by alllllyzing dnta in the k space domain, where rotationnl and translational motion estimates are com· puted from magnitude 3nd phase information. respectively, Furthennore, registered images llrc genemted using frequency regridding for rotational offsets. lind phuse plane conjugation for translalional offset<;, resulting in less error introduced into the registered images thi:ln with the use of conventional bilinear interpolation. Echo planar phantom image data demonstrate the accuracy of the algorithm (rotation: -0.09 0 ± 0.11 0 , translation: ·0.035 :t 0.054 pixels [0.055 ± 0.084 mOl)). Additionally, dlltll from human brain Ilctivation studies arc presenled to demonstmte the utility of this method in the pre-processing of rMRI data selS.
48 I. COCAINE REDUCES CEREBRAL BLOOD VOLUME IN HUMANS: DETECTION WITH DSC MRI
M.l Kaufman I , J.M. Levin,,2, J.D. Christensen 2, S.E. Lukas l , J,H. Mendelson l , L C. Maas2 , ., 2 2 S.L. Rose-, B.M. Cohen, & P.F. Renshaw I Alcohol and Drog Abuse Research Center and :lBrain Imaging Center, McLean Hospital, Harvard Medicat School. Delmont. MA 02178
Coclline is a vasoactive drug, and its usc has been associated with both trllnsJcnt and long lasting changes in cerebral perfusion. The present
BIOI. PSYCHIATRY
64[
1996;39:500-666
study utilized dynamic susceptibility contra!\! magnetic resonance imilging (DSC MRI) (Magn. Reson. Med. 34:655, 1995) to examine the effects of acute cocnine administration on cercbrnl blood volume (CBY). DSC MRI provides superior sputial and temporal resolution for evaluation of cerebml hemodynamics compared with either PET or SPECf. Ten healthy
482. DYNAMIC CONTRAST MRI OF CEREBRAL BLOOD VOLUME COMPARED WITH SPECT IN ALZHEMER'S DISEASE OJ. Harris l , R.F. Lewis I, A. SatIin 2 , C.D. English 2 , T.M. Scotti, D,A. Yurgelun-Todd 2, & P.F. Renshaw 2 INcuroimaging Research Laboratory, Department of Psychiatry. New England Medical Center. Boston, MA 02111: :'!Brain Imaging Center and Memory Diagnostic Clinic. McLean Hospital. Belmont. MA 02178 Dynamic susceptibility conlrllst magnetic resonance imaging (DSC MRI) is a new method for viewing cerebral blood volume that uses fast acquisition echo-planar MRI with a non-radioactive contrast agent to gcneriltc high resolution 0-2mm) functional images.The purpose of this study was to investigate tile potential effectiveness of DSC MRI to discriminate Alzheimer's disease (AD) patients from elderly compari!'>on subjects. ond to compare results obtained with DSC MRI and single photon emission computed tomography (SPECJ'). using similar image analysis tcchniques for both modalities. Regional cerebral blood volumc (rCBY) imogcs were generated from echo planar mllgnetic resonance imaging using the dynamic susceptibilily contrnst (DSC MRI) method in thirteen Alzheimer's disease patienls lind lhirteen comparison subjects group·matched on age and gender. Tempore-parietal blood volume nonnalizcd to ccrclx!lIum was reduced 17% bilaterally in AD. Sensorymotor regional rCBV WIlS reduced only 8.5% in AD. With SPEcr imaging in 20 AD palients lind 8 controls. results were nearly idcntical to those obtained wilh the DSC MRI blood volume imaging: mcthod: tcmporo-pal'ielal ~1"rllsioll 3S u pcn:cntayc of cerebellar blood flow WIlS reduced 18% on averllge (compared with 17% reductions measured using DSC MRI). and sensory· motor perfusion WllS 8.5% below that of the comparison group (also 8.5% dericit with DSC MRt). Discriminnnt function analysis (89% correct group cl:lssifieation) and t-test compari. son signlncance levels were similar using DSC MRI or SPEeT. DSC MRJ Is promising as tI non-radi03ctive. lower cost alternative to other functional neurolmaging methods for evaluoting Alzheimer's disease.
Abstracts
moL PSYCIUATRY
1996:39:500-666
477. AUTOMATED MEASUREMENT OF SUBCORTICAL BRAIN MR STRUCTURES IN SCHIZOPHRENIA D.V. losifescu I, M.E. Shenton l , R. Kikinis 2 , J. DengJer2, S.K. Warfield 2, & R.W. McCarley I IClinical Neuro~cience Division. Laboralory of Neuroscience. Brocklon VA Medical Center, Department of Psychiatry; 2S urgical Planning Laboratory, MRI Division, Dllpanment of Radiology. Harvard Mctlic:lI School The possibility of detecting very small volume and shape abnomt.alities on MR biJin images is currently burdened by the laborious task of manually tr.lcing bmln struClW'e contours. Ela.~tic malching. an nutomated MR image processing technique. provides not only an exception:!l increuse in speed (hundreds of an:nocnic regions can be measured in a short period of lime) but also completc reproducibility. We used elaslic matching to warp an MR brain alia... onto 28 new MR brain images from 14 schizophrenia patienls and 14 110nn:ll controls, m:ltchcd for age, gender and ham.lt:Jness. We uppiled two successive image processing algorithms for each new brain image. First. :I rigid body transformalion removed tr.1nslational. rotational nnd scale differences between the atlas and the new image. Second. clasI:c matchinG corrected Ihe local shape differences by m:llehing the edges of the white and gray maltcr surf:lccs. This n:sulled in uutomalic identificalion of all the alias anatomical structures on Ihe new brain image. To a.c;sess accumey in measurcmenl, we compared the thal:lmic and the busnl ganglia volumes, measured with clastic malching. with the same volumes traced m:munl1y on the MR images. The accuracy of the match. measured as the percent difference between matched and true volumes. avcr.lged 98% for gray and white malter volumes. 94% for thalamus and 91 % for ba...al ganglia. Using ela.c;lic mmching. we found un 11 % increase in MR volume for the bil~i:ll ganglin in schizophrenics but no difference in thalamic MR volume compared with nornla) controls. which reproduces previous results in our lubor.Jtory.
478. CLUSTER ANALYSIS OF MRI REGIONAL BRAIN VOLUMES IN SCHIZOPHRENIA D.H. Mathalon, L. Marsh, K.O. Lim, & A. Pfefferbaum Depurtmcnl of (l.~ychiatry. Stanford University School of Medicine and VA Palo Aho Health Cure System, Palo Alto, CA 94304 We report an approach to identify biologically homogeneous subgroups of schizophrenia lhat may reflect distinct pathophysiological entities nnd account for lhe clinical heterogeneity. We oblained rcgionnl bmln volumes with MRI in 78 men with DSM-IIIR schizophrenill lind performed two clust!!r analyses (Ward's method). In Ihe first analysis. 78 schizophrenics were clustered into two subgroups based 011 the extent of deficits in six axi:1I MRI cortical gray mailer regions (prefrontal. frontal. frontul-tempoml, Icmpoml·parielnl, p:ltiClal·occipital). The hij;h-d.:::fict! cluster (n=46) consistently had more s.cvcrc gruy mailer deficits than the low-deficit cluster (n=32) across nil of the cortical regions (nIl p<.OOO2), nnd also showed worse BPRS scores on somatic concern. emotional withdrawal, and IIncoopcralivcnes5, Agr. and clinical c:ours~ did not distinguish lhese c1ustm. In the second analysis, a subset or 39 schizophre!1ics wilh coronal and llxial MRls were clustered based on the
axial regions and coronal volumcs of left and right temporal lobe gmy mailer. hippocampus tissue. lemporal hom and third ventricle. Before clustering, the regions were reduced to 6 orthogonal faclors based 011 a principal components analysis: Venlocular System, Temporal Horns, Temporal Lobe Gray, Hippocampus, Frontal-Temporal-Parictal Gray, and Prefrontal.Occipila! Gmy. The 39 schizophrenics were clustelcd into two subgroups based on dislingu~hable pallcms llnd extent of volume abnonnalhics. The high conical-low temporal cluster (n""21) showed significantly grealer deficits in the two COl1ical Gray faclors but smaller deficilS in the Temporal Oray and Tempoml Hom factoff, liS well as being signilicantly older and showing worse BPRS scores on somatic concern, mannerisms llnd posluring. 1lu;~ low cortical·high tempornl cluster (n"" 18) showedlhe opposile pattern, These clusters did not differ in the Ventricular System or the Hippocampas. These results provide preliminary evidence for neuroanatomical subgroups in schizophrenia based on severit}' and pnllcm of regional involvement. Supported by MH 30854. Department of Veterans Affairs
479. AUTOCOVARIANCE BASED ANALYSIS OF FUNCTIONAL MRI DATA
L.e. Maas·,2.3, B.deB. Frederick l ,:!, D.A. Yurgelun-Todd 1,2, & P.F. Renshaw 1•2 'Brain Imaging Center, Mclean Hospit.at. Belmont. MA; 2Consolidatcd Department of Psychiatry. Harvard Medical School, Boston, MA; JHarvanl University· Massnchusetts Instilute of Technology Division of Health Scicnces ami Technology, Cambridge. MA Functional magntltic resonance imaging (fMRI) is nn important 1001 for the lloninvasivc mapping of bmht llctivity. In blood oxygenalion level dependent (BOLD) IMRI, post-processing of tMRI data sets is required to Identify activated regions wilhout thc need for Cx.0benous contrast ngllnts or radiotroc:c~. Most currcnlllpplicalions employ difference Hest. cOlTClalion coefficient. or Fourier methods to generate activation mllps. However, the sensitivity of difference and correlation coefficient methods is compromised by spatially varying hemouynilmic clclays :md riSe or fnll limes associated with activation. and both methods require a priori assumptions about these characteristics, Furthennore. quantification of :lctivation energy by Fourier methods is difficult due to the windowing effects of finite data Sel length and the harmonics introduced by non-sinusoidaillclivation proliles. We have developed 11 method for the detcction and quantification of activation in BOLD fMRI data sets based on pixel-wise autocovariance analysis of the time series data. For a periodic experimental stimulus, any activation profile with the same periodicily as the slimulus can be detected and quantified without a priori knowledge of its shape or hemodynamic delays. The SCIlsitivity of Ihe ACV lechniquc is not reduced by spatilll variation of activation characteristics. and energy quantification is not complic:lted by signal shape or sampling effects. Additionally. a measure of the activation contrnsHonoise ratio can be extracted, and the statistical significance of llctivation Clln be assessed, To demonstrnte the advantages of this ACV ba.c;cd method, two seLe; of data wcre collected in a human pholic stimulation study with alternating Iiglll and dark intervals lasting 5 or 10 seconds each. As the stimulus nnd rest intervals were shol1ened, 11 reduced number of activated pixels wns detected with both correlation cocfncient (125 pixels for lOs vs. 85 pixels for 55) nnd t·test (127 vs. 90) rncthods, whereas ACV analysis, whose sClIsilivily increases with the number of experimental cycles studied, delected additional activated pixels (124 vs. 135). Additionally, the three comparison methods yielded lower esli· miltes of activation magnitUde at individUll1 pbtcls than ACV energyba!\cd qutll1tificalion, As the limitations described above for Ihese
Abstmcts
comparison methods tend to lead to underestimation, this finding suggests that ACV analysis may more accur:ucty qu:mtitutc activation.
480. DECOUPLED ROTATIONAL AND TRANSLATIONAL IMAGE REGISTKATION FOR FUNCTIONAL MRI DATA SETS L.C. Maas 1•2 •3, a.deB. Frederick l •2, & P.F. Renshaw l •2 IBrain Imaging Center. McLean Hospital, Belmont. MA; 'Consolidated Department of Psychiatry. HllrVard Medical School, Boston, MA: 3Harvanl University - Massachuselts Inslitute of Technology Division of HCllhh Sciences and Technology, Cambridge. MA Functional magnctic resonance imaging (fMRl) is an impommt tool for lhe noninvasive mapping of human brain activity, as tlctivlltion maps can be geneI'llted rapidly, without the use of radiotrllcers. Howevcr, accurate posl-processing of these data sets is required to generate usefuillctivation maps, In echo planar IMRI, fmmc-lo-fmme subject motion during the acquisition of the image sct C3n IC3d to significant artifacts during post-processing. Motion correl3ted with the presentation of experimental stimuli or the performance ofmsks is nn especially important problem. as it may lead to false activation artifacts. While a variety of imllge registl'lltion algorithms exist for the correction of frame·lo·frame motion, the large number of images in typical lMRI dilta scts make mcthods requiring the identificlllion of landmarks or surfllces imprnctical. In methods where an error measure is iteratively minimized. careflll choices of inhilll parometers arc required for convergence. Bnd convergence time Cllnnot be predicted. Typically, these methods simultaneously estimate rotational nnd translational motion, increasing the complexity of the estimation. We have developed an in-plane frame-to·frame inmge registration algorithm suited to the difficultics of large fMRI data sets. By decoupling the estimation of rotational and translational motion. fast non-iterativc methods can be u!'>ed to determine corrcction panlmcters sequentiillly. The dccoupling is effected by alllllyzing dnta in the k space domain, where rotationnl and translational motion estimates are com· puted from magnitude 3nd phase information. respectively, Furthennore, registered images llrc genemted using frequency regridding for rotational offsets. lind phuse plane conjugation for translalional offset<;, resulting in less error introduced into the registered images thi:ln with the use of conventional bilinear interpolation. Echo planar phantom image data demonstrate the accuracy of the algorithm (rotation: -0.09 0 ± 0.11 0 , translation: ·0.035 :t 0.054 pixels [0.055 ± 0.084 mOl)). Additionally, dlltll from human brain Ilctivation studies arc presenled to demonstmte the utility of this method in the pre-processing of rMRI data selS.
48 I. COCAINE REDUCES CEREBRAL BLOOD VOLUME IN HUMANS: DETECTION WITH DSC MRI
M.l Kaufman I , J.M. Levin,,2, J.D. Christensen 2, S.E. Lukas l , J,H. Mendelson l , L C. Maas2 , ., 2 2 S.L. Rose-, B.M. Cohen, & P.F. Renshaw I Alcohol and Drog Abuse Research Center and :lBrain Imaging Center, McLean Hospital, Harvard Medicat School. Delmont. MA 02178
Coclline is a vasoactive drug, and its usc has been associated with both trllnsJcnt and long lasting changes in cerebral perfusion. The present
BIOI. PSYCHIATRY
64[
1996;39:500-666
study utilized dynamic susceptibility contra!\! magnetic resonance imilging (DSC MRI) (Magn. Reson. Med. 34:655, 1995) to examine the effects of acute cocnine administration on cercbrnl blood volume (CBY). DSC MRI provides superior sputial and temporal resolution for evaluation of cerebml hemodynamics compared with either PET or SPECf. Ten healthy
482. DYNAMIC CONTRAST MRI OF CEREBRAL BLOOD VOLUME COMPARED WITH SPECT IN ALZHEMER'S DISEASE OJ. Harris l , R.F. Lewis I, A. SatIin 2 , C.D. English 2 , T.M. Scotti, D,A. Yurgelun-Todd 2, & P.F. Renshaw 2 INcuroimaging Research Laboratory, Department of Psychiatry. New England Medical Center. Boston, MA 02111: :'!Brain Imaging Center and Memory Diagnostic Clinic. McLean Hospital. Belmont. MA 02178 Dynamic susceptibility conlrllst magnetic resonance imaging (DSC MRI) is a new method for viewing cerebral blood volume that uses fast acquisition echo-planar MRI with a non-radioactive contrast agent to gcneriltc high resolution 0-2mm) functional images.The purpose of this study was to investigate tile potential effectiveness of DSC MRI to discriminate Alzheimer's disease (AD) patients from elderly compari!'>on subjects. ond to compare results obtained with DSC MRI and single photon emission computed tomography (SPECJ'). using similar image analysis tcchniques for both modalities. Regional cerebral blood volumc (rCBY) imogcs were generated from echo planar mllgnetic resonance imaging using the dynamic susceptibilily contrnst (DSC MRI) method in thirteen Alzheimer's disease patienls lind lhirteen comparison subjects group·matched on age and gender. Tempore-parietal blood volume nonnalizcd to ccrclx!lIum was reduced 17% bilaterally in AD. Sensorymotor regional rCBV WIlS reduced only 8.5% in AD. With SPEcr imaging in 20 AD palients lind 8 controls. results were nearly idcntical to those obtained wilh the DSC MRI blood volume imaging: mcthod: tcmporo-pal'ielal ~1"rllsioll 3S u pcn:cntayc of cerebellar blood flow WIlS reduced 18% on averllge (compared with 17% reductions measured using DSC MRI). and sensory· motor perfusion WllS 8.5% below that of the comparison group (also 8.5% dericit with DSC MRt). Discriminnnt function analysis (89% correct group cl:lssifieation) and t-test compari. son signlncance levels were similar using DSC MRI or SPEeT. DSC MRJ Is promising as tI non-radi03ctive. lower cost alternative to other functional neurolmaging methods for evaluoting Alzheimer's disease.