Assessing Tissue Oxygenation and Predicting Tissue Outcome in Acute Stroke

Transcription

1 The Danish National Research Foundation s Center of Functionally Integrative Neuroscience Aarhus University / Aarhus University Hospital - DENMARK Assessing Tissue Oxygenation and Predicting Tissue Outcome in Acute Stroke Kim Mouridsen Associate Professor, PhD Head, Neuroinformatics Research Center for Functionally Integrative Neuroscience Aarhus University Hospital

2 Acute Stroke At 5 hours... After 5 days... Within 3 hours: Intravenous rtpa to dissolve blood clots in Ischemic strokes Risk of causing intracranial Hemorrhage!

3 Acute Stroke Size of problem die annually in EU increasing survive, many with severe disabilities Major socioeconomic burden estimated 34 Billion Diagnostic challenges Automated analysis of diagnostic imaging Rapid integration of imaging and clinical information Access to diagnostic expertise Therapeutical challenges Better drugs faster and cheaper development Use test data in an efficient way Individualize therapy

4 Acute Stroke

5 Mismatch Darby et al Stroke. 30: (1999) Mismatch T2 DWI MTT 5-day followup DWI 90 % risk of infarct growth

6 Powers, Ann. Neurol CBF decrease Perfusion MRI in Acute Stroke CBV decrease MTT increase MTT CBF CBV

7 Residue function R (t) describes tracer wash-out R(0)=1 Tracer Kinetics

8 Singular Value Decomposition

9 Oxygen Extraction and Metabolism B A A: Autoregulation. Precapillary vasodilation B: Clinical symptoms of brain dysfunction Powers, Ann. Neurol. 1991

10 Capillary Transit Time and Oxygen Extraction

11 Capillary Transit Time and Oxygen Extraction v1 v2 vn

12 Capillary Transit Time and Oxygen Extraction

13 Capillary Transit Time Heterogeneity Capillary Transit Time and Oxygen Extraction Vasodilation (flow increase) The Danish National Research Foundation s Center of Functionally Integrative Neuroscience Aarhus University / Aarhus University Hospital Jespersen, Østergaard. JCBFM 2011

14 Capillary Transit Time Heterogeneity Capillary Transit Time and Oxygen Extraction Flow increase can reduce CMRO2 Vasodilation (flow increase) The Danish National Research Foundation s Center of Functionally Integrative Neuroscience Aarhus University / Aarhus University Hospital Jespersen, Østergaard. JCBFM 2011

15 Vascular Model h(t) t Capillary network Vascular model Transport function

16 Vascular Model h(t) t Capillary network Vascular model Transport function

17 Vascular Model h(t) t Capillary network Vascular model Transport function R(t) Observed concentration: Bayesian estimation Residue function t

18 Estimation Results!"#$%&'()*+,-%$./%&0(/$1(-2+,-%$ Mouridsen et al, NeuroImage 2006

19 Digital Phantom OEC values OEC values in digital phantom Transit Time Heterogeneity 1 0 Transit Time Heterogeneity 1 0 Mean Transit Time Mean Transit Time

20 Digital Phantom Each field 14x14 voxels 2 2 A concentramon curve is simulated with noise according to the model for each voxel MTT = 2, 5,,20 sec TTH = 2, 5,, 20 sec SNR = 100, 40, 15 CBV=4% Transit Time Heterogeneity (MTT, TTH) = (8, 17) OEC=42% 1 0 Mean Transit Time

21 Results EsMmaMng Transit Time DistribuMon MTT TTH OEC True TTH TTH TTH MTT MTT MTT Es8mated TTH TTH TTH MTT MTT Low within- field variamon in MTT, TTH and OEC esmmates. Low bias except for combinamon of low MTT and high TTH (SNR=100) Mouridsen et al, ISMRM 2011 MTT

22 OEC EsMmaMon at Different SNRs SNR = 100 SNR = 40 SNR = 15 Phantom EsMmated OEC EsMmated OEC EsMmated OEC TTH TTH TTH MTT MTT MTT True OEC True OEC True OEC Mouridsen et al, ISMRM 2011

23 CTTH in Acute Stroke MTT CTTH OEF Follow- up 20

24 CTTH in Acute Stroke MTT CTTH OEF Follow- up 21

25 CTTH in Acute Stroke MTT CTTH OEF Follow- up 22

26 Example Glioblastoma

27 Salvageable tissue Perfusion - Diffusion Basic prediction of outcome: Identify mismatch between PWI and DWI

28 Perfusion Lesion Identification Thresholding Isolated hyperintensities Ventricles Wholes in infarct Brain boundary (movement) Mean Transit Time

29 Active contours

30 Active contours

31 Change in topology

32 Level Sets Level Set function φ : R 2 R φ(x, y) =0 N Gradient φ = φ x, φ y. φ(x, y) < 0 φ(x, y) > 0 Level set equation dφ dt + F φ =0

33 Contour Evolution

34 Sensitivity to Initial Estimate

35 Optimization E(Γ) = + Inside Γ Outside Γ (u(x, y) C 1 ) 2 dxdy (u(x, y) C 2 ) 2 dxdy Minimize E(Γ) over over smooth contours Γ

36 Optimization E(Γ) = + Inside Γ Outside Γ (u(x, y) C 1 ) 2 dxdy (u(x, y) C 2 ) 2 dxdy Minimize E(Γ) over over smooth contours Γ Γ C 1 = C 2 = Γ C 1 = C 2 = E(Γ) > 0 E(Γ) =0

37 Example

38 Ischemic Core and Follow-Up Thresholding is insufficient

39 DWI Examples Thresholding Gen. Morphology

40 Validation Study I 14 stroke patients MTT calculated using osvd Manual perfusion lesion outline by 4 neuroradiologists Assess agreement with expert consensus

41 Agreement with Expert Consensus (a) (b) (c) (d) Nr Experts (e) (f) (g) (h) (i) (j) (k) (l)

42 Agreement with Expert Consensus

43 Validation Study II 120 acute stroke patients UK, Germany, France, Spain, Denmark Comparison with expert manual outlining

44 Penumbra R 2 =0.86 R 2 =0.63 Courtesy Kartheeban Nagenthiraja

45 Validation Study III Applies tuning parameters from Validation Study II 168 acute stroke patients UK, Germany, France, Spain, Denmark Manual penumbra outlining by 3 radiologists, 1 neuroradiologists

46 TTP Volumetric comparison, APS Volumetric comparison, STM 400 A 400 B APS mask volume [ml] STM mask volume [ml] Manual mask volume [ml] Manual mask volume [ml] R 2 =0.93 R 2 =0.89 Courtesy Kartheeban Nagenthiraja

47 DWI Volumetric comparison of DWI lesions 100 Volumetric comparison of DWI lesions 100 APS mask volume [ml] STM mask volume [ml] Difference Manual mask volume [ml] Manual mask volume [ml] 1 R 2 =0.95 R 2 =0.80 Courtesy Kartheeban Nagenthiraja

48 Penumbra Volumetric comparison of penumbra Volumetric comparison of penumbra APS mask volume [ml] Penumbra volume manual [ml] STM mask volume [ml] Penumbra volume manual [ml] R 2 =0.88 R 2 =0.67 Courtesy Kartheeban Nagenthiraja

49 Examples Courtesy Kartheeban Nagenthiraja

50 Flaskehalsen: For mange data 1400 images 2 minutes Experience from 1500 patients The Danish National Research Foundation s Center of Functionally Integrative Neuroscience Aarhus University / Aarhus University Hospital

51

52 Predicting Outcome in Acute Stroke Acute and final lesion volume Acute lesion volume and follow up neurological score Acute perfusion Outcome Outcome score (mrs) Acute perfusion Outcome Acute perfusion lesion volume w w w. i - k n o w - s t r o k e. e u

53 Predicting Outcome in Acute Stroke Acute and final lesion volume Acute lesion volume and follow up neurological score Acute perfusion Outcome Outcome score (mrs) Acute perfusion Outcome Acute perfusion lesion volume Acute perfusion lesion volume does not predict final lesion w volume w w. i - k n or o w neurological - s t r o k e. e u outcome score

54 Predicting Tissue Outcome after Stroke Imaging Clinical Genetic Database Biochemical Epidemiological Follow up data Acute Patient Data Imaging Clinical Genetic Biochemical Epidemiological Patient Outcome Clinical score Lesion volume w w w. i - k n o w - s t r o k e. e u

55 Predicting Tissue Outcome after Stroke Imaging Clinical Genetic Database Biochemical Epidemiological Follow up data Acute Patient Data Imaging Clinical Genetic Biochemical Epidemiological Voxel-wise infarct probability w w w. i - k n o w - s t r o k e. e u

56 Predicting Tissue Outcome after Stroke Imaging Clinical Genetic Database Biochemical Epidemiological Follow up data > Data points < 50 Data points Acute Patient Data Imaging Clinical Genetic Biochemical Epidemiological Voxel-wise infarct probability w w w. i - k n o w - s t r o k e. e u

57 Computer model of Tissue Infarction T2 DWI CBV Risk Images CBF MTT Heterogenity Risk factors Duration of symptoms Diabetes? Hypertension? Fever? etc Available in patient records The Danish National Research Foundation s Center of Functionally Integrative Neuroscience Aarhus University / Aarhus University Hospital

58 Computer model of Tissue Infarction T2 DWI CBV Risk Images CBF MTT Heterogenity Risk factors Duration of symptoms Diabetes? Hypertension? Fever? etc Available in patient records The Danish National Research Foundation s Center of Functionally Integrative Neuroscience Aarhus University / Aarhus University Hospital

59 Stroke Progression Model DATABASE MODEL CBF DWI MTT CBF DWI MTT ACUTE IMAGES 3 MONTHS FOLLOW-UP PREDICTED RISK 1.0 CBF DWI MTT RISK MAP 0.0

60 Modeling Class Boundary

61 Modeling Class Boundary

62 Outcome Predicted with DWI Risk of Infarct Risk Observed Model (DWI) N=24 non-reperfusers DWI

63 Outcome Predicted with DWI Modeled and Actual Risk of Infarct Risk Observed Model (DWI) N=24 non-reperfusers Mean Transit Time

64 Outcome Predicted with DWI Modeled and Actual Risk of Infarct Risk Observed Model (DWI) 50% actual difference 10% predicted difference N=24 non-reperfusers Mean Transit Time

65 Outcome Predicted with DWI & MTT Modeled and Actual Risk of Infarct Risk Observed Model (DWI, MTT) N=24 non-reperfusers Mean Transit Time

66 Gain in risk prediction Reperfusers Modeled and Actual Risk of Infarct Modeled and Actual Risk of Infarct Risk Observed Model (DWI) Risk N=32 N= Observed Model (DWI, MTT) Mean Transit Time Mean Transit Time

67 Multimodal Predictive Model 0.85 Model performance AUC DWI TTP MTT CBF CBV OEF ADC LOWB FH Courtesy Kartheeban Nagenthiraja

68 Longitudinal Modeling - Risk Progression No reperfusion 2 hours 24 hours 3 months Reperfusion 2 hours 24 hours 3 months Jonsdodr, Hjort, Østergaard, Mouridsen. Proc ISMRM 2008

69 IdenMfy groups of paments with similar dynamic characterismcs Inter- pa8ent similarity in infarct evolu8on Arbitrary Units M M M R R R R Individual Subject Models R R Non- reperfuser M No acute mismatch Arbitrary Units Mouridsen et al. Proc Intl. Stroke Conf. 2009

70 IdenMfy groups of paments with similar dynamic characterismcs Inter- pa8ent similarity in infarct evolu8on Arbitrary Units M M M R R R R Individual Subject Models R R Non- reperfuser M No acute mismatch Arbitrary Units Mouridsen et al. Proc Intl. Stroke Conf. 2009

71 IdenMfy groups of paments with similar dynamic characterismcs Inter- pa8ent similarity in infarct evolu8on Arbitrary Units M M M R R R R Individual Subject Models R R Non- reperfuser M No acute mismatch Arbitrary Units Mouridsen et al. Proc Intl. Stroke Conf. 2009

72 Risk Progression Profiles High risk NormalizaMon No reperfusion Increasing risk! Low risk Reperfusion Automa)cally iden)fy clinically relevant characteris)cs Differen)ate regions of risk increase from risk decrease by correla)on with clinical informa)on Jonsdo&r, Hjort, Østergaard, Mouridsen. Proc ISMRM 2008

73 RelaMve Importance of Risk Markers DWI odds ramo changed from 3.1 to 8.4 in the 0 24h interval Perfusion parameter odds ramos decrease over Mme (CBF, MTT, Delay) w w w. i - k n o w - s t r o k e. e u Jonsdodr, Hjort, Østergaard, Mouridsen. Proc ISMRM 2008

74 OEF biomarker offers better prediction of outcome than existing perfusion metrics in acute stroke The Danish National Research Foundation s Center of Functionally Integrative Neuroscience Aarhus University / Aarhus University Hospital

75 Can we identify active drugs in a Neuroinformatics: Standard treatment more cost-effective way? Does a new drug candidate have beneficial effects in humans? Current method (low sensitivity): N1 patients N2 patients 100 % 50 % Does it reduce fatalities and severe disability? More stroke victims live independently? Proof requires over 1500 patienter and time Phase III test of therapy: mio DKK 10 % Reduction of risk for loss of brain tissue in individual patient because of new drug Proof in only 30 patients Cost-effective test of new therapies Test-data used in subsequent product phase Individualized treatment The Danish National Research Foundation s Center of Functionally Integrative Neuroscience Aarhus University / Aarhus University Hospital Wu, Sumii, Asahi, et al. JCBFM 27: (2007)

76 Untreated DWI < PWI T2 DWI ADC CBF CBV MTT Delay 71 y.o. male Time to MR 3.4 h Wu, et al. ISMRM. 2004

77 Untreated DWI < PWI T2 DWI ADC Predicted CBF CBV MTT Delay 71 y.o. male Time to MR 3.4 h Wu, et al. ISMRM. 2004

78 Untreated DWI < PWI T2 DWI ADC Predicted CBF CBV MTT Delay 71 y.o. male Time to MR 3.4 h 7-day Follow-up Wu, et al. ISMRM. 2004

79 Treated DWI < PWI T2 DWI ADC CBF CBV MTT Delay 62 y.o. male imaged at 3.5 h Wu, et al. ISMRM. 2004

80 Treated DWI < PWI T2 DWI ADC Predicted CBF CBV MTT Delay 62 y.o. male imaged at 3.5 h Wu, et al. ISMRM. 2004

81 Treated DWI < PWI T2 DWI ADC Predicted CBF CBV MTT Delay 62 y.o. male imaged at 3.5 h 7-day Follow-up Wu, et al. ISMRM. 2004

82 Summary Metabolic parameters and capillary distribution characteristics can be obtained with DSC-MRI Viable tissue can be identified based on PWI/DWI mismatch MRI modalities can be combined into one risk image, which substantially improves estimates of infarct risk Predictive models can be used to estimate tissue response to treatment