Machine Learning Algorithms for Neuroimaging-based Clinical Trials in Preclinical Alzheimer s Disease

Transcription

1 Machine Learning Algorithms for Neuroimaging-based Clinical Trials in Preclinical Alzheimer s Disease Vamsi K. Ithapu Wisconsin Alzheimer s Disease Research Center University of Wisconsin-Madison April 2, 2017 (BRAIN Initiative Symposium 2017) Learning methods for enrichment 1 / 18

2 A Clinical Trial The work flow Randomized Controlled Trial Study Population (BRAIN Initiative Symposium 2017) Learning methods for enrichment 2 / 18

3 A Clinical Trial The work flow Randomized Controlled Trial At enrollment Study Population Controls Intervention (BRAIN Initiative Symposium 2017) Learning methods for enrichment 2 / 18

4 A Clinical Trial The work flow Randomized Controlled Trial Study Population At enrollment Controls Trial ends Controls Intervention Intervention (BRAIN Initiative Symposium 2017) Learning methods for enrichment 2 / 18

5 A Clinical Trial The work flow Randomized Controlled Trial Study Population At enrollment Controls Trial ends Controls Do Controls differ from interneved Intervention Intervention (BRAIN Initiative Symposium 2017) Learning methods for enrichment 2 / 18

6 Setting up a clinical trial My work Who is participating in the trial? Clinical Trial Enrichment How to differentiate control from intervened? Trial Outcome Design (BRAIN Initiative Symposium 2017) Learning methods for enrichment 3 / 18

7 Setting up a clinical trial My work Who is participating in the trial? Clinical Trial Enrichment How to differentiate control from intervened? Trial Outcome Design (BRAIN Initiative Symposium 2017) Learning methods for enrichment 3 / 18

8 Setting up a clinical trial My work Who is participating in the trial? Clinical Trial Enrichment How to differentiate control from intervened? Trial Outcome Design trials aimed for Alzheimer s Disease (BRAIN Initiative Symposium 2017) Learning methods for enrichment 3 / 18

9 Alzheimer s Disease Destroys memory and cognition Irreversible. Strongest risk factor is age Diagnosis { Age, Family History, Cognitive/Neuropsych/Physical Exams, Brain Scans } (BRAIN Initiative Symposium 2017) Learning methods for enrichment 4 / 18

10 Alzheimer s Disease Destroys memory and cognition Irreversible. Strongest risk factor is age Diagnosis { Age, Family History, Cognitive/Neuropsych/Physical Exams, Brain Scans } (BRAIN Initiative Symposium 2017) Learning methods for enrichment 4 / 18

11 Alzheimer s Disease Destroys memory and cognition Irreversible. Strongest risk factor is age Diagnosis { Age, Family History, Cognitive/Neuropsych/Physical Exams, Brain Scans } Mild Cognitive Impairment (MCI) Normal/Healthy Preclinical Dementia (BRAIN Initiative Symposium 2017) Learning methods for enrichment 4 / 18

12 Landscape of AD Clinical Trials Clinicaltrials.gov lists 485 recruiting studies 225 in US; 147 in Europe; 68 are in Phase III and IV (BRAIN Initiative Symposium 2017) Learning methods for enrichment 5 / 18

13 Landscape of AD Clinical Trials Clinicaltrials.gov lists 485 recruiting studies 225 in US; 147 in Europe; 68 are in Phase III and IV Very little success... more than 550 trials since 2002 (Cummings 2014) (BRAIN Initiative Symposium 2017) Learning methods for enrichment 5 / 18

14 Landscape of AD Clinical Trials Clinicaltrials.gov AD diagnosislists itself 485 is messy recruiting studies 225 in US; Early 147diagnosis in Europe; is much harder 68 are incn Phase vs. III MCI and IV 70% Very little success... more than 550 trials since 2002 (Cummings 2014) (BRAIN Initiative Symposium 2017) Learning methods for enrichment 5 / 18

15 Landscape of AD Clinical Trials Clinicaltrials.gov AD diagnosislists itself 485 is messy recruiting studies 225 in US; Early 147diagnosis in Europe; is much harder 68 are incn Phase vs. III MCI and IV 70% < 20% of MCIs convert to AD Very little success... more than 550 trials since 2002 (Cummings 2014) = 8 out of 10 trial subjects are not-eligible!! (BRAIN Initiative Symposium 2017) Learning methods for enrichment 5 / 18

16 ... but there is light Imaging to the rescue Cognitive decline follows atypical brain scans (BRAIN Initiative Symposium 2017) Learning methods for enrichment 6 / 18

17 ... but there is light Imaging to the rescue Cognitive decline follows atypical brain scans Risk Factors Clinical Markers Alzheimer s Disease High Dimensional Imaging Machine Learning Optimal Enricher Design (BRAIN Initiative Symposium 2017) Learning methods for enrichment 6 / 18

18 Population enrichment Healthy Discarded Enrichment Cut-Off Included AD Worsening Disease (Enrichment Criterion) (BRAIN Initiative Symposium 2017) Learning methods for enrichment 7 / 18

19 Population enrichment Healthy Discarded Enrichment Cut-Off Included AD Worsening Disease (Enrichment Criterion) Good enrichment criterion High correlation with disease Practical enrichment criterion High predictive power (BRAIN Initiative Symposium 2017) Learning methods for enrichment 7 / 18

20 Designing a good enricher Given some marker δ : Longitudinal change σ : Pooled Variance (BRAIN Initiative Symposium 2017) Learning methods for enrichment 8 / 18

21 Designing a good enricher Optimal Enricher Given some marker δ : Longitudinal change σ : Pooled Variance Small σ + Large δ (BRAIN Initiative Symposium 2017) Learning methods for enrichment 8 / 18

22 Designing a good enricher Optimal Enricher Given some marker δ : Longitudinal change σ : Pooled Variance Low-Variance + Un-Biased An Ensemble + Neural Networks (BRAIN Initiative Symposium 2017) Learning methods for enrichment 8 / 18

23 Randomized deep networks for enrichment (BRAIN Initiative Symposium 2017) Learning methods for enrichment 9 / 18

24 Randomized deep networks for enrichment T networks L Layered NN T outputs Block 1 L Layered NN L Layered NN (BRAIN Initiative Symposium 2017) Learning methods for enrichment 9 / 18

25 Randomized deep networks for enrichment T networks L Layered NN T outputs Block 1 L Layered NN L Layered NN Block 2 T networks L Layered NN L Layered NN T outputs L Layered NN (BRAIN Initiative Symposium 2017) Learning methods for enrichment 9 / 18

26 Randomized deep networks for enrichment T networks L Layered NN T outputs Block 1 L Layered NN L Layered NN f 1 Block 2 T networks L Layered NN L Layered NN L Layered NN T outputs f 2 f 1 f B f f f 3 6 2f4 f 5 Block B T networks L Layered NN L Layered NN L Layered NN T outputs f B (BRAIN Initiative Symposium 2017) Learning methods for enrichment 9 / 18

27 Randomized deep networks for enrichment T networks L Layered NN T outputs Block 1 L Layered NN L Layered NN Block 2 T networks L Layered NN T outputs L Layered NN L Layered NN Linear Combination Final Output T networks L Layered NN T outputs Block B L Layered NN L Layered NN (BRAIN Initiative Symposium 2017) Learning methods for enrichment 9 / 18

28 Randomized deep network Markers rdm Training baseline rdm Inputs MRI and PET Images Labels AD 0, healthy 1 (BRAIN Initiative Symposium 2017) Learning methods for enrichment 10 / 18

29 Randomized deep network Markers rdm Training baseline rdm Inputs MRI and PET Images Labels AD 0, healthy 1 rdm at test time Predict on MCI (BRAIN Initiative Symposium 2017) Learning methods for enrichment 10 / 18

30 Randomized deep network Markers rdm Training baseline rdm Inputs MRI and PET Images Labels AD 0, healthy 1 rdm at test time Predict on MCI Choose a cut-off t [0, 1] & filter out subjects with rdm prediction > t (BRAIN Initiative Symposium 2017) Learning methods for enrichment 10 / 18

31 Predictive power of baseline rdm Baseline rdm versus change (12 and 24 months) in outcomes Spearman correlation coefficient (and p-value) Marker 12m 24m MMSE , p = , p = ADAS , p = , p < 10 4 MOCA , p > , p = 10 4 RAVLT , p = , p = PsyMEM , p < , p = HippoVol , p , p 10 4 CDR-SB , p , p 10 4 DXConv 1 > 20, p 10 4 > 20, p ANOVA test results are reported since this variable is categorical (BRAIN Initiative Symposium 2017) Learning methods for enrichment 11 / 18

32 Predictive power of baseline rdm Baseline rdm versus change (12 and 24 months) in outcomes Spearman correlation coefficient (and p-value) Marker 12m 24m MMSE , p = , p = ADAS , p = , p < 10 4 MOCA , p > , p = 10 4 RAVLT , p = , p = PsyMEM , p < , p = HippoVol , p , p 10 4 CDR-SB , p , p 10 4 DXConv 1 > 20, p 10 4 > 20, p ANOVA test results are reported since this variable is categorical (BRAIN Initiative Symposium 2017) Learning methods for enrichment 11 / 18

33 Predictive power of baseline rdm Baseline rdm versus change (12 and 24 months) in outcomes Spearman correlation coefficient (and p-value) Marker 12m 24m MMSE , p = , p = ADAS , p = , p < 10 4 MOCA , p > , p = 10 4 RAVLT , p = , p = PsyMEM , p < , p = HippoVol , p , p 10 4 CDR-SB , p , p 10 4 DXConv 1 > 20, p 10 4 > 20, p ANOVA test results are reported since this variable is categorical (BRAIN Initiative Symposium 2017) Learning methods for enrichment 11 / 18

34 Predictive power of baseline rdm Baseline rdm versus change (12 and 24 months) in outcomes Spearman correlation coefficient (and p-value) Marker 12m 24m MMSE , p = , p = ADAS , p = , p < 10 4 MOCA , p > , p = 10 4 RAVLT , p = , p = PsyMEM , p < , p = HippoVol , p , p 10 4 CDR-SB , p , p 10 4 DXConv 1 > 20, p 10 4 > 20, p ANOVA test results are reported since this variable is categorical (BRAIN Initiative Symposium 2017) Learning methods for enrichment 11 / 18

35 Predictive power of baseline rdm Baseline rdm versus change (12 and 24 months) in outcomes Spearman correlation coefficient (and p-value) Marker 12m 24m MMSE , p = , p = ADAS , p = , p < 10 4 MOCA , p > , p = 10 4 RAVLT Very strong , correlations p = 0.04 across , all markers p = PsyMEM , p < , p = HippoVol , p , p 10 4 CDR-SB , p , p 10 4 DXConv 1 > 20, p 10 4 > 20, p ANOVA test results are reported since this variable is categorical (BRAIN Initiative Symposium 2017) Learning methods for enrichment 11 / 18

36 Predictive power of baseline rdm Mean longitudinal change in MMSE & CDR Important trial outcomes MMSE Change m 24m rdam Cut Off CDR Change m 24m rdam Cut Off (BRAIN Initiative Symposium 2017) Learning methods for enrichment 12 / 18

37 Baseline rdm vs. alternate enrichers Sample sizes per arm 80% power, 25% improvement from treatment Sample Outcome measure enricher MMSE ADAS MOCA RAVLT PsyMEM HipVol CDR-SB DxConv HipVol 500 > > FDG > AV > > FAH 296 > > MKLm rdm MKLm is the current state-of-the-art based on SVMs (BRAIN Initiative Symposium 2017) Learning methods for enrichment 13 / 18

38 Baseline rdm vs. alternate enrichers Sample sizes per arm 80% power, 25% improvement from treatment Sample Outcome measure enricher MMSE ADAS MOCA RAVLT PsyMEM HipVol CDR-SB DxConv HipVol 500 > > FDG > AV > > FAH 296 > > MKLm rdm MKLm is the current state-of-the-art based on SVMs (BRAIN Initiative Symposium 2017) Learning methods for enrichment 13 / 18

39 Baseline rdm vs. alternate enrichers Sample sizes per arm 80% power, 25% improvement from treatment Sample Outcome measure enricher MMSE ADAS MOCA RAVLT PsyMEM HipVol CDR-SB DxConv HipVol 500 > > FDG > AV > > FAH 296 > > MKLm rdm rdm has smallest estimates across all outcomes 3 MKLm is the current state-of-the-art based on SVMs (BRAIN Initiative Symposium 2017) Learning methods for enrichment 14 / 18

40 Baseline rdm vs. alternate enrichers Sample sizes per arm 80% power, 25% improvement from treatment Sample Outcome measure enricher MMSE ADAS MOCA RAVLT PsyMEM HipVol CDR-SB DxConv HipVol 500 > > FDG Baseline rdm can 579 detect >2000weak832 treatment 752 effects AV > > FAH 296 > > MKLm rdm MKLm is the current state-of-the-art based on SVMs (BRAIN Initiative Symposium 2017) Learning methods for enrichment 14 / 18

41 The proposed enricher AD Spectrum Normal/Healthy Preclinical Mild Cognitive Impairment (MCI) Dementia (BRAIN Initiative Symposium 2017) Learning methods for enrichment 15 / 18

42 The proposed enricher AD Spectrum Training Regimes Label 1 Mild Cognitive Impairment (MCI) Label 0 Preclinical Normal/Healthy Dementia (BRAIN Initiative Symposium 2017) Learning methods for enrichment 15 / 18

43 The proposed enricher AD Spectrum Training Regimes Label 1 Mild Cognitive Testing Impairment Regime (MCI) Label 0 Preclinical Normal/Healthy Dementia (BRAIN Initiative Symposium 2017) Learning methods for enrichment 15 / 18

44 The proposed enricher AD Spectrum Training Regimes Label 1 Mild Cognitive Testing Impairment Regime (MCI) Label 0 Preclinical Normal/Healthy Dementia Two Issues (BRAIN Initiative Symposium 2017) Learning methods for enrichment 15 / 18

45 The proposed enricher AD Spectrum Training Regimes Normal/Healthy Label 1 Mild Cognitive Testing Impairment Regime (MCI) Label 0 Preclinical Dementia Two Issues Disease spectrum is continuous Labels somewhat artificial Supervised models are sensitive (BRAIN Initiative Symposium 2017) Learning methods for enrichment 15 / 18

46 The proposed enricher AD Spectrum Training Regimes Normal/Healthy Label 1 Mild Cognitive Testing Impairment Regime (MCI) Label 0 Preclinical Dementia Two Issues Disease spectrum is continuous Labels somewhat artificial Supervised models are sensitive Bio-markers interact differently in preclinical vs. AD (BRAIN Initiative Symposium 2017) Learning methods for enrichment 15 / 18

47 The proposed enricher AD Spectrum Training Regimes Label 1 Mild Cognitive Testing Impairment Regime Label 0 (MCI) Can we instead select subjects without information Preclinical transfer from AD stage? Normal/Healthy Dementia Two Issues Disease spectrum is continuous Labels somewhat artificial Supervised models are sensitive Bio-markers interact differently in preclinical vs. AD (BRAIN Initiative Symposium 2017) Learning methods for enrichment 15 / 18

48 An alternate View Sampling Select atypical subjects The more unique a subject is... the more information they contribute to trial Some typical points also needed (BRAIN Initiative Symposium 2017) Learning methods for enrichment 16 / 18

49 An alternate View Sampling Select atypical subjects The more unique a subject is... the more information they contribute to trial Some typical points also needed AD Imaging Features AD Clinical/Neuropsych Scores (BRAIN Initiative Symposium 2017) Learning methods for enrichment 16 / 18

50 An alternate View Sampling Select atypical subjects The more unique a subject is... the more information they contribute to trial Some typical points also needed Very Rich Block (Hierarchical) Structure AD Imaging Features AD Clinical/Neuropsych Scores (BRAIN Initiative Symposium 2017) Learning methods for enrichment 16 / 18

51 Multiresolution Matrix Factorization Salsa Chutney Ketchup Salad Strawberry Shortcake Pita Chapati Bannock Margarine Saute Limpa (BRAIN Initiative Symposium 2017) Learning methods for enrichment 17 / 18

52 Multiresolution Matrix Factorization Salsa Chutney Ketchup Salad Strawberry Shortcake Pita Chapati Bannock Chapati Chutney Pita Margarine Saute Limpa `=1 Limpa Bannock (BRAIN Initiative Symposium 2017) Learning methods for enrichment 17 / 18

53 Multiresolution Matrix Factorization Salsa Chutney Ketchup Salad Strawberry Shortcake Salsa Saute Salad Pita Chapati Bannock Chapati Chutney Pita Margarine Saute Limpa `=1 `=2 Limpa Bannock (BRAIN Initiative Symposium 2017) Learning methods for enrichment 17 / 18

54 Multiresolution Matrix Factorization Salsa Chutney Ketchup Ketchup Strawberry Salad Strawberry Margarine Shortcake Salsa Saute Salad Pita Chapati Shortcake Bannock Chapati Chutney Pita Margarine Saute `=5 Limpa `=4 `=1 `=2 Limpa Bannock (BRAIN Initiative Symposium 2017) Learning methods for enrichment `=3 17 / 18

55 Multiresolution Matrix Factorization Salsa Salad Chutney Strawberry Ketchup Maximally different from the rest Ketchup Strawberry Margarine Shortcake Salsa Saute Salad Pita Chapati Shortcake Bannock Chapati Chutney Pita Margarine Saute Limpa Limpa Bannock More Outlier-ish (BRAIN Initiative Symposium 2017) Learning methods for enrichment 17 / 18

56 Thank you... Questions? I., V. Singh, O. C. Okonkwo, S. C. Johnson, A predictive multi-modal imaging marker for designing efficient and robust AD clinical trials, Clinical Trials on Alzheimer s Disease (CTAD), 2014 I., V. Singh, S. C. Johnson, Randomized deep learning methods for clinical trial enrichment and design in Alzheimer s disease, Deep Learning for Medical Image Analysis (1st Edition) ISBN: ; Chapter 15 I., V. Singh, O. C. Okonkwo, R. J. Chappell, N. M. Dowling, S. C. Johnson, Imaging based enrichment criteria using deep learning algorithms for efficient clinical trials in MCI, Alzheimer s and Dementia, 2015 I., R. Kondor, S. C. Johnson, V. Singh, The Incremental Multiresolution Matrix Factorization Algorithm, Computer Vision and Pattern Recognition (CVPR), Acknowledgements: Vikas Singh, Sterling Johnson, Chris Hinrichs, Risi Kondor, Barbara Bendlin, Ozioma Okonkwa NIH AG040396, NSF CAREER , NSF CCF , UW ADRC AG (BRAIN Initiative Symposium 2017) Learning methods for enrichment 18 / 18