modelling the neuroanatomical progression of alzheimerâ s ... › ~razvan › talk › martinos2019...
TRANSCRIPT
![Page 1: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/1.jpg)
Modelling the Neuroanatomical Progression of Alzheimers Disease andPosterior Cortical Atrophy
Razvan Valentin Marinescu
Medical Vision Group, Massachusetts Institute of Technology
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Introduction 1 / 20
![Page 2: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/2.jpg)
About me
I Grew up in Pitesti, Romania
I 2010-2014: Studied a 4-year MEng in Computer Science at Imperial College London
I 2014-2019: PhD in Medical Imaging at UCL (with Daniel Alexander)
I 2019: Postdoc at MIT with Pollina Golland (working on image analysis of stroke)
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Introduction 2 / 20
![Page 3: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/3.jpg)
Progression of Neurodegenerative Diseases (POND)
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Introduction 3 / 20
![Page 4: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/4.jpg)
POND Aim: Develop Computational Models for Disease Progression
Event-Based Model(Fontejin et al., Neuroimage, 2012)
Differential Equation Model(Oxtoby et al., submitted, 2017)
Gaussian-Process Regression(Lorenzi et al., IPMI, 2015)
Subtype and Stage Inference(Young et al., submitted, 2017)
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Introduction 4 / 20
![Page 5: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/5.jpg)
POND Aim 2: Apply the Models to Distinct Neurodegenerative Diseases
typical AD(Young et al., Nature Comms., 2018)
Familial AD(Oxtoby et al., Brain, 2018)
Multiple sclerosis(Eshaghi et al., Brain, 2017)
Huntington’s disease(Wijeratne et al., Ann. Clin. Neurol;, 2018)
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Introduction 5 / 20
![Page 6: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/6.jpg)
Alzheimer’s Disease is a Devastating Disease
I 46 million people affected worldwide
I No treatments available that stop or slow down cognitive decline
I Q: Why did clinical trials fail? A: Treatments were not administered early enough
I Q: How can we then identify subjects early in order to administer treatments?
I A: Biomarkers ...
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Introduction 6 / 20
![Page 7: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/7.jpg)
Alzheimer’s Disease is a Devastating Disease
I 46 million people affected worldwide
I No treatments available that stop or slow down cognitive decline
I Q: Why did clinical trials fail? A: Treatments were not administered early enough
I Q: How can we then identify subjects early in order to administer treatments?
I A: Biomarkers ...
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Introduction 6 / 20
![Page 8: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/8.jpg)
Alzheimer’s Disease is a Devastating Disease
I 46 million people affected worldwide
I No treatments available that stop or slow down cognitive decline
I Q: Why did clinical trials fail? A: Treatments were not administered early enough
I Q: How can we then identify subjects early in order to administer treatments?
I A: Biomarkers ...
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Introduction 6 / 20
![Page 9: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/9.jpg)
Biomarker Evolution creates a Unique Disease Signaturethat can be used for Staging Individuals in Clinical Trials
ADNI website
I Accurate disease staging → better patient stratification
I Problem: This is a ”hypothetical” (i.e. qualitative) disease progression model
I Why construct a quantitative model?
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Introduction 7 / 20
![Page 10: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/10.jpg)
Benefits of Quantitative Disease Progression Models
I Basic biological insight
I Staging can help stratification in clinical trials
I Differential diagnosis and prognosis
How can we build such a disease progression model?
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Disease Progression Modelling 8 / 20
![Page 11: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/11.jpg)
Benefits of Quantitative Disease Progression Models
I Basic biological insight
I Staging can help stratification in clinical trials
I Differential diagnosis and prognosis
How can we build such a disease progression model?
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Disease Progression Modelling 8 / 20
![Page 12: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/12.jpg)
Benefits of Quantitative Disease Progression Models
I Basic biological insight
I Staging can help stratification in clinical trials
I Differential diagnosis and prognosis
How can we build such a disease progression model?
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Disease Progression Modelling 8 / 20
![Page 13: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/13.jpg)
Benefits of Quantitative Disease Progression Models
I Basic biological insight
I Staging can help stratification in clinical trials
I Differential diagnosis and prognosis
How can we build such a disease progression model?
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Disease Progression Modelling 8 / 20
![Page 14: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/14.jpg)
Building a Quantitative Disease Progression Model is difficult
Challenges:
I Patients are at unknown disease stages
I X-axis are not the same (need to construct the disease stage axis)
I Biomarkers have different trajectory shapes
I Cohort is heterogenous
I Control population not well defined
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Disease Progression Modelling 9 / 20
![Page 15: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/15.jpg)
Building a Quantitative Disease Progression Model is difficult
what we have what we want
Challenges:
I Patients are at unknown disease stages
I X-axis are not the same (need to construct the disease stage axis)
I Biomarkers have different trajectory shapes
I Cohort is heterogenous
I Control population not well defined
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Disease Progression Modelling 9 / 20
![Page 16: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/16.jpg)
Building a Quantitative Disease Progression Model is difficult
what we have what we want
Challenges:
I Patients are at unknown disease stages
I X-axis are not the same (need to construct the disease stage axis)
I Biomarkers have different trajectory shapes
I Cohort is heterogenous
I Control population not well defined
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Disease Progression Modelling 9 / 20
![Page 17: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/17.jpg)
Building a Quantitative Disease Progression Model is difficult
Challenges:
I Patients are at unknown disease stages
I X-axis are not the same (need to construct the disease stage axis)
I Biomarkers have different trajectory shapes
I Cohort is heterogenous
I Control population not well defined
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Disease Progression Modelling 9 / 20
![Page 18: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/18.jpg)
Building a Quantitative Disease Progression Model is difficult
Challenges:
I Patients are at unknown disease stages
I X-axis are not the same (need to construct the disease stage axis)
I Biomarkers have different trajectory shapes
I Cohort is heterogenous
I Control population not well defined
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Disease Progression Modelling 9 / 20
![Page 19: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/19.jpg)
Building a Quantitative Disease Progression Model is difficult
Model
Challenges:
I Patients are at unknown disease stages
I X-axis are not the same (need to construct the disease stage axis)
I Biomarkers have different trajectory shapes
I Cohort is heterogenous
I Control population not well defined
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Disease Progression Modelling 9 / 20
![Page 20: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/20.jpg)
Building a Quantitative Disease Progression Model is difficult
Model
Challenges:
I Patients are at unknown disease stages
I X-axis are not the same (need to construct the disease stage axis)
I Biomarkers have different trajectory shapes
I Cohort is heterogenous
I Control population not well defined
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Disease Progression Modelling 9 / 20
![Page 21: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/21.jpg)
My PhD Aim
1. Study the progression of atrophy in two diseases (using existing models):I typical Alzheimer’s Disease (tAD)I Posterior Cortical Atrophy (PCA)
2. Develop novel disease progression models (DPMs)
p(X |S) =∏J
j=1
[∑Nk=0 p(k)
(∏ki=1 p
(xs(i),j |Es(i)
)∏Ni=k+1 p
(xs(i),j |¬Es(i)
))](1)
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Disease Progression Modelling 10 / 20
![Page 22: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/22.jpg)
My PhD Contributions
1. Modelled progression of PCA and tAD 2. Spatio-temporal Progression Modelling
3. Disease Knowledge Transferacross Neurodegenerative Diseases
normal
abnormal
Disease 1 progression
Temporal Dysfunctio
n
Frontal Dysfunctio
n
Occipital
Dysfunction
normal
abnormal
Disease 2 progression
Temporal
Dysfunction
Frontal Dysfunctio
n
normal
abnormal
Temporal Dysfunction
Amyloid temporal
Tau temporal
MRI temporal
normal
abnormal
Occipital Dysfunction
Amyloid -
occipital
Tau occipital
MRI occipita
l
Disease 1 (e.g. tAD) Disease 2 (e.g. PCA)
Dis
ease
Agn
ostic
Dis
ease
Spe
cific
Occipital D
ysfunction
...
Dys
func
tion
scor
e
Dys
func
tion
scor
e
Bio
mar
ker
valu
e
Bio
mar
ker
valu
e
Temporal Unit Occipital Unit
4. TADPOLE Competition
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Disease Progression Modelling 11 / 20
![Page 23: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/23.jpg)
Overview
1. Modelled progression of PCA and tAD 2. Spatio-temporal Progression Modelling
3. Disease Knowledge Transferacross Neurodegenerative Diseases
normal
abnormal
Disease 1 progression
Temporal Dysfunctio
n
Frontal Dysfunctio
n
Occipital
Dysfunction
normal
abnormal
Disease 2 progression
Temporal
Dysfunction
Frontal Dysfunctio
n
normal
abnormal
Temporal Dysfunction
Amyloid temporal
Tau temporal
MRI temporal
normal
abnormal
Occipital Dysfunction
Amyloid -
occipital
Tau occipital
MRI occipita
l
Disease 1 (e.g. tAD) Disease 2 (e.g. PCA)
Dis
ease
Agn
ostic
Dis
ease
Spe
cific
Occipital D
ysfunction
...
Dys
func
tion
scor
e
Dys
func
tion
scor
e
Bio
mar
ker
valu
e
Bio
mar
ker
valu
e
Temporal Unit Occipital Unit
4. TADPOLE Competition
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Disease Progression Modelling 12 / 20
![Page 24: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/24.jpg)
Clinical question: Find the order in which GM regions become atrophied
I in PCA
I in tAD
Why? No previous studies modelled disease progression in PCA
Demographics:
I cohort from the Dementia Research Centre with uniquely large PCA population (70)
# SubjectsGender
M/FAge at baseline
(years)Years from
onset (years)Controls 89 33/56 60.5 ± 11 -
PCA 70 27/43 63.0 ± 7 4.4 ± 2.8AD 65 34/31 66.3 ± 8 4.8 ± 2.6
Data: Structural MRI scans
Impact: the first major investigation of PCA disease progression
How? The Event-Based Model ...
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Cross-sectional Modelling 13 / 20
![Page 25: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/25.jpg)
Key Idea: The Event-Based Model Estimates an Atrophy Sequence fromInformative Patient Snapshots
I Event-Based Model (EBM): Fontejin et al., Neroimage, 2012.
I Aim: Region 1 → Region 2 vs Region 2 → Region 1
Patient 1 Patient 2 Patient 3Region 1 1.1 0.9 0.1Region 2 0.95 0.0 0.05
Patient 1 Patient 2 Patient 3Region 1 normal normal abnormalRegion 2 normal abnormal abnormal
Region 1 Region 2
Estimated Sequence: Region 2 → Region 1
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Cross-sectional Modelling 14 / 20
![Page 26: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/26.jpg)
Key Idea: The Event-Based Model Estimates an Atrophy Sequence fromInformative Patient Snapshots
I Event-Based Model (EBM): Fontejin et al., Neroimage, 2012.
I Aim: Region 1 → Region 2 vs Region 2 → Region 1
Patient 1 Patient 2 Patient 3Region 1 1.1 0.9 0.1Region 2 0.95 0.0 0.05
Patient 1 Patient 2 Patient 3Region 1 normal normal abnormalRegion 2 normal abnormal abnormal
Region 1 Region 2
Estimated Sequence: Region 2 → Region 1
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Cross-sectional Modelling 14 / 20
![Page 27: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/27.jpg)
Key Idea: The Event-Based Model Estimates an Atrophy Sequence fromInformative Patient Snapshots
I Event-Based Model (EBM): Fontejin et al., Neroimage, 2012.
I Aim: Region 1 → Region 2 vs Region 2 → Region 1
Patient 1 Patient 2 Patient 3Region 1 1.1 0.9 0.1Region 2 0.95 0.0 0.05
Patient 1 Patient 2 Patient 3Region 1 normal normal abnormalRegion 2 normal abnormal abnormal
Region 1 Region 2
Estimated Sequence: Region 2 → Region 1
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Cross-sectional Modelling 14 / 20
![Page 28: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/28.jpg)
Key Idea: The Event-Based Model Estimates an Atrophy Sequence fromInformative Patient Snapshots
I Event-Based Model (EBM): Fontejin et al., Neroimage, 2012.
I Aim: Region 1 → Region 2 vs Region 2 → Region 1
Patient 1 Patient 2 Patient 3Region 1 1.1 0.9 0.1Region 2 0.95 0.0 0.05
Patient 1 Patient 2 Patient 3Region 1 normal normal abnormalRegion 2 normal abnormal abnormal
Region 1 Region 2
Estimated Sequence: Region 2 → Region 1
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Cross-sectional Modelling 14 / 20
![Page 29: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/29.jpg)
Key Idea: The Event-Based Model Estimates an Atrophy Sequence fromInformative Patient Snapshots
I Event-Based Model (EBM): Fontejin et al., Neroimage, 2012.
I Aim: Region 1 → Region 2 vs Region 2 → Region 1
Patient 1 Patient 2 Patient 3Region 1 1.1 0.9 0.1Region 2 0.95 0.0 0.05
Patient 1 Patient 2 Patient 3Region 1 normal normal abnormalRegion 2 normal abnormal abnormal
Region 1 Region 2
Estimated Sequence: Region 2 → Region 1
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Cross-sectional Modelling 14 / 20
![Page 30: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/30.jpg)
The EBM assumes a subject at stage k has first k biomarkers ”abnormal” and thelast N − k biomarkers ”normal”
I Evaluate data likelihood under normal and abnormal distributions:I normal - p
(xs(i),j |¬Es(i)
)I abnormal - p
(xs(i),j |Es(i)
)I Compute likelihood of one subject j being at stage k given sequence S :
p(Xj |S, k) =k∏
i=1
p(xs(i),j |Es(i)
) N∏i=k+1
p(xs(i),j |¬Es(i)
)I Marginalise stage k:
p(Xj |S) =N∑
k=0
p(k)
k∏i=1
p(xs(i),j |Es(i)
) N∏i=k+1
p(xs(i),j |¬Es(i)
)I Extend to all subjects:
p(X |S) =J∏
j=1
N∑k=0
p(k)
k∏i=1
p(xs(i),j |Es(i)
) N∏i=k+1
p(xs(i),j |¬Es(i)
)I Sequence and uncertainty estimated with MCMC sampling
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Cross-sectional Modelling 15 / 20
![Page 31: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/31.jpg)
The EBM finds a Distinct Atrophy Sequence in PCA compared to tAD
I PCA → early occipital and superior parietal atrophy
I tAD → early hippocampal and inferior temporal atrophy
PCA
Stage 4 Stage 8 Stage 16 Stage 24 Stage 32abnormal
normal
tAD
Stage 4 Stage 8 Stage 16 Stage 24 Stage 32abnormal
normal
Firth, Marinescu and Primativo, in first revision (Brain)
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Cross-sectional Modelling 16 / 20
![Page 32: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/32.jpg)
Atrophy Patterns Resemble Previous Studies from the Literature
I PCA → early occipital and superior parietal atrophy
I tAD → early hippocampal and inferior temporal atrophy
PCA
Stage 4 Stage 8 Stage 16 Stage 24 Stage 32abnormal
normal
Lehmann et al., 2012
tAD
Stage 4 Stage 8 Stage 16 Stage 24 Stage 32abnormal
normal
Lehmann et al., 2012
Firth, Marinescu and Primativo, in first revision (Brain)
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Cross-sectional Modelling 17 / 20
![Page 33: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/33.jpg)
PCA Subtypes show Different Atrophy Progressions, providing Evidence forHeterogeneity within PCA
Initial hypotheses
1. Basic visual impairment → early atrophy in occipital lobe
2. Space perception impairment → early atrophy in superior parietal lobe
3. Visuoperceptual impairment → early atrophy in inferior temporal lobe
1. Basicvisualimpairment(n=21)
Stage 1 Stage 2 Stage 3 Stage 4
2. Spaceperceptionimpairment(n=21)
Stage 1 Stage 2 Stage 3 Stage 4
3. Visuo-perceptualimpairment(n=22)
Stage 1 Stage 2 Stage 3 Stage 4
Firth, Marinescu and Primativo, in first revision (Brain)
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Cross-sectional Modelling 18 / 20
![Page 34: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/34.jpg)
The Differential Equation Model reconstructs Biomarker Trajectories fromShort-term Longitudinal Measurements
What we want What we have
lim∆t−→0∆x
∆t=δx
δt= f (x)
Solve for x using the Eulermethod:
t1 = t0 + δt
x1 = x0 + f (x0)δt
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Cross-sectional Modelling 19 / 20
![Page 35: Modelling the Neuroanatomical Progression of Alzheimerâ s ... › ~razvan › talk › martinos2019 › pres.pdfModelling the Neuroanatomical Progression of Alzheimers Disease and](https://reader035.vdocuments.mx/reader035/viewer/2022070820/5f1d44fbf945e325c62190bb/html5/thumbnails/35.jpg)
Model Recapitulates Differences in PCA vs tAD Atrophy Progression
I PCA: rapid and extensive atrophy in occipital and parietal regions
I tAD: global atrophy pattern, with early hippocampal involvement
(a) PCA (b) tAD
Firth, Marinescu and Primativo, in first revision (Brain)
Razvan V. Marinescu [email protected] http://razvan.csail.mit.edu Cross-sectional Modelling 20 / 20