Techniques

Question 1

Name three things that diffusion does

Answer 1

Diffusion techniques allow measurement of water motion
DWI provides evaluation of tissular motility
DTI allows demonstration motion strength and direction

Question 2

What is BOLD in fMRI

RSN

Answer 2

Blood Oxygenation Level Dependent (effect)

RSN= Resting State Network

Question 3

The fMRI signal

The fMRI activations that we see

Answer 3

Is NOT direct neuronal activation
IT IS YES indirect neurovascular response (BOLD effect)
The results depend on fMRI data quality

It it YES indirect statistical analysis of a neurovascular response
Know the statistics

Question 4

Pros and cons of activation/Task fMRI

Answer 4

PROS: Activation related to a specific task

Cons:

• Only one task at a time
• Experimental setup
• Patient compliance

Question 5

Pros and cons of SEED Based Functional connectivity

Answer 5

PROS:
Easy
Reproducible

CONS:
Only one or few networks
Seed not necessarily the best spot for functional connectivity network

Question 6

ICA (Independent Component Analysis) functional connectivity pros and cons

Answer 6

PROS:
+Multiple RSN networks at same time
+Automatic selection

CONS:

• results depend on input data and ICA parameters
• not reproducible
• how many components?
• what do with individual patient?

Question 7

Atlas Based Functional Connectivity pros and cons

Answer 7

PROS:
+Multiple networks
+Automatic selection
+ Reproducible

Cons:

• Atlas ROI not necessarily matches functional region/not necessarily best spot for RSN
• Results depend on atlas

Question 8

1 dataset of 64 x 64 to 128 x 128, around 30 slices contains

How many neurons on a typical unfiltered fMRI voxel?

Answer 8

up to 500.000 voxels

5.5million neurons

Question 9

Data analysis : Differences between research fMRI and clinical fMRI

Answer 9

Research fMRI correction of false positive (+fMRI results are evaluated sceptically) (από τα πολλά positive που βγαινουν θελεις να τα ελαττωσεις) Errors of excessive scepticism

Clinical fMRI correction of false negative ( +do not miss relevant activations) (if there is no activation you want to make sure there is no activation) Errors of excessive activation

Question 10

What is AI? Machine Learning and Deep Learning

Answer 10

Artificial Intelligence: Mimicking the Intelligence or behavioural pattern of humans or any other living entity

Machine Learning: A technique by which a computer can learn from data. Mainly based on training a model from datasets.

Deep Learning: A technique to perform machine learning inspired by our brain’s own network of neurons

DL subgroup of ML which is subgroup of AI

Question 11

Pros and cons of supervised vs unsupervised learning

Answer 11

Supervised:
\+smaller number of data
\+can classify a disease
-need expert annotated data
-will not detect unexpected diseases

Unsupervised:
\+can detect unexpected diseases
\+no annotation needed
-larger dataset
-detected clusters/pattern not necessarily equal to diseases

Question 12

Cross validation technique

Answer 12

The data may be small lets say 100 participants - you do a ten times cross validation technique , dataset in 10 parts. 9 PARTS FOR TRAINING , 1 FOR TESTING

Question 13

Training vs testing dataset

Answer 13

Best is to have completely different training and testing datasets

The training set to optimize the machine learning and the testing set to test the model and get the results