Untitled 22

let's do the same here.
here's my most improtant work a large chapter

which I will layout in details here. After taking those details in mind, I will
paste you the existing chapter in LateX and you should again, act as a
supervisor, do give any feedback and then incorporate that feedback in
the writing and editing process, and spit out a latex code with the final
text i need to put into my thesis as a chapter. 1) preprocessing of a
massive dataset 1450 total subjects from ABIDE I and ABIDE II. fmriprep
was used in a parallel CPU fashion, in king's college london NaN multi-cpu
serverNeuroimaging Analysis Network (NaN) this was a very lengthy
process so again, SGE scripts were used (attach them to the appendix).
The total subjects from the ABIDE was 2200 but they were excluded based
on various parameters such as First we will exclude: - 20% percentage
with FD more than 0.2mm, using the threshold from Parkes 2018 [1] -
subjects with mean FD more than 0.5mm (2sd of sample mean (0.23+-
0.27)) [2] - sites with less than 10 participants to avoid skewing - subjects
with FIQ less than 80 [3] ~~subjects with FIQ outside of 2s.d. of the
overall ABIDE sample mean (108+_15)~~ References: - [1] Parkes 2018 -
[2] "The autism brain imaging data exchange: towards a large-scale
evaluation of the intrinsic brain architecture in autism, Nature 2013 A Di
Martino" - [3] Potential Locations for Noninvasive Brain Stimulation in
Treating Autism Spectrum Disorders—A Functional Connectivity Study, Y.
Huang et al frontiers in Psychiatry, 2020 after the preprocessing Hermes
was used to calculate in parallel the voxelwise degree centrarlity for each
of the subjects 1b) Here we run GLM using FSL randomise (with 10000
iterations) for two things a) Are the degree centrality maps of the autistic
subjects correlated voxelwise with non-autistic? I have two plots here. One
shows Are there any brain areas that differentiate significantly between
autistic vs non-autistic in ABIDE I/II? and the other in my subset of AIMS2-
TRIALS. I can find significant voxels, especially for ABIDE in Frontal lobe,
motor function area and parietal lobe 2) AIM: Can we use deep learning to
classify between autistic and non-autistic participants?I performed
Classification of autism using a deep learning (densenet-121) classifier
network which was parameter hyperoptimised to achieve 70.2accuracy
and 0.72 roc/auc with nested 5-fold cross validation 3) AIM: Can we use
saliency maps to identify regions in the brain that may be important for
autism classification? Here I used smoothgrad to create individualised or
PERSONALISED SALIECNY MAPS (I call them PSAMs) from now on.
Smoothgrad5: Explainable Artificial Intelligence algorithm which was
applied on the Densenet-121 with seven consecutive dense blocks to
extract individual and personalised saliency maps (PSAM), indicating the
voxelwise significance for the classification.Using those PSAMs i went to
the following aim: 4) AIM: Can we apply dimensionality reduction and
clustering to the personalised saliency maps to further categorise autism
into subgroups? A data driven-approach of discovering subtypes ws
chosen using non-linear dimensionality reduction using Uniform Manifold
Approximation and Projection (UMAP7) and clustering using k-means8,
informed by their silhouette score and their Calinski-Harabasz Index. I still
have not all the results here but the four subtypes different significantly in
between ages (and I already have corrected on age because my initial
dataset traininig has been cross validated as well as divided into
homogeneous subgroups before sampling (based on age). That was not
possible for sex because there's a big discrepancy for sex, something i
need to consider on my discussion. OK so until here we got four different
subgroups and I have brain images whih show significant differences
between those groups in specific areas (not finalised those yet), let's put it
as a draft and we can focus on this later. Finally, one thing i am currently
doing is that I am using Yeo Atlas to see for ROI differences between each
of those four subtypes and how those can be different. Again let's keep
this as a draft and we can discuss it later. If you are all OK with the
previous I am now going to paste the chapter. A note about the language:
I am not using ASC or ASD or anything like that. I am calling it Autism
(with capital A). And when talking about the individuals i am saying
autistic and non-autistic participants. Also you should never mention that
it is a disability, or patients or they need any kind of intervention. It's
about making their lives easier and exploring what is best for them. But of
course its important for the scientific knowledge to be able to better
understand Autism as well as the individuality of the people etc

Untitled 22

Uploaded by

Copyright:

Available Formats

Untitled 22

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Untitled 22

Uploaded by

Copyright:

Available Formats

let's do the same here.

here's my most improtant work a large chapter

You might also like