[Neuroimaging] [Nipy-devel][Dipy] A new circle of Google Summer of Code starts - time for new proposals

[Samuel St-Jean]

At the same time, a sample of project proposal that I could mentor if
anyone is interested/want to add stuff.

Currently, there is two noise estimation [1] methods in dipy, which both
have their strengths and limitations. Having other methods which can
complement these could enhance the performance of the RESTORE dti fitting
or the nlmeans denoising currently, in addition to any other module which
benefits from modelling noise uncertainty in its fitting process.

Firstly, estimate_sigma works by predicting a single value for a whole
volume, which is suboptimal for acquisition which produces spatially
varying noise. It is also only designed for Rician noise.
Secondly, piesno circumvent this problem by predicting a per slice value
and works for both Rician/noncentral chi noise, but still assumes that the
background contains a single noise distribution, which can fail when it was
masked by the scanner. Moreover, there is no surefire way currently to
estimate the degree of freedom arising from the noise distribution, which
is left to the user and it reconstruction dependent.

A cool project would be to
1. Implement a true 3D noise estimation function which works for
rician/noncentral chi noise.
2. Implement a function to estimate the distribution/degrees of freedom
producing the noise.
3. Enhance the examples by comparing the effect of these different
estimation techniques on both RESTORE and nlmeans.

For starters, [2] and [3] seems to fix task no 1 and 2. Of course any other
worthwhile algorithm is welcomed.

[1] https://github.com/nipy/dipy/blob/master/dipy/denoise/noise_estimate.py
[2] Veraart, J., Fieremans, E., & Novikov, D. S. (2015). Diffusion MRI
noise mapping using random matrix theory. Magnetic Resonance in Medicine,
http://doi.org/10.1002/mrm.26059
[3]
https://www.lpi.tel.uva.es/~santi/personal/docus/noise_survey_tec_report.pdf

2016-02-03 10:43 GMT+01:00 Samuel St-Jean <stjeansam at gmail.com>:

> For 2, it already exists as you mentioned, and there is also the recently
> released [1] that propose another way to do it. It also has a python port
> (same original author as the matlab version) using cvxopt under the hood,
> but I don't know where the discussion is at for proper inclusion in dipy.
>
> [1] Deslauriers-Gauthier, S., P. Marziliano, M. Paquette, and M.
> Descoteaux <http://scil.dinf.usherbrooke.ca/?author=3&lang=en>. "The
> application of a new sampling theorem for non-bandlimited signals on the
> sphere: Improving the recovery of crossing fibers for low b-value
> acquisitions.
> <http://scil.dinf.usherbrooke.ca/wp-content/papers/deslauriers-etal-media16.pdf>
> " Medical Image Analysis, 2016.
>
> http://scil.dinf.usherbrooke.ca/wp-content/papers/deslauriers-etal-media16.pdf
>
> 2016-02-03 1:59 GMT+01:00 Julio Villalon <jevillalonr at gmail.com>:
>
>> Hi Eleftherios + Nipy community,
>>
>> This is really great. GSoC 2015 was a very rewarding experience for me. I
>> am willing to be a co-mentor with Eleftherios, Ariel and Omar this year.
>>
>> I have some ideas which I would like to share with you.
>>
>> 1. Bias field/non-uniformity correction of T1-weighted images. There are
>> many freely available tools that do this: SPM, FSL-FAST, N3 (MINC and
>> Freesurfer), N4 (ANTS), BrainVoyager, etc. The idea is to implement the
>> best one of these and include it as part of the processing pipeline of
>> Dipy. Bias Field correction of T1 images allows for better segmentation and
>> consequently for better partial volume estimation of brain tissue types,
>> which ultimately has a direct impact on novel tractography techniques such
>> as Anatomically-Constrained Tractography (ACT).
>>
>> - Does anyone know is there is anything available in Python?
>> - Which of the mentioned methods is better?
>>
>>
>> 2. Recovery of local intra-voxel fiber structure for DTI-like
>> sampling/acquisition schemes (6-40 samples, b<=1200 s/mm2). Most of the
>> acquired Diffusion MRI (DMRI) data available nowadays is data acquired for
>> clinical/neuroscientific  studies looking at the effects of many diseases
>> on the brain (schizophrenia, bipolar disorder, HIV, autism, Alzheimer's
>> Disease, etc). The vast majority of this data has been acquired with
>> sampling schemes with less than 40 samples and a single shell of less than
>> 1200 s/mm2. With recent advancements in dictionary learning and sparse
>> recovery techniques (e.g. Merlet et al, 2013), the idea is to make these
>> tools available to the general public, especially to those who have this
>> type of data and can make a better use of it.
>>
>> Please let me know what you think.
>>
>> Thanks
>>
>> Julio
>>
>>
>> 2016-01-31 13:35 GMT-08:00 Eleftherios Garyfallidis <
>> garyfallidis at gmail.com>:
>>
>>> Hello all,
>>>
>>> Taking part in Google Summer of Code (GSoC) is indeed rewording for our
>>> project as it allows for new algorithms to be merged and at the same time
>>> grow our development team with excellent contributors.
>>>
>>> After I believe a successful GSoC participation last year a new circle
>>> starts for this year (2016).
>>>
>>> Last year it was Ariel and me who did most the mentoring. This year we
>>> would like to hear others' ideas too. Therefore, we welcome other
>>> developers/scientists who would like to mentor or propose new projects. For
>>> those who want to mentor we will be happily co-mentors to help them with
>>> the process and give extra feedback to the relevant students.
>>>
>>> In the following link we started adding projects that we think would be
>>> interesting for this year's GSoC
>>>
>>> https://github.com/nipy/dipy/wiki/Google-Summer-of-Code-2016
>>>
>>> Be happy to add your projects at the wiki or suggest ideas in this
>>> thread. We also welcome the previous participants of GSoC (Julio and
>>> Rafael) to take part as mentors this year.
>>>
>>> Finally, this year I am hoping to be able to get more than 2 projects
>>> funded. Hopefully 4 but that is not certain.
>>>
>>> Waiting for your ideas/suggestions. What would you like to see in Dipy
>>> that could be developed by a student during this summer?
>>>
>>> Best regards,
>>> Eleftherios
>>>
>>>
>>>
>>> _______________________________________________
>>> Neuroimaging mailing list
>>> Neuroimaging at python.org
>>> https://mail.python.org/mailman/listinfo/neuroimaging
>>>
>>>
>>
>> _______________________________________________
>> Neuroimaging mailing list
>> Neuroimaging at python.org
>> https://mail.python.org/mailman/listinfo/neuroimaging
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/neuroimaging/attachments/20160203/207e8b29/attachment.html>