---------- Forwarded message ----------
From: Rupin Dalvi <rupin.dalvi@gmail.com>
Date: Wed, Oct 20, 2010 at 8:28 AM
Subject: Re: Super resolution
To: Negar Mohaghegh Harandi <negar.mohaghegh@gmail.com>, Tricia Pang <triciajpang@gmail.com>


Hiya,
Maybe you could just demonstrate this as proof-of-concept - perhaps that would allow you to use 
artificial data and remove the need to register. Though I admit, lack of real data does make it 
less substantial as a project.
Anyway, in case you do go ahead with this as a project, these were two of the papers I was reading 
that Rafeef hasn't already suggested to you:

http://iacl.ece.jhu.edu/~yingbai/pubs/CISS04.pdf

Maximum A Posteriori Estimation of Isotropic High-Resolution Volumetric MRI from Orthogonal 
Thick-Slice Scans
Ali Gholipour, Judy A. Estroff, Mustafa Sahin, Sanjay P. Prabhu and Simon K. Warfield

Regards,

Rupin




On Thu, Oct 14, 2010 at 6:45 PM, Negar Mohaghegh Harandi <negar.mohaghegh@gmail.com> wrote:
>
> Hi Rupin,
>
> As I have understood, the super-resolution method involves mostly three steps or registration 
(motion estimation,...), interpolation and Restoration. Also as you mentioned, we need to acquire 
new data set which will take lots of time in a coarse project framework. I am kind of not sure if 
it can be done in less than 2 month, regarding the fact that the topic is new for m and if the 
result is good enough... What do you think?
> I may shift to another topic like upper airway segmentation... But still waiting for your ideas.
>
> Thanks,
> Negar

>
> On Thu, Oct 14, 2010 at 2:26 PM, Rupin Dalvi <rupin.dalvi@gmail.com> wrote:
>>
>> Hey Negar, Tricia,
>> Sorry I took so long to reply, life is kinda busy :).
>> I was looking at the super resolution very briefly, so I may not be the best person to ask, 
but I'll dig up the papers I was reading and send them to you. 
>> Also, the whole reason I was trying to do orthogonal super-resolution was to achieve one of two 
things: 1) increase the speed of acquisition of the MRI data, 2) improve the signal to noise 
ratio. The idea for the improvement of the SNR was as follows:
>> If your goal is to get a 300*300*300 volume (each pixel of 1mm^3), then:
>> Instead of taking a full 300*300*300 scan, take 3 scans (300*300*100 (of pixel size 1mm*1mm*3mm), 
300*100*300 (of pixel size 1mm*3mm*1mm) and 100*300*300 (of pixel size 3mm*1mm*1mm)) and then put 
then together using orthogonal super-resolution.
>> The motivation for this is: since high resolution MRI data typically has low SNR, by reducing 
the resolution in each direction by a third, we can improve the SNR by, I believe, 9 times. 
Moreover, we should be able to use the three orthogonal slices to recover the target 1mm*1mm*1mm 
resolution. Hence, the final output would be high resolution, high SNR volumes without sacrificing 
speed.
>> If this sounds like a good project, let me know. I have written some simple code in MATLAB that 
I could give you that makes an attempt at the interpolation (super-resolution). I could send that 
over as well if you want.
>> Regards,
>> Rupin
>>
>>
>>
>> On Tue, Oct 12, 2010 at 8:07 PM, Tricia Pang <triciajpang@gmail.com> wrote:
>>>
>>> Hi Rupin,
>>>
>>> Negar in our lab is looking into doing some work on creating super resolution images. I know 
you were looking into this... when you have a chance, could you pass her some of the papers, etc 
you have on the topic?
>>>
>>> Thanks,
>>> Tricia
>>>
>>>
>>>
>>> Sent via carrier pigeon.