[Mristudio-users] Fibrebundle - help

[susumu mori]

Hi Nagulan,

-------------------------------------------------------------------------------------------------

>Considering a deterministic tracking method and one seed point,

>Is it really natural to represent a whole bundle by a single line? is this
representation not at risk of being misinterpreted clinically? What is the
exact interpretation of these lines?

>Please can you suugest some idas or references about these questions.

------------------------------------------------------------------------------------------------

It's an important and fundamental question and please allow me to share my
though with other users in the mailing list. What I'm going to describe
below is mostly my opinion and I'll try to make clear which is fact and
which is my opinion.

First of all, when we talk about tracts and fibers and bundles and
connections, we are talking about axons. There are 100 billion neurons in
the brain and each neuron has one axon to communicate with other regions of
the brain. The axon branches at its terminal but also along the trajectory.
Combined with the dendrite network, the notion of fibers and connections
could be very complicated; even if we can reconstruct a complete picture of
one neuron, its anatomy and connection are far from simple one-to-one
communication. If one neuron communicate with 10 neurons in average, it is a
question of the size of 100 billion x 10 (or maybe 100 billion^10?).

Apparently, this is way beyond our ability for complete delineation of the
system. Therefore, we need to resort to some kind of methods to reduce the
dimension of the information. During this process, it is inevitable that
much information is degenerated. For example, if we obtain 10GB worth of
anatomical information from one brain, we still loose an astronomical amount
of information. So in general, when we talk about axonal anatomy and
connectivity, we need to accept whatever the technique we use gives only
approximation with HUGELY degenerated information.

The above argument started from a microscopic cellular level point of view.
When we do MRI, we usually approach from the other end of the spectrum. We
observe properties of water molecules (which is a very indirect indicator of
any cellular entity) with 1-3 mm spatial resolution, acquired within 10 min,
and has merely 10 MB of data. We can understand that a huge amount of
anatomical information is already lost at the point of data acquisition,
which is impossible to recover using any sophisticated mathematical tools.
We can say any biological statement from this kind of measurements can only
be an approximation. On the other hand, we can also argue that it is an
excellent method to systematically, consistently, and quantitatively reduce
the unmanageably large system to 10 MB within 10 min so that we can
investigate and compare the status of the anatomy, which can't be done by
any microscopic (histological) methods.

There is a very common issue that arises in this kind of situations; between
the microscopic cellular studies and macroscopic imaging methods, there is a
gap  so large that we often (well, I would say usually) can't connect these
two sides of information; e.g. if we find one line generated by a
deterministic tracking, we don't know what it exactly means in terms of a
cellular level entity.

In my opinion, there are two types of approaches we usually take to deal
with this "information gap". First, we often use EMPIRICAL RELATIONSHIP. I
believe many radiological diagnosis is based on this. Certain appearance in
brain MRI suggests brain tumor even if it doesn't necessary tell where
exactly is the tumor. Decrease in diffusion constant suggest stroke
infarction.  These are the examples of useful empirical relationship we use
every day. Second is GROUP DIFFERENCE. T1-weighted images of AD patients may
suggest cortical thinning. This could be real thinning or maybe due to
contrast change. Even if we don't know the real cellular event, if MRI can
reproducibly detect differences, it is very useful information. As for fiber
tracking, one may find that a particular tract in a disease group is
smaller. This could be due to FA decrease or size reduction of the tract. We
can't immediately tell what is really happening, but the fact the
DTI/tractography can detect such changes non-invasive and reproducibly may
prove very useful.

Now let's go back to the question of "what does the line mean". There are
also related questions like, "is the line really true (validation)" or "does
it really represent connection". Before answering this question, we first
have to define what is exactly the anatomical entity we are going after.
Without defining the target, we can't do validation studies. First of all,
it is obvious that we are not talking about a single axon which is only few
micron in diameter. Nobody would argue that DTI-based tractography
represents a single axon and I'm sure nobody wants to study 100 billion
axons either. Then what is exactly we want to study? Suppose we have a dream
imaging method that can give complete pictures of whatever we want to image.
Then what is the exact anatomical entity we want to study? Well, as long as
we are talking about the white matter, it must be "a bundle of axons". One
may argue, "we know that the corticospinal tract does exist and this is what
we want to know". There are many bundles of axons that were described for
many years such as SLF, cingulum, thalamic radiation, etc. These structures
have been described using histology samples and visual inspection of the
macroscopic configuration of the white matter. Those named fibers are mostly
"anatomical features", with loose definitions about the boarders; many axons
may merge and exit along the tract. Because they are merely features without
clear definition, it becomes a very difficult to task to use them as the
target of the study; if the anatomical target is vague, how can we validate
our DTI-tractography results against it? For example, the CST is one of the
most well-characterized tracts. We know that it exists and also know the
location quite well. However, there are so many other fibers running
parallel to it that it is very difficult to define it in the cerebral
hemisphere. At the pons level, it has much more compacted shape with much
less contaminating axons, although there still contain a lot of
corticopontine and corticobulbar axons. Even at the pons level, histology
reveals that the CST axons are interdigitated with a significant number of
pontine crossing fibers and the CST axons are distributed from the high
concentration at the center to gradual reduction of the concentration toward
the edge. So, even if we have histology, it is very difficult to define
boundary and the clear entity of the CST. So, if we can obtain line (or
lines) seemingly representing the CST, it is hard to tell the validity or
what it exactly represents.

So, in summary, the problem we are facing is, the DTI-tractography only
gives macroscopic and indirect (water property) measure of white matter
features, while the anatomy we are interested in is also vaguely defined.
Therefore any efforts that try to attach true (whatever the "true" means)
anatomical meaning to DTI results often hit a big wall. There are other
similar situations when we ask, "what is the meaning of decreased FA" and
"can we study brain connectivity"; in both cases, the measurements and
targets are not well defined. In this kind of situation, in my opinion, I
feel that we have to resort to "empirical relationship" and "group
differences" to extract some PRACTICAL conclusions. This  may be enough if
one wants to contribute to clinical diagnosis. If one is interested in
neuroscience (or ground truth), DTI/MRI may give a good indication about
what could be wrong in which area so that we can direct subsequent analyses
using, for example, histology or other modalities.

Another common and specific question about tractography is, "then how we can
believe tractography results when we do tumor surgery". If we have to take
"real or not" approach, it should be not too difficult to come up with many
cases showing how tractography fails, and thus dismiss tractography. In this
kind of questions, I think it's very important to define practical goals
rather than asking a simple deterministic question like "is it real or not".
For example, the usefulness or accuracy of tractography can be judged
differently depending on whether we are asking 5 mm accuracy or merely
asking "is the CST left or right side of the tumor". This is a good example
that our answer requires clear statement about the end goal. Without
defining the goal and a required accuracy level, it is sometimes sterile to
discuss about the validity of tools. In this case, we are actually asking
"does the availability of tractography results on the CST improve the
outcome of the tumor surgery?". When we ask this kind of questions, we
rarely expect improvement in 100% cases. Just like drug trials, we should
talk about statistically significant improvement and evaluate cost/benefit
functions including the rate of adverse impacts. In this situation, I think
a simple question like "does tractography give real CST" is a too naive
question.

I guess I'm not giving you a straight answer, but hope this would give you a
good background about DTI/Tractography research.

Susumu


On Mon, Feb 14, 2011 at 5:52 AM, N.Ratnarajah <rn54 at kent.ac.uk> wrote:

> Dear Susumu
>
> Considering a deterministic tracking method and one seed point,
>
> Is it really natural to represent a whole bundle by a single line? is this
> representation not at risk of being misinterpreted clinically? What is the
> exact interpretation of these lines?
>
> Please can you suugest some idas or references about these questions..
>
> Sincerely
> Nagulan
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.mristudio.org/pipermail/mristudio-users/attachments/20110219/0748e732/attachment.html