Procedure Code to identify a Study Type?

[Sri]

Hello Dicom folks,

I have a conceptual question and would appreciate your views on your
solution approach:

Question: I need to identify a study type (or Procedure type) of a
study that was performed by the modality. We interface with modality
using Query (C-FIND) or MPPS. Is there a way to identify it across
modality vendors and hospital sites? - How do workstation vendor
determine this?

(The "Study Description" tag [0008,1030] does not help unfortunately
as it is in natural language in all the hospital sites - this is hard
to parse & understand by machines).

Possible Solution: There is DICOM tag "Procedure Code
sequence" (0008,1032) obtained from the Modality, through which I
intend to identify the procedure type ("Code Value" tag has this
procedure-code, as shown below).

As I understand, every hospital or site has their unique procedure-
codes defined to identify each procedure (e.g., "CTNECKW" for which
the study-description (0008,1030) would be "CT NECK WITH INTRAVENOUS
CONTRAST"; "CTNECKWO" meaning "CT NECK WITHOUT INTRAVENOUS CONTRAST").

Procedure Code Sequence:
---------------------------------------
Procedure Code Sequence 0008,1032>
Code Value (0008,0100)
Coding Scheme Designator (0008,0102)
Code Meaning tag (008,0104)


The challenge in the above approach is that:
- It's not "guaranteed" to be obtainable from the modality. (Not all
modality might support this using in their Query Imformation model or
MPPS messages).
- This approach would require us to map the procedure-codes of each
site to a common procedure-code dictionary!

Are there any other way to determine the procedure type?

Thanks for your time and effort in responding to this post.

Sridhar

[David Clunie]

Hi Sri

This is the same dilemma that all of us who do multi-site clinical
trials or multi-site radiation dose reporting face.

The plain text strings like Study Description may contain useful
information, but different text and abbreviations are used, and
when the sites are multi-national, one is faced with text and
abbreviations in different languages too !

There is no common use of standard codes for Requested Procedure
Code Sequence or Procedure Code Sequence, unfortunately, and
each site configures their own, usually in the RIS and this
may get copied via the modality worklist; often even within a
site these are malformed, not unique, and missing coding scheme
designators or using illegal coding scheme designators (that
do not start with 99 for private ones, etc.).

There are standard codes available, in SNOMED, in the RadLex
Playbook (though this is a work in progress) and in various
countries (like the UK interim codes), but in most countries
these are not used, and for many sites they are not sufficient
to define sufficient granularity for ordering local customized
procedures (like David Clunie's favorite brain tumor perfusion
protocol as distinct from the standard one).

Even if one does recognize the code set in use, the value populated
in the headers is often wrong (e.g., it will claim to be just a
CT Chest, but is actually a CT Chest/Abdo/Pelvis).

In the com.pixelmed.anatproc package I have a very crude and
unreliable attempt at multi-lingual guessing of the right choice.

Graham Warden in his GROK extensions of my tools uses the Z extent
to further refine likelihood of being one body part or another.

Another approach would be to actually analyze the image content,
e.g., to use the density histogram pattern to match against
known patterns, or to actually segment the organs and/or do
some sort of registration against an atlas of organs or body
shapes or whatever; probably too much work and difficult to
generalize to all modalities and body shapes and patient ages
for your use case though.

The ACR Dose Index Registry folks have taken the simple expedient
of mapping all supplied strings for Study Description to their
standard internal codes once the pattern has been recognized or
specified by the site; their goal is to use RadLex (and extend
it as needed) as their internal target.

If one tries to go down to the series or procedure step or
acquisition or irradiation event level, not just the study
level, the problem gets even harder.

David

On 1/25/12 4:20 PM, Sri wrote:
> Hello Dicom folks,
>
> I have a conceptual question and would appreciate your views on your
> solution approach:
>
> Question: I need to identify a study type (or Procedure type) of a
> study that was performed by the modality. We interface with modality
> using Query (C-FIND) or MPPS. Is there a way to identify it across
> modality vendors and hospital sites? - How do workstation vendor
> determine this?
>
> (The "Study Description" tag [0008,1030] does not help unfortunately
> as it is in natural language in all the hospital sites - this is hard

> to parse& understand by machines).

[Sri]

Hi David, Thank you very much for sharing your thoughts.

As far as the parsing of study-description (or procedure-type) text
string, we only need to determine, if a given study is "with Contrast"
or "without Contrast".
We have 2 solutions as you mentioned:

1. Parse the Study-Description string coming from the modality's MPPS
messages or C-FIND responses
2. Map all study-description strings for each site to internal codes

How is your experience with parsing procedure-description strings?
i.e.,Do you think it would be possible to determine if a study is
"with Contrast" or "without Contrast", without any errors?
if not, we'll have to go on the mapping path. (which is simple but
complex configuration job for each site)

Thanks again for your time in responding to my questions David.

Sridhar

[David Clunie]

On 1/27/12 11:06 AM, Sri wrote:

> As far as the parsing of study-description (or procedure-type) text
> string, we only need to determine, if a given study is "with Contrast"
> or "without Contrast".
> We have 2 solutions as you mentioned:
>
> 1. Parse the Study-Description string coming from the modality's MPPS
> messages or C-FIND responses
> 2. Map all study-description strings for each site to internal codes
>
> How is your experience with parsing procedure-description strings?
> i.e.,Do you think it would be possible to determine if a study is
> "with Contrast" or "without Contrast", without any errors?
> if not, we'll have to go on the mapping path. (which is simple but
> complex configuration job for each site)

Hi Sri

Ah, that is a much simpler problem.

It is not really practical to determine whether contrast has been
administered at the study level, since this is often not recorded
or encoded in Study Description or a study-level coded attribute
like Procedure Code Sequence. These values are often generic and
the same text or code may be used whether IV contrast was used or
not.

The best thing to do is to look to see if any of the images in the
study contain a value for Contrast/Bolus Agent (0018,0010).

In most cases, a study will contain a mixture of images, some with
contrast, some not, and this attribute should reflect that.

Since this is typically an operator entered free text field,
sometimes it will not be populated, and sometimes the operator
or machine will do something silly (and illegal per the standard)
like enter text of "no contrast".

I am not sure how detailed your requirement is, but theoretically
you should be able to distinguish between "without contrast",
"with contrast" and "both with and without contrast", depending
on which image is populated with what for this field (and excluding
localizers which are usually pre-contrast but don't count).

If Contrast/Bolus Agent is not populated, you may need to consider
Series Description, which may contain "+C" or "C+" or "POST" or
"IV" or "Gd" or something similar, but this is less reliable.

So you may need a site-specific (or even vendor/model/software version
or machine-specific) mapping, but at more than just the study level.

David

[Sri]

Thank you David.

I'm thinking in the direction of site-specific mapping of DICOM Study
Description to a common procedure description for each procedure type
used in the hospital site. (I'm not sure how successful parsing these
text strings would be, given the complexity of internationlization).
This opens door to another issue - How does one collect these site-
specific procedure descriptions from each site (in order to map)?

I'm thinking of a way to device a simple C-FIND Query application to
query the RIS (MWL) or the modality (standard DICOM Query) - for a
lengthy window of periods, (hoping I would get most of the studies
performed in the site, this way) - once we obtain all of these
procedure descriptions from the RIS DB, we can then manually map and
configure every procedure types and what it means.

Could you share your opinion about this approach?

I would love to hear about the implementation experience of this
mapping effort done by ACR Dose Index Registry folks you mentioned
about.

Thank you David once again for your kind help,

Sridhar