1. AIM unique identifier questions.
Question by John:
Does AIM not currently support uniquely tying an annotation to a nodule? In the "ImageAnnotation" tag there is the attribute "uniqueindentifier" that appears to be restricted to DICOM UID format.
LIDC is more nodule-centric, you could say, (and RIDER is even more so, tracking nodules over time). How do we use "ImageAnnotation" to track a unique nodule over time?
Answer by Channin, David (NMH)
Consider a nodule N1 found in patient P1 at time T1 by reader R1. N1P1T1R1; that gets AIM Image Annotation UID: 123
Then N1P1T1R gets AIM Image Annotation UID: 345
The adjudicator comes along and says AnnotationOfAnnoation UID: 567 references UID:123 and UID: 345, are the same object, average of two readers = "xyz", etc.
Notes from telconv 2/17/09:
Each annotation gets a unique ID
Second annotation gets its own unique id.
Difference between LIDC and AIM:
LIDC assigns a physical ID to each nodule. AIM relies on AnnotationofAnnotation for nodule tracking.
Mike McNitt G says he worked through it with Bob Schwanke at the last face to face...
What is the attribute in which you store that nodule mapping information:
Next action - discuss Bob Schwanke and Pat Mongolowat on the phone at the same time.
2. The new annotation type definition we would like to add to AIM model.
Question by Qinyan:
Here is the new annotation type definition we would like to add to AIM model. The new type is marked red:
<xs:simpleType name="ImageAnnotationIdentifier">
<xs:restriction base="xs:string">
<xs:enumeration value="RECIST_Baseline_Target_Lesion"/>
<xs:enumeration value="RECIST_Baseline_Non-target_Lesion"/>
<xs:enumeration value="RECIST_Fellow-up_Target_Lesion"/>
<xs:enumeration value="RECIST_Fellow-up_Non-target_Lesion"/>
<xs:enumeration value="LIDC_Chest_CT_Nodule"/>
<xs:enumeration value="LIDC_Chest_CT_NonNodule"/>
<xs:enumeration value="Teaching"/>
<xs:enumeration value="Quality_Control"/>
<xs:enumeration value="Clinical_Finding"/>
<xs:enumeration value="Other"/>
<xs:enumeration value="Brain_Tumor_Baseline_Target_Lesion"/>
<xs:enumeration value="Brain_Tumor_Baseline_Non-Target_Lesion"/>
</xs:restriction>
</xs:simpleType>
Answer by Pat:
For nodule case: you can use New_Lesion or Target_Lesion
For nonNodule case: you can use Non_Target_lesion.
The new values domain for ImageAnnotation and AnnotationOfAnnotaiton
are as follows.
ImageAnnotation
Clinical_Finding
New_Lesion
Non_Target_Lesion
Non_Lesion_at_Baseline
LIDC_Chest_CT_Nodule
Quality_Control
Target_Lesion
Teaching
AnnotationOfAnnotation
Interval_Change
Summary_statistic
Sum_of_Longest_Diameter
Notes from telconv 02/17/09:
Mike McNitt: Target_Lesion and Non_Target_Lesion have specific RECIST Meanings. Non-Nodule in LIDC means "we've seen this, it's not a nodule, don't worry about it".
Q: Can AIM allow you to re-purpose, or clarify your application-specific definitions?
Q: Is there a dictionary for the meaning of Non_Target_Lesion? Are the meanings defined - linking in CDE browser with LIDC - Stuart Turner? this is not in Radlex.
3. imageZ position vs SOPInstanceUID
Question by Justin:
Quick questions in the LIDC annotation you list both an Image Z position and the SopInstanceUID. Here is an example of information on a particular nodule:
<roi>
<imageZposition>-147.979996</imageZposition>
<imageSOP_UID>1.2.826.0.1.3680043.2.656.2.1.1.242.53</imageSOP_UID>
<inclusion>TRUE</inclusion>
<edgeMap>
<xCoord>130</xCoord>
<yCoord>179</yCoord>
</edgeMap>
</roi>
If a reader only had the SOPInstance UID and the X/Y coordinates would they be able to find the nodule without the Z position? We're considering using this format to convey the location information in AIM:
<SpatialCoordinate
CoordinateIndex="0"
Id="0"
imageReferenceUID="1.2.826.0.1.3680043.2.656.2.1.1.242.53"
xsi:type"TwoDimensionalSpatialCoordinate"
x="130"
y="179" />
Do you think this would work? Or is the Z position required to make sense of the nodule location?
Answer by Mike:
If I remember this correctly (and there is significant probability that I don't), Peyton Bland of Michigan (cc'ed on this email to correct me) had some experience that there was some issue with the image Z position in the DICOM header and that the image SOP_UID was a much more reliable value for finding the right slice.
That said, I think we included the z position just in case as it is much more intuitive value and we had the SOP UID as backup in case these were confusing/missing/uninformative for some reason (you can just never tell with DICOM).
Anyone remember something similar to that ? (Peyton? Richie? David? Sam?).
Answer by Peyton:
Based on the fact that you mentioned the word "reliable" in reference to finding z position, I think it goes like this: Based on an experience we had in the past, the DICOM tag "Slice Location" (0020 1041) is not reliable: We saw some cases where the sign seemed to be wrong. Plus, it is a Type 2 tag (I think) which means that it might not have a value. But no problem... "Image Position Patient" (0020 0032) is the tag to use: It is a Type 1 tag (required to have a value), and in our experience, has the "correct" z value. So presumably this DICOM tag is used for the value of *our* <imageZposition> XML tag.
So more directly to Justin's question... Can "SOP Instance UID" (0008 0018) be used? I guess so. It just means that the image having the indicated "SOP Instance UID" would have to be parsed for the value of the "Image Position Patient" tag. This tag has a value like "-143.600006-122.000000-260.434998"; we grab the 3rd element, "-260.434998" since CTs are axials. (And be careful with those "\"s when parsing!) Presumably this value is used by the drawing tools to populate the <imageZposition> XML tag.
So if I can read a little between the lines of Justin's question, I think he's saying that having both <imageSOP_UID> and <imageZposition> is redundant since the latter can be derived from the former. I know we had discussions about this type of redundancy in the LIDC Implementation group (but can't remember specifics). I will say, however, that *not* having a separate <imageZposition> tag would have made some things *much* harder to do in the course of doing QA/QC, writing MAX, etc., but maybe that's just me (although your 2nd parag, Mike, supports my bias toward including the z value).
Notes from telconv 02/17/09:
Comments:
"If you just give me Z position - some people use slice number, some people use SOPInstanceUID, some people use Z position - shall we keep some redunancy?"
How Z position works:
They put the patient in a position, make an arbitrary reference point.
"Anatomic Landmark"
when sam sends out his QA -
Q - how is z position re-created on a scan (how is the Anatomic Landmark referenced?)
