a

[shepardcommander1]

-

---

Members don't see this ad.

 

[Cognovi]

Reasonable idea. There is a ceiling for the AUC for what CT is able to define morphologically. Within that, there may be room for improvement, such as a combination of short and long axis in transverse plane, longest axis in all three planes, some ratio thereof, combination with internal attenuation/enhancement/texture, perinodal stranding, loss of reniform shape or fatty hilum, etc. The more complex you make it for the purpose of discrimination, the less practical it is, the harder it is to remember, and/or the less interrater reliability there is. Within a given ROC curve, then you probably also have to set an operating point, which depends on whether you need to be more sensitive or specific. The decision depends on the clinical scenario / patient population - do you use this only on people with known disease or people who are getting scans for all reasons and the adenopathy is an incidental finding? Your performance will greatly depend on the spectrum of disease in that population.

Regarding using PET/CT as the reference in particular, you have to carefully choose how you define a positive node and how much error is in that based on the existing literature. SUVmax above liver reference level? above 1.5x liver level? The additional problem with using PET as the standard for the problem of small nodes is that the smaller the node, the more likely you'll have a false negative simply because of the limited spatial resolution of PET. If you combine the PET uptake with the size you can compensate (clinically we might say it is intensely avid for its size).

 

[IRattending2021]

This is where AI come in. A future AI maybe able to reformat a node in the background into data sets including 3D data set and just flag the abnormal.

 

[Dave1980]

No idea if this is the appropriate place for this but....

What about the idea of using a node-to-node correlation between PET/CT and pretreatment planning CT to optimize CT criteria for detecting nodal metastasis for sites (middle-cinome/lower income countries/rural areas/etc) without access to PET. The CT scans are based on pretreatment planning CT. Basically, using the observation that the standard 10 mm short-axis cutoff has a horrible sensitivity for the nodal stations (with regards to specific cancer we are studying) we are examining, and since PET/CT positive nodes for this nodal station have an interesting short-axis distribution (many FDG avid nodes under 7 mm short-axis) perhaps we can rework the CT criteria to potential predict PET positive nodes using ROC analysis for this specific nodal station within this specific cancer. Additionally, a study like this would give rad onc docs a better idea what to do when they see lymphadenopathy in this nodal station no (i.e. gross nodal irradiation or not)?

Some concerns I want to address: I know our reference standard (PET/CT) is a limitation as these nodes are not biopsied. But for the nodal region we would study, the specificity is high 98% (meta-analysis), sensitivity is alright at 81%.

Perhaps a potential way to bridge the gap in sensitivity between PET/CT and CT?

Any thoughts?

What exactly would you be doing? The problem with CT only is that affected nodes are often small and large nodes are often unaffected. Lymph node swell and shrink as part of their normal function.

As a radiologist who has been looking at nodes for years I don't think there is anyway of having any reasonable certainty without the PET portion.

 

[shepardcommander1]

What exactly would you be doing? The problem with CT only is that affected nodes are often small and large nodes are often unaffected. Lymph node swell and shrink as part of their normal function.

As a radiologist who has been looking at nodes for years I don't think there is anyway of having any reasonable certainty without the PET portion.

The experimental cohort is patients with PET positive nodes in the particular nodal station. Negative control cohort is patients with same cancer, with PET negative nodes (determined via nuc med doc + rads) in that same nodal station. We merge PET scan with pretreatment planning CT ( 2 mm slice thickness). Measurements including short-axis, long-axis, L/S ratio, and nodal volume (using rad onc treatment planning system) to determine optimal criteria on the basis of ROC analysis.

PET positive nodes = increased FDG uptake relative to blood pool, since a SUV cutoff has been shown to not be reliable

The main goal is to make the case that for this particular cancer for this particular nodal station, the 10 mm short-axis cutoff is too big. ROCs give us about 6 mm short-axis (73% sensitivity, 90% specific, relative to PET avidity). Also if the rad onc doc is examing the node to giving irradiation too, they can use nodal volume which is considerable more sensitivity (what we found) than short-axis (80% sensitive, 92% specific, relative to PET.

With your thoughts on nodes swell and shrink, is this still the case for nodes draining cancer?

Thanks

 

[shepardcommander1]

Reasonable idea. There is a ceiling for the AUC for what CT is able to define morphologically. Within that, there may be room for improvement, such as a combination of short and long axis in transverse plane, longest axis in all three planes, some ratio thereof, combination with internal attenuation/enhancement/texture, perinodal stranding, loss of reniform shape or fatty hilum, etc. The more complex you make it for the purpose of discrimination, the less practical it is, the harder it is to remember, and/or the less interrater reliability there is. Within a given ROC curve, then you probably also have to set an operating point, which depends on whether you need to be more sensitive or specific. The decision depends on the clinical scenario / patient population - do you use this only on people with known disease or people who are getting scans for all reasons and the adenopathy is an incidental finding? Your performance will greatly depend on the spectrum of disease in that population.

Regarding using PET/CT as the reference in particular, you have to carefully choose how you define a positive node and how much error is in that based on the existing literature. SUVmax above liver reference level? above 1.5x liver level? The additional problem with using PET as the standard for the problem of small nodes is that the smaller the node, the more likely you'll have a false negative simply because of the limited spatial resolution of PET. If you combine the PET uptake with the size you can compensate (clinically we might say it is intensely avid for its size).

Thanks for your thoughts. AGree on the idea of practicality (hence why we would like to use short-axis, although we do make measurements of short-axis, long-axis, L/S ratio, and nodal volume (via rad onc treatment planning system, since rad oncs will ultimately be deciding to irradiate the node or not).

PET positive nodes will be determined by nuc med (rads) doc based on increased FDG uptake relative to blood pool (appropriate for our respective nodal station). We are combing FDG uptake with size (will merge pretreatment planning CT (what rad onc uses, 2mm slice thickness) with PET)

 

[Dave1980]

The experimental cohort is patients with PET positive nodes in the particular nodal station. Negative control cohort is patients with same cancer, with PET negative nodes (determined via nuc med doc + rads) in that same nodal station. We merge PET scan with pretreatment planning CT ( 2 mm slice thickness). Measurements including short-axis, long-axis, L/S ratio, and nodal volume (using rad onc treatment planning system) to determine optimal criteria on the basis of ROC analysis.

PET positive nodes = increased FDG uptake relative to blood pool, since a SUV cutoff has been shown to not be reliable

The main goal is to make the case that for this particular cancer for this particular nodal station, the 10 mm short-axis cutoff is too big. ROCs give us about 6 mm short-axis (73% sensitivity, 90% specific, relative to PET avidity). Also if the rad onc doc is examing the node to giving irradiation too, they can use nodal volume which is considerable more sensitivity (what we found) than short-axis (80% sensitive, 92% specific, relative to PET.

With your thoughts on nodes swell and shrink, is this still the case for nodes draining cancer?

Thanks

It sounds like you are describing a PET/CT which is the standard of care here in the US. Are you not from the US? It just seems odd that you are describing combining the PET with the pre-op CT as a novel idea.

It's not a bad study to look at the dimensions of PET+ and PET- lymph nodes and describe what you think a cutoff should be. The downside of this project is that it has been extensively researched and we all know that the cutoff misses a ton of positive nodes. A lower cutoff wouldn't be very specific although would be more sensitive. In the end I don't see anyway of the patient avoiding needing a PET/CT.