Pls help me ....

SDN

New Member
#1
CT measurement
9.8 mm²
3.8 mm²
6.0 mm²
0.9 mm²
5.4 mm²
1.4 mm²
16.2 mm²
19.5 mm²
2.1 mm²
0.94 mm²
43.80 mm²
3.9 mm²
1.7 mm²
Histopathology
7.6 mm²
1.7 mm²
2.4 mm²
0.49 mm²
4.7 mm²
1.4 mm²
12.20 mm²
10.8 mm²
3.27 mm²
2.16 mm²
28.3 mm²
2.0 mm²
1.4 mm²

Hi friends,
I have two sets of data. One is CT scan measurement and other is Histopathological measurement of same tissue. I want to correlate CT measurement with Histopathology measurement. Could you please tell me which test would be appropriate here? Person's, spearman, others?
 

Miner

TS Contributor
#3
Always start with graphing your data on a scatter plot to see whether there is a potential relationship, and whether it is linear, monotonic or nonlinear. Then you may choose a correlation coefficient appropriate to the relationship. Your data appears linear. Therefore, a Pearson correlation would be appropriate. If you then desire a model, you could perform a least squares regression.

One caution, your data is mostly from the lower range and becomes sparser in the higher range. Consequently, there is one data pair that will have a high influence on the results. If possible, you should obtain additional data for the higher end.

1594816093261.png
 

katxt

Active Member
#5
I agree with Miner about the graph, and the comment about the far right point. The "stretched out " look often suggests a log log transformation.
Now the graph looks a little better behaved. Just a suggestion. kat
1594861801576.png
 

obh

Active Member
#6
Hi Kat :)

For me, the linear look better ...
I believe that when the linear look good, you shouldn't look for a more sophisticated method, even if it has a minor improvement
 

SDN

New Member
#7
As I can't obtain or drop data of high range which statistical test might be appropriate for my study? Is Pearson's correlation only an option?
 

obh

Active Member
#9
Hi SDN
I assume that when you run a simple regression with only one IV, the F test for the regression also tests the correlation between the DV and the IV
 

SDN

New Member
#10
Hi Obh,
thanks a lot for your response. Really helpful. I am measuring same parameters(tissue area) with two different modalities (CT and histology). Now which of them are my IV or DV ? Both looks independent to each other?
 

obh

Active Member
#11
Hi SDN,

Please try both options and see the values of R and p-value in both cases (hint, you can guess the answer...)

PS what method is more reliable? histology?
 

obh

Active Member
#13
Histology is gold standard (reference standard)
So if for example, the Histology is the closure measurement for the "correct" measurement you may define Histology as the dependent variable
and the CT as the IV, so you will be able to "fix" the CT result.
Histology = 0.4442+0.6293*CT

But from the linear regression perspective, the least square will be minimized the same way also if the CT will be the DV
 

SDN

New Member
#15
Hi,
I have three data sets of CT measurements measured by three different radiologists and one histological measurement ( as reference standard) for same tissue. Any suggestions?
 

obh

Active Member
#19
The variance between the 3 radiologists is disturbing ...
Maybe you should ask each radiologist to evaluate the list several times ...(in a different day)?