Results for Binary Categorical Variables in Correlational Tractography

[Brady Williamson]

Hi Frank,

I am running an analysis that involves testing a few different binary outcomes and I have concerns about how they are being handled in the statistical analysis. I noticed for every binary variable, the track count for the lower valued group is exponentially lower and rarely contains significant findings. To test this further, I ran two separate analyses using dummy coded columns and found the same thing was happening and contradictory results. In the examples in the screenshot, I have a variable indicating presence of chronic lacunar infarcts (LI). As you can see, when LI =1, there are no results for LI = 0 and an expected track length distribution for LI = 1. If I keep the exact same settings and data, only switching the study variable to No LI = 1, the results contradict the previous finding. In other words, the results are the same both when LI=0 and when No LI=0, which is not possible. I would expect the track length distribution to mirror each other. For example, I would expect to see the same curves in blue when running the reverse dummy coded variable (No LI = 1 instead of LI = 1). This situation also applies if the coding is 1 and 2 instead of 0 and 1. Hopefully, that makes sense. Please let me know if you have any questions or if I can provide more info. 

Thanks for your help!
Brady

[Brady Williamson]

I wanted to add more information to better understand why this might be happening. Here is the methods description:
MRI Acquisition

A DTI diffusion scheme was used, and a total of 64 diffusion sampling directions were acquired. The b-value was 1000 s/mm². The in-plane resolution was 2.6 mm. The slice thickness was 2.6 mm. The accuracy of b-table orientation was examined by comparing fiber orientations with those of a population-averaged template (Yeh et al. Neuroimage, 2018). The diffusion data were reconstructed in the MNI space using q-space diffeomorphic reconstruction (Yeh et al., Neuroimage, 58(1):91-9, 2011) to obtain the spin distribution function (Yeh et al., IEEE TMI, ;29(9):1626-35, 2010). A diffusion sampling length ratio of 1.25 was used. The output resolution in diffeomorphic reconstruction was 2.6 mm isotorpic. The restricted diffusion was quantified using restricted diffusion imaging (Yeh et al., MRM, 77:603–612 (2017)). The tensor metrics were calculated using DWI with b-value lower than 1750 s/mm².

Correlational Tractography Analysis

Correlational tractography (Yeh, et al. Neuroimage 245 (2021): 118651) was derived to visualize pathways that have nqa correlated with LI. A nonparametric Spearman correlation was used to derive the correlation. The statistical significance of the correlation was examined using a permutation test (Yeh et al. NeuroImage 125 (2016): 162-171). Subjects with Side.of.the.ICH is 1 were selected. A total of 42 subjects were included in the analysis. A T-score threshold of 2.5 (effect size = 0.27) was used in the fiber tracking algorithm (Yeh et al. PLoS ONE 8(11): e80713, 2013). Cerebellum was excluded. An ROI was placed at hgb_right_hemi (25,48,44). A seeding region was placed at whole brain (39,53,40). The tracks were filtered by topology-informed pruning (Yeh et al. Neurotherapeutics, 16(1), 52-58, 2019) with 8 iteration(s). An FDR threshold of 0.05 was used to select tracks. To estimate the false discovery rate, a total of 8000 randomized permutations were applied to the group label to obtain the null distribution of the track length.

Also, the analysis works as I would expect for continuous variables with the same settings, which is another reason I thought it might be due to how binary categorical variables are handled. 

Thanks!

[Frank Yeh]

Thanks for providing the details.

I will review the code to see possible causes.

Frank

--
You received this message because you are subscribed to the Google Groups "DSI Studio" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dsi-studio+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/dsi-studio/08c5d313-e746-4d14-97c5-6dddaad7cb99n%40googlegroups.com.