deletions | additions       diff --git a/begin_enumerate_item_textbf_can__.tex b/begin_enumerate_item_textbf_can__.tex  index 4ce9224..f488603 100644  --- a/begin_enumerate_item_textbf_can__.tex  +++ b/begin_enumerate_item_textbf_can__.tex 
...
\begin{enumerate}  \item\textbf{can't find  Roche et al 2014} 2014 - was about partial volume estimation using a bayesian MAP formulation that extends the mixel model. Our segmentation results do not account for partial voluming effects. In the text we added:  "Another factor not accounted for in our segmentation results was the effects of partial voluming, which adds uncertainty to tissue volume estimates. In (Roche 2014), researchers developed a method to more accurately estimate partial volume effects using only T1-weighed images from the ADNI dataset, which resulted in higher classification accuracy between Alzheimer’s disease (AD) patients and mild cognitively impaired (MCI) patients from normal controls (NL)."  \item Wolz et al 2014: LEAP algorithm on hippocampal volumes, compared between 1.5T and 3T found small bias (1.17\% mean signed difference) between field strengths on ADNI data. When we look at the scaling factors of our data on hippocampal volumes, we find that the average of scaling factors for the 1.5T scanners is 0.99, while the average of the 3T scanners is 1, showing that the 1.5T scanner volumes, on average, are around 1\% larger than the 3T (with only 4 1.5T scanners, this is not a significant difference), which is very similar to the findings of Wolz et al, though the algorithm used in Wolz et al are much more accurate.  \item Jovicich 2009 - they found that test-retest reproducibility does not change much cross platforms and field strengths. We also found that all our scanners were very reliable, with the except of a few sites with bad reproducibility of the thalamus.   \item Wyman et al. 2013: This paper emphasized the use of ADNI's standardized data sets, which they say add   \begin{enumerate} 
...
In order to enable other researchers to compare our methods, evaluate robustness and replicate our study, we will provide, in the supplemental materials, the raw data on MRI volumes produced from Freesurfer, along with the python and R code to calculate scaling factors, leave-one-out calibration, and between-/within- site ICC.   \item Whitwell et al. 2012: \textbf{CAN'T FIND} Found different rates of hippocampal atrophy in the ADNI cohort than the Mayo Clinic Study of Aging cohort, even though there were no differences in hippocampal volumes between the matched cohorts. This is attributed to sampling of different populations rather than differences in hippocampal volumes due to differences in acquisition parameters. In the text I added:  "The other limitation of this study is that we assumed that subjects across all sites will come from the same population, and that stratification occurs solely from systematic errors within each site. In reality, sites may recruit from different populations, and the true disease effect will vary even more. \textbf{For example, in a comparison study between the matched ADNI cohort and a matched Mayo Clinic Study of Aging cohort, researchers found different rates of hippocampal atrophy even though no differences in hippocampal volumes was detected (Whitwell 2012). This could be attributed to sampling from two different populations.} This added site-level variability requires a larger site-level sample size, for an example of modeling this, see (Han 2011). "  \end{enumerate}