None of the "brains make minds" claims of neuroscientists depend on imagery, there being no photos that do anything to substantiate either the claim that brains make minds or that brains store memories, So someone might claim that the degree of fake or mislabeled images in neuroscience has no relevance to whether brains make minds. But there is a possible relevance: if there are honesty problems or accuracy problems with many of the images used in neuroscience research, that is another reason for distrusting cognitive neuroscientists; and the less trustworthy cognitive neuroscientists are, the more we should doubt their dogmatic claims.
A recent article at the Retraction Watch site (https://retractionwatch.com/) has the title "Widespread image reuse, manipulation uncovered in animal studies of brain injury." The "image reuse" being referred to is when some scientific paper has an image that it claims is data from new original research done by its authors, even though the image seems to be the same image published much earlier in some other scientific paper.
We read this:
"More than 200 papers on ways to prevent brain injury after a stroke contain problematic images, according to an analysis published today in PLOS Biology...René Aquarius and Kim Wever, of the Radboud University Medical Center in the Netherlands, first noticed these patterns in 2023 when they started working on a systematic review of animal studies in the field...Of the 608 studies they analyzed, more than 240, or 40 percent, contained problematic images...At first, the pair tried to check the images manually, but the work was too slow. So they turned to ImageTwin, which cross-checks uploaded images against a database, making the process 'more efficient and accurate,' Wever said. The results showed a sprawling network of images that not only appeared in articles on early brain injury, but also showed up labeled under different experiment conditions across studies on Alzheimer’s disease, epilepsy and lung cancer, and other unrelated fields....In total, their analysis found 37 of these papers in research fields other than early brain injury. Overall, 133 of the 608 articles contained an image that also appeared in another publication, a pattern typical of paper mills or image reuse among an author group, Aquarius said...In their new publication, the researchers took a conservative approach to identifying image reuse, so Aquarius called the 40 percent estimate a 'best-case scenario.' "
The Retraction Watch article is based on the scientific paper here, entitled "High prevalence of articles with image-related problems in animal studies of subarachnoid hemorrhage and low rates of correction by publishers." That paper states this:
"Estimates of the prevalence of inappropriate image duplication in (biomedical) research remains uncertain and are dependent of the body of literature that is being investigated. Reports are sparse and cover widely different literature samples. Out of >20,000 articles from 40 scientific journals, 4% contained problematic figures [18], while Danish researchers detected inappropriate image duplication in 19% of preclinical depression publications [19]. Image-related issues were identified in 6.1% of the assessed articles published in Molecular and Cellular Biology [20] and in 16% of articles published in Toxicology Reports [21]. Finally, in a sample of articles published in the journal Bioengineered, >25% contained inappropriate image duplication [22]. A synthesis of the sparse data estimates the combined misconduct rate (including fabrication, falsification, and plagiarism) to be 14%, 1 in 7 research articles [23]. The 40% prevalence observed in our study far exceeds these figures, suggesting an alarming level of integrity issues in the preclinical subarachnoid hemorrhage literature."
Sometimes you have imagery malfeasance that is so common in some particular field that some researcher can use an "everybody does it" kind of defense. An example is what goes on in brain scan visuals that purport to show areas of "superior activation" in the brain. There massively occurs misleading visuals that give the impression that the amount of variation is much greater than it is. Typically the variation will be some very small amount such as 1 part in 200. But you will see visuals that try to make such differences look much greater than 1 part in 200.
Page 68 of a scientific paper ("More Than Meets the fMRI: The Unethical Apotheosis of Neuroimages" by Eran Shifferman) has a quote talking about the kind of shady business that goes on when these visuals are produced:
"The time series of voxel changes may be motion-corrected, coregistered, transformed to match a prototypical brain, resampled, detrended, normalized, smoothed, trimmed (temporally or spatially), or any subset of these, with only a few constraints on the order in which these are done. Furthermore, each of these steps can be done in a number of ways, each with many free parameters that experimenters set, often arbitrarily. After preprocessing, the main analysis begins. . In a standard analysis sequence, experimenters define temporal regressors based on one or more aspects of the experiment sequence, choose a hemodynamic response function, and compute the regression parameters that connect the BOLD signal to these regressors in each voxel. This is a whole-brain analysis, and it is usually subjected to one of a number of methods to correct for multiple comparisons… the wholebrain analysis is often the first step in defining a region of interest in which the analyses may include exploration of time courses, voxelwise correlations, classification using support vector machines or other machine learning methods, across-subject correlations, and so on. Any one of these analyses requires making crucial decisions that determine the soundness of the conclusions."
After the quote, the paper author says, "This detailed description shows that BOLD-fMRI NIs [neuroimages] represent mathematical constructs rather than physiological reality (Burock 2009)." Page 70 of the same paper states this:
"A ubiquitous statistical error in functional neuroimaging is the non-independence error (aka double dipping): using the same data for selecting the voxels of interest and then using these voxels for the secondary analysis, the one upon which the functional conclusions are based9 . Double dipping violates random sampling because the test statistics are not inherently independent of the selection criteria of the region of interest, thus statistically guaranteeing the outcome of the second analysis and rendering them useless (Kriegeskorte et al. 2009; Vul et al. 2009). Similarly, as mentioned before, statistical tests in neighboring voxels are not independent of one another, because time series in neighboring voxels are intercorrelated (Peterson 2003). Analyses have shown that the non-independence error is widespread in BOLD-fMRI studies (40-50% of published papers) and that the severity of the distortions of the results presented in these papers could not be assessed. This necessitates replications and reanalysis (Kriegeskorte et al. 2009) or the results of these studies “mean almost nothing”, since they are 'using seriously defective research methods and producing a profusion of numbers that should not be believed' (Vul et al. 2009)."
On page 71 we have this complaint about the use of way-too-small study group sizes in brain scan studies:
"Yet another sizeable statistical concern is unfitting sample sizes: most published fMRI studies have sample sizes that would be considered exceedingly small by conventional standards (Yarkoni 2009; Button et al. 2013; Ingre 2013), if they include sample size calculations at all (Guo et al. 2014). It is established that in fMRI studies, small studies (n=16) fail to reliably distinguish small and medium-large effect sizes from random noise as do larger studies (n=100) (Ingre 2013). However, Wager et al. (Wager et al. 2009) report that across 415 fMRI studies reviewed, the average group size was smaller than 12, with some using only 4 subjects."
Page 73 refers to a "localization project," by which the author means attempts to show that particular brain regions activate more strongly when some type of cognitive activity is performed. We read, "The cumulative effect of these types of data variability is a serious impediment on the localization project, suggesting that there are no macroscopic-level delineations corresponding to cognitive performance, and that they are probably a methodological artifact (Gonzalez-Castillo et al. 2012; Thyreau et al. 2012)."
The author of the scientific paper is apparently suggesting that you cannot actually find any evidence that particular regions of the brain are more active during particular cognitive activities. On page 76 the author refers to those running brain scan studies, saying, "This practice all too often amounts to unethical science, one where the generators of data overlook known shortcomings of their tools of the trade and press forward with producing claims too strong to be supported by exploiting the strong appeal of their meticulously crafted images."
No comments:
Post a Comment