 |
 |
Hi David, the .txt output is a bit complicated to read, but it is “sensible”. Here is what I mean: Basically, for each component, after the number, you have the classification provided by FIX (signal, Unknown, Unclassified noise). The true/false after that refers to what FIX would do when removing the bad components: False = not to be removed, True = To be removed. So the components you were judging non-artefactual would actually not be removed by FIX (being either Signal or Unknown). :) The last line is the list of components that are classified as noise and in the example below it’s not empty, so if you want to check that those are actually bad, we can say that the classification itself is not the problem. I guess that the issue is the calculation of TPR and TNR. A few of things you can check: - In your hand_labels_noise.txt file, make sure that in the list of components the numbers are separated by a comma AND a space. - Make sure that you didn’t change/move/rename/delete the folder containing the components, as FIX will look at the file filtered_func_data.ica/melodic_IC to calculate the indices. - Have a look if the TPR is calculated correctly, i.e. number of signal components correctly detected (present both in your labels and in FIX .txt file), divided by the number of signal components in your labels. - Have a look at the log files .fixlist and .Rlog1 to see if they contain any error. Hope it helps, Ludovica — Ludovica Griffanti, PhD Analysis Postdoctoral Research Assistant Oxford Centre for Functional MRI of the Brain (FMRIB) Nuffield Department of Clinical Neurosciences, University of Oxford John Radcliffe Hospital Oxford, OX3 9DU, UK email: [log in to unmask] On 8 Jul 2016, at 12:01 pm, David Watson <[log in to unmask]> wrote: > Hi Ludovica > Thanks for your response. When I look in the txt files you mentioned they all look something like this: > > filtered_func_data.ica > 1, Unknown, False > 2, Unknown, False > 3, Unknown, False > 4, Unclassified Noise, True > 5, Unknown, False > 6, Unknown, False > 7, Unclassified Noise, True > 8, Unclassified Noise, True > 9, Unknown, False > 10, Unknown, False > 11, Unclassified Noise, True > 12, Signal, False > 13, Signal, False > 14, Unknown, False > 15, Unknown, False > 16, Unknown, False > 17, Signal, False > 18, Unknown, False > 19, Unknown, False > 20, Unclassified Noise, True > 21, Signal, False > 22, Unclassified Noise, True > 23, Signal, False > 24, Unknown, False > 25, Signal, False > 26, Unknown, False > 27, Unknown, False > 28, Signal, False > 29, Signal, False > 30, Unknown, False > 31, Signal, False > 32, Unclassified Noise, True > 33, Unclassified Noise, True > 34, Unclassified Noise, True > 35, Signal, False > [4, 7, 8, 11, 20, 22, 32, 33, 34] > > Does this look sensible? I’m not sure I completely understand the labelling. By the way the data I’m looking at is from a simple signal detection task. > For this example 14,17,19,21-26,28,29 and 31 all look to me like legitimate non-artefactual components - which does not seem to be reflected in this output L > > Regards David > > From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Ludovica Griffanti > Sent: 08 July 2016 10:47 > To: [log in to unmask] > Subject: Re: [FSL] Training Data for FIX > > Dear David, > getting 0 as TNR could mean that all your components have been labelled as signal (no correct noise identification). > Have a look within each subject’s folder, where you will find the .txt file of FIX classification for each threshold tested (e.g. fix4melview_Training_LOO_thr1.txt, … fix4melview_Training_LOO_thr50.txt) > Here you will see if FIX actually classified everything as signal. > > Otherwise, if these files look sensible to you, look for other errors in the following files within the output directory created: > .fixlist --> should contain the list of subjects included in the training dataset (to check if they've been all loaded/recognised properly) > .Rlog1 --> contains errors from R about the generation of the .RData file > > Hope it helps, > Ludovica > > — > Ludovica Griffanti, PhD > Analysis Postdoctoral Research Assistant > Oxford Centre for Functional MRI of the Brain (FMRIB) > Nuffield Department of Clinical Neurosciences, University of Oxford > John Radcliffe Hospital > Oxford, OX3 9DU, UK > email: [log in to unmask] > > On 8 Jul 2016, at 10:26 am, David Watson <[log in to unmask]> wrote: > > > Hi Enrico > Thanks for getting back. Your suggestion has removed the Nans but I now get zeros in their place as shown below. > > 84.6 0.0 82.1 0.0 74.4 0.0 69.2 0.0 59.0 0.0 48.7 0.0 43.6 0.0 43.6 0.0 > 93.1 0.0 89.7 0.0 82.8 0.0 75.9 0.0 69.0 0.0 62.1 0.0 55.2 0.0 41.4 0.0 > 88.2 0.0 79.4 0.0 73.5 0.0 67.6 0.0 55.9 0.0 52.9 0.0 32.4 0.0 26.5 0.0 > 84.4 0.0 78.1 0.0 65.6 0.0 50.0 0.0 46.9 0.0 40.6 0.0 34.4 0.0 28.1 0.0 > 72.4 0.0 62.1 0.0 55.2 0.0 48.3 0.0 44.8 0.0 44.8 0.0 31.0 0.0 31.0 0.0 > 95.0 0.0 90.0 0.0 85.0 0.0 80.0 0.0 62.5 0.0 57.5 0.0 47.5 0.0 42.5 0.0 > 47.6 0.0 38.1 0.0 28.6 0.0 14.3 0.0 14.3 0.0 14.3 0.0 14.3 0.0 9.5 0.0 and so on > > Something is still wrong :( > > David > > -----Original Message----- > From: Enrico Premi [mailto:[log in to unmask]] > Sent: 08 July 2016 00:32 > To: [log in to unmask]; David Watson > Subject: Re: Training Data for FIX > > Hi David, > I had exactly the same problem one week ago. I solved it by removing the space at the end of the hand label txt. That's what I mean: > > for each subject, in its hand label txt you have: > > [1,3,5,9,11,13,etc...] > > after the second bracket (that close the series of noise components) don't leave any space and don't press Enter. You should have the cursor blinking just after the second bracket, without a second text line. > > Best, > > Enrico Premi, MD > Neurology Department, University of Brescia, Italy > > ___________________________________________________________ > This email has been scanned by MessageLabs' Email Security System on behalf of the Brighton & Sussex Medical School. For more information see: > https://staff.brighton.ac.uk/is/computing/Pages/Email/spam.aspx > > ___________________________________________________________ > This email has been scanned by MessageLabs' Email Security System > on behalf of the Brighton & Sussex Medical School. For more information see: > https://staff.brighton.ac.uk/is/computing/Pages/Email/spam.aspx > > > ___________________________________________________________ > This email has been scanned by MessageLabs' Email Security System > on behalf of the Brighton & Sussex Medical School. For more information see: > https://staff.brighton.ac.uk/is/computing/Pages/Email/spam.aspx > > ___________________________________________________________ > This email has been scanned by MessageLabs' Email Security System > on behalf of the Brighton & Sussex Medical School. For more information see: > https://staff.brighton.ac.uk/is/computing/Pages/Email/spam.aspx