Updates

One viewer for every kind of image and recording, a quality check of every anatomical and diffusion image, MRI series named from how they were acquired, and a preview of the dataset before it is written. Some of these change what a conversion writes, so read Before you upgrade at the end of this card.

New

• A preview of the dataset before it is converted. The panel left of the table, titled by the view it shows, has a third view, BIDS preview, that lists every file the conversion and the metadata step will write, as each converter names them. Select a file to see its predicted content, each value coloured by where it comes from: the scan, BIDS Manager, your own answers, the converter, or a TODO placeholder. The preview runs the conversion's own steps on what the scan found, and a test converts the multimodal sample and requires every predicted file and value to match. See Checking before you convert.

• An Overview in place of the table. Inspection has two views, Table and Overview. The Overview counts what was found, compares subjects with each other (a recording most subjects have and one lacks, an extra one, a label spelled two ways, a label that looks like a name or an ID, sessions on the same day, two recordings claiming one name), and draws a grid of subjects by kind of recording. Show on a finding opens the table on its rows.

• Two more views beside the table. Structure is the tree it was; Recordings lists each kind of recording across subjects, with one checkbox to leave a kind out everywhere. Right-click (Control-click on a Mac) in either for the table's row menu.

• One viewer for every kind of image and recording. Images, PET, spectroscopy, physiological recordings, EEG and MEG now open in one viewer with one set of controls: layouts (one plane, three planes, a large view with small ones, a mosaic, 3-D), overlays and atlases, the time course of a 4-D image with its events, physiological recordings and quality plots aligned beneath it, MEG and EEG with an overview bar, filters, butterfly and scale bars, bad channels and segments written to the dataset in one step, and spectra with automatic phase. How you last looked at each kind of image is remembered. See Viewing data.

• The 3-D view renders on the graphics card, with overlays, the crosshair, picking, and six clip planes that crop or cut away.

• bidsmgr-view opens any image from a terminal, and --render writes a picture of it without opening a window, which works on a server. Save > Show the command line... in the viewer writes the command that reproduces what is on screen.
• A fast quality check of anatomical and diffusion images. Noise and contrast, artefacts in the air around the head, tissue separation, bias, coverage, smoothness, a face still present; for diffusion, the gradient table (flipped directions included), motion, slice dropouts and spikes. It runs in seconds with tools BIDS Manager already brings, docks beside the image, and explains every measure where it is shown. From the viewer's QC menu, Tools > Quality check... in the Editor, after a conversion, or with bidsmgr-qc. Every method and threshold is a setting. See Checking image quality.

• MRI series are named from how they were acquired. A new classifier reads the echo, repetition and inversion times, the flip angle, the readout, the number of volumes, the b-values, the phase encoding, a contrast agent and the protocol, scores every possible name, looks at the session as a whole (single-band references, reversed-phase references, magnitude and phase parts), and decides. On 256 hand-checked series it named 99.2 % right, against 89.1 % before. The table gains a confidence column, and Why this name in Properties shows the evidence and the alternatives. It names MR images only, never spectroscopy or PET. Every number it tests is a setting under Settings > Classification, and bidsmgr-classify scores a table again. See The classifier.

• Fieldmap links that fMRIPrep finds. Fieldmaps are linked to the runs they correct with B0FieldIdentifier and B0FieldSource as well as IntendedFor, grouped the way sdcflows expects. The targets are chosen by shim settings, then by time, and can be changed per fieldmap in Properties. See Fieldmap links.

• Eye tracking. SR Research EyeLink recordings (.edf, .asc) convert to BIDS physiological recordings and events, one pair per eye, with the calibration, and can be paired with the run they were recorded during. See Eye tracking.

• A DICOM header viewer, in the Editor and from the raw tree and the table: search the header, list the frames of an enhanced file, compare two files, and save the header as text or JSON without its identifiers.

• Source data. The conversion can copy each subject's raw files into the dataset's sourcedata/ folder, as they are or as one archive per series, with a checksum list (Settings > Convert, or --sourcedata).

• A log of every run. Each scan, conversion, metadata step and validation writes a log into the project's .bidsmgr/logs/ folder; the history icon of the Log lists them.

• New subject numbers for a second batch that came back numbered from sub-001 again: right-click its rows, Give new subject numbers..., in one step you can undo.
• A preview for every kind of row: fieldmaps as their three files, physiological recordings, spectroscopy as a spectrum, PET, ECAT, EEG and MEG.

• The Editor opens any folder. A BIDS dataset can be opened as the project, or only viewed; any other folder opens to view only, and nothing is written into it.

BIDS-Manager

• Documents in the Editor. README, CHANGES and Markdown files are shown the way a repository shows them, and text files can be edited; .bval and .bvec files show the gradient table (directions on a hemisphere, the b-value of each volume, and a table of volumes).

• Search in sidecars, by field name, value or description.

• Nineteen themes, in four groups, each tested for contrast; your own accent, tint, icon style and file colours over any of them. The GUI uses Inter and JetBrains Mono, which come with BIDS Manager, so it looks the same on every system.

Fixed

• TODO placeholders follow the setting. The scope you chose for placeholders was not saved, unticking the option still wrote them, and a narrower scope never took back what a wider one had written. The dataset now holds exactly what the setting says; your own values and n/a are never touched.
• Each file is asked only for what applies to it. Requirement levels are read from the standard for each file: a T1w without a task is not asked for TaskName, and Authors is not asked for when the dataset has a CITATION.cff. No placeholder is written into a field that names a file, which made every MEG recording fail validation on AssociatedEmptyRoom.
• Spectroscopy sidecars get their repairs, answers and placeholders. The lists of datatypes predated mrs; they are now read from the standard.
• 2-D slices are drawn to scale. A scan with thick slices looked flattened, because slices were fitted by voxel count rather than in millimetres.
• A scan remembers where its data is. Choosing another raw folder could overwrite the open scan's record of where its data was. Moved data is now detected, and Locate... relinks it by series.

Changed

• The Converter's bottom panel is the Log alone: the old preview tree and the one-line statistics are replaced by the BIDS preview and the Overview.
• Side panels fold from a strip along their edge; panels are drawn as cards, and every divider has a grip.
• The scan table has 66 columns: the fieldmap targets and three classification columns are added at the end.
• bidsmgr-metadata --fill-scope accepts none.
• New dependencies, installed with BIDS Manager: SciPy, and eyelinkio for eye tracking. On Windows the EyeLink reader needs the Microsoft Visual C++ 2008 runtime, which the Windows installer adds.

Before you upgrade

• Some MRI series get a different name than in 1.4.1, because the classifier is applied by default. To keep the old names, set Use the acquisition parameters under Settings > Classification to Only to report or Not at all, or scan with --classifier shadow.
• Fieldmaps carry B0FieldIdentifier and B0FieldSource. bidsmgr-convert --no-b0-field-identifier writes IntendedFor alone, as before.
• Placeholders are removed when the option is off. With Write TODO into missing fields off, the metadata step removes the TODOs an earlier run wrote.
• The issues known in 1.5.0 are listed, with what to do about each, in Known issues.

A head scan contains a face, and a face can be rendered from one. BIDS Manager can now remove it, during the conversion itself or afterwards, keep the original so the operation can be undone, and show you the before and the after side by side so you can confirm that the face went and the brain did not.

New

• Defacing, in three places. Tick Deface in Settings and every conversion removes faces from anatomical and PET images before the subject is written, so the identifiable image never enters the dataset at all. Or do it afterwards, from Tools > Deface on a whole dataset or a selection, or by right-clicking any image in the tree. Same engine in all three; only the moment differs. The engine is niimath, whose program ships inside BIDS Manager, so there is no FSL, no FreeSurfer and no ANTs to arrange.

• Skull stripping, and it writes a derivative. Tools > Remove the skull keeps only the brain, using either the mindgrab network run through brainchop, or a mask from the same atlas the defacer uses. The result goes to derivatives/bidsmgr-skullstrip/ as desc-brain, with the dataset_description.json that makes the folder a real derivative dataset, and your raw scan is left exactly as it is. A skull-stripped image is not raw data, and a raw tree that quietly contains one is a dataset nobody can interpret. Overwriting in place is offered, and is not the default.
• A synchronised before-and-after viewer. Every defacing tool tells you to inspect the result and then leaves you to find your own viewer. This one is built in: two images side by side, driven as one. The crosshair, the slice, the plane, the volume, the 4-D graph, the 3-D camera, the rendering effect and the cut plane all stay together, so what you see is the difference between the two images rather than the difference between two viewers.

• Compare any two images, not only a defacing pair. Tools > Compare images puts any two NIfTI files side by side: raw against preprocessed, two echoes, a derivative against the scan it came from, one subject against another. Images of different sizes work too, because the crosshair is matched by position in the scanner rather than by voxel index.

• Choosing an image no longer sends you to the file dialog. The picker lists the dataset's own images, in the dataset's own shape, with a filter whose words can come in any order: 014 t1 finds sub-014/…/sub-014_T1w.nii.gz. Derivatives are listed too, since comparing one against its source is the common case, and Browse is still there for a file from outside the dataset.
• Putting the face back. Tools > Put the face back works however many changes came afterwards, restoring from the copy kept in .bidsmgr/ or in sourcedata/, and taking the deidentification entries back out of the sidecar so the image can be defaced again later.
• The viewer opens in the layout you left it in. The plane and the view mode are remembered between sessions, and a new scan opens in the multi-planar view, or the multi-planar 3-D view where there is a GPU, rather than on a single slice.

Fixed

• Checkboxes follow the theme. An unstyled checkbox fell back to the platform control, which does not know about the palette: in the dark theme a plain unchecked box painted a light grey block that read as disabled, or as already ticked. Every checkbox and radio button in BIDS Manager now takes its colours from the theme, in every state.
• A wide figure in the documentation sat half its width off to the left and was clipped at the edge of the browser window.

Changed

• Convert provenance moved to the dataset root. It used to be written into a hidden folder inside every subject, which cost more than it was worth: deleting a subject removed every recording and left the folder standing, because BIDS Manager's own state is excluded from a deletion. It now lives with everything else BIDS Manager writes, at .bidsmgr/provenance/. Datasets converted by an earlier version still have theirs in the subject folder, and the Editor's delete now clears those too.
• The defacing and skull-stripping engines are ordinary dependencies rather than optional extras, because these features are reached from the GUI and somebody who installed by double-clicking an installer has no terminal in which to add one. On Python 3.14 the defacing engine has no wheel yet, so the menu entry is greyed out with the reason; everything else is unaffected.

The defacing guide

The Editor stops being a place you look at a converted dataset and becomes a place you can change its shape. Entities can be added and removed, sessions created and dissolved, and recordings deleted, and every one of those carries the files that point at them along with it.

New

• Add or remove an entity. A recording can gain a label it should have had, or lose one it never needed: an acq- to tell two protocols apart, a run- that was only ever going to be one run. Only entities the standard permits for that kind of file are offered, so an EEG recording is never offered echo-, and a _bold is never allowed to lose its task-, because a file without one has no reading. The new entity is placed where the standard puts it rather than where you typed it.

• Sessions can be created and dissolved. A session is a folder and a label at the same time, which is why moving recordings into or out of one by hand goes wrong so easily. Now it is one action: pick the recordings, name the session, and the folder, the filenames, the *_scans.tsv and every reference move together. Taking the last session out again works the same way.

• Delete recordings, datatypes and sessions. Deleting the files is the easy half. What leaves a dataset broken is everything still naming them, so the same step removes the *_scans.tsv rows, the IntendedFor entries, the participants row of a subject with nothing left, and any folder the deletion empties. A sidecar that sat above several recordings and now describes none of them goes too.

• Everything is one step in the project's history. A deletion that touches forty files is one entry in the Editor's history, not forty, with every file as it was kept in .bidsmgr/ (the GUI has no command to undo it: Known issues).
• The preview looks like your dataset. Before any of this happens you see what it would do, arranged the way the data is: subject, then session, then datatype, then files. Untick anything you want left alone. What follows automatically, the table rows and the references, is listed but not offered as a choice, because a half-applied cross-reference is worse than either outcome.

Fixed

• Renaming from a right-click now offers what you clicked. Choosing to rename a subject from one file used to tick every file that subject has, across every datatype. It now starts with that recording and the files that belong to it, and widening is a click away.
• Choosing to merge no longer discards your selection. Narrowing a rename and then ticking "merge them into one subject" silently reselected everything.
• Merging two subjects leaves nothing behind. The folder that was merged away used to survive as an empty shell. Anything inside it that was not named after the subject, an inherited sidecar, a note you left, stayed with it, and the sidecar then described recordings that had moved. All of it now travels, and the record of how the merged subject was converted is kept beside the target's own rather than overwriting it.
• A rename now carries everything that names the old value. Renaming a task left TaskName in the sidecar saying the old one. Renaming a session left *_sessions.tsv listing a session that no longer existed. Fieldmap and empty-room pointers, and the Sources fields that record what a file was derived from, were left aimed at names nothing had. None of that was reported by any validator, which is what made it worth finding.
• Renaming a subject fixes the whole reference, not half of it. Where a pointer was written in the bids:: form, the file name was updated and the folder in the same path was not, leaving a reference that resolved to nothing.

Changed

• The power-line frequency is no longer guessed. When nothing stated one, EEG and MEG sidecars used to be written with PowerLineFrequency set to 50. That is correct in Europe and wrong across most of the Americas and in western Japan, and nothing in the file distinguished it from a value somebody had measured. Filtering at 50 on 60 Hz data leaves the interference exactly where it was. The field is now written as n/a unless the recording's own header, the scan table, or your metadata template supplies a value. That is still valid, and it is true. If your recordings are 50 Hz, state it once in Dataset metadata and every recording will carry it.
• Deleting is deliberately limited to what sits inside a subject. The dataset description and the Editor's own history are not removable from the GUI, because losing either turns a dataset back into an ordinary folder.

The largest release since 1.0. A whole modality arrives, and the form that asks you for metadata stopped being written by hand and started being built from the standard itself.

New

• PET support. Positron emission tomography is now a first-class modality. PET studies scan, convert and validate alongside everything else, from either of the two formats scanners produce: DICOM, and the ECAT files that Siemens HRRT and older CTI systems write. Hybrid studies convert in one pass. On a PET/MR study the MR half is ordinary anatomical and functional data, and BIDS Manager converts it in the same run. On a PET/CT study the CT half is set aside with a note explaining why: BIDS has no place for CT in a raw dataset yet.

• The metadata a scanner cannot know. A PET sidecar needs around forty fields, and the scanner header holds about half of them. It records how the image was reconstructed, but not how much tracer was injected, in what form, or when. Those now have a home: a Tracer, Radiochemistry, Acquisition and Reconstruction block in Dataset metadata, with per-scan overrides in the properties panel, and the same inherit-from-the-dataset behaviour the EEG and MEG fields already have.

• The scan reads what it can and proposes it. Tracer, radionuclide, injected dose and reconstruction settings are read out of the source and offered as suggestions beside the fields they inform. They are never filled in silently, because a vendor writes these as free text and a wrong tracer is worse than a blank one.

• Bring your dose table. If the radiochemistry already exists as a spreadsheet, point bidsmgr-convert --pet-spreadsheet at it. Column names are matched loosely, so Injected Dose and injected_radioactivity both work.

• A recording can state any field its file may carry. If the standard declares a field for that datatype and suffix, the form offers it, at its real requirement level, with the standard's own description on hover.

• What the conversion already answers is folded away. The scan runs the converter once, quietly, and reads what came out, so the form can separate the fields that will be filled for you from the ones only you can answer. The settled block is still editable: correcting a value there is how you tell BIDS Manager its header was wrong.
• The BIDS version you choose is honoured everywhere. Not only in validation. It decides which fields the form asks for, which entities a filename may carry, and what is written into dataset_description.json.
• Answers take the shape the standard declares. A field typed as a number reaches the sidecar as a number and an array of numbers as an array of numbers, whether you typed it into the form or into a cell in the table.

Fixed

• Scans no longer skip DICOM files with dots in their names. Philips exports name each file after its internal identifier, and GE Signa uses a .img ending. Both are ordinary DICOM, and both were being passed over, whatever the modality. If a past scan of a Philips or GE dataset came back emptier than expected, it is worth rescanning.
• Patient details are stripped from PET sidecars. Name, identifier, birth date and referring physician are removed on conversion. The study and series identifiers stay, so an image can still be traced back to the series it came from.
• Numbers typed into the metadata form stay numbers. A list of numbers, frame times or reconstruction parameters, was being stored as a list of text. Worse, opening the form and saving it rewrote values the conversion had already got right, so filling in the metadata could add errors to a sidecar that had been correct. Both are fixed, and a template already saved that way is repaired the next time it is applied: nothing has to be typed in again.
• A blood table's sidecar is named for the sampling it belongs to. BIDS requires the recording entity on both the table and its sidecar. It was being written on the table only, which left a valid table beside a sidecar that no tool could match to it.
• Acquisition times with a missing leading zero are repaired. A converter can write 11:22:1.995, which is not a time any validator will accept, and it was copied on into scans.tsv, so one slip was reported twice.
• One project's metadata stays inside that project. Answers given for one dataset no longer follow you into the next.
• Windows: the settled metadata block is filled in again. On Windows, "Already answered by the conversion" came up empty for every MRI kind, while the same dataset filled it on macOS and Linux. The values were never missing from the data. The pass that measures them works inside a folder whose path could run past the length Windows allows, and the converter does not fail gracefully there: it stops without a message, and a measurement nobody could take looked the same as a measurement there was nothing to take. Those paths are now short and of fixed length. If that block came up empty for you, rescanning fills it.
• Windows: the converter is found however you started BIDS Manager. The dcm2niix that ships with the Python package carries a different name on Windows, and BIDS Manager looked only for the other one. Started from an activated environment a fallback found it anyway; started from a desktop shortcut or a bundled interpreter it did not, and both the classifier that proposes BIDS names and the pass that measures metadata were skipped without saying so.
• A series whose output does not name itself is no longer dropped. On some acquisitions, a Siemens XA30 localiser among them, the converter writes an identifier into the sidecar that does not match the one the series carries. Those outputs belonged to no row and their measurements were lost. Each is now attributed to the series it was actually converted from.
• A dropdown inside a folded section is wide enough to read. A menu field sized itself while it was still hidden, so a value the conversion had answered could appear as a stub too narrow to show its own text.
• A path too long for Windows now says so. The one failure that produced no message at all now reports the limit and the length of the path that exceeded it, so the next time it happens it is a message rather than a silence.

Changed

• The scan table gained five read-only columns carrying the PET details read from the source. They are appended, so every existing column keeps its position. In the GUI those values are offered inside the fields they inform rather than shown as columns.
• pet2bids is a new dependency, installed with BIDS Manager. Nothing to do on your side.
• A dataset's name now means one thing. It is the folder name, stated once when the project is created, and the GUI says what a rename would affect rather than silently keeping two answers.