Check the app and upload access first
Choose the app and intended analysis, then check its supported input formats and required metadata. Medical imaging workflows are for research and education only. You must accept this restriction and have permission to use the data, including any approvals required for your study. Hosted uploads are not open yet. Supported file formats and account permissions will be confirmed before access opens; public files and recorded examples do not establish upload access.
Keep an identified source and its companions
Preserve the original source, version or accession, attribution and file identities before making a working copy. Keep required headers, calibration, gradient or other companion files with their matching measurements. Similar filenames, equal row counts or a changed file extension do not establish compatible contents. If an import is refused, retain the reason and resolve the missing requirement rather than inventing a header or replacing it with an unrelated file.
Describe what each dimension represents
Record whether a row or array dimension represents time samples, channels, acquisitions, spatial positions, atoms or sequence records. Preserve order and names. An image’s loader array order may differ from its source dimension order; use the recorded geometry instead of relabeling axes by appearance. A table of spatial soundings does not acquire a timebase merely because it opens in an engineering product.
Confirm units, timing and coordinate conventions
Carry sampling rate, physical units, coordinate system, image orientation and sign conventions through import. Changing a sampling-rate label is not resampling. Degrees are not projected metres, and positive-up elevation differs from positive-down depth. Keep calibration separate from later standardization; a transformed output can have different units from its measured input. Use the app-specific guide to check the resolved contract.
Separate measurements from references and selections
Name the measured channel or input separately from annotation events, comparison sequences, masks and reference images. Preserve the selected annotation layer instead of merging labels silently. Record whether an app selects a whole volume, one plane, a channel prefix, the first sequence record or one coordinate model. A dataset name does not select an anatomical region, and a matching shape does not prove spatial correspondence.
Record every preparation change
Retain the operation, settings and order for filtering, scaling, normalization, resampling, coordinate conversion or other preparation. Keep source and prepared identities separately and carry companion changes with the corresponding measurements. Treat missing values separately from measured zeros; preserve warnings and document exclusions or replacements. A transformation used for one named example is not a default recipe for another input.
Review the import before running
Inspect the loader’s actual dimensions, channel or record selection, units, companion identities and warnings. Then record the selected routes and their prerequisites. A successful import establishes that the loader read that representation; it does not establish anatomical accuracy, annotation truth, adequate acquisition geometry or scientific validity. Preserve the input record with the eventual saved results.
Prepare one named input without losing its meaning
These examples link to detailed instructions and exact evidence inventories. Their source, version, access and review limits still apply; this table adds no new analysis or recording acceptance.
| Input requirement | Named example and instructions |
|---|---|
| WFDB companions and annotation layers | BIDMC 01 uses matching .hea, .dat and .breath files. Confirm 60,001 samples, five channels and 125 Hz, then select the measured “RESP,” channel, preserving the comma in its source label. ann1 is the recorded GUI comparison layer; ann2 remains a separate historical comparison. |
| Diffusion acquisitions and gradients | The OpenNeuro example is 128 × 128 × 88 × 67. The fourth dimension contains 67 acquisitions; preserve matching .bval, .bvec and .json files and their volume order. B-values use s/mm². The saved tensor comparison covers 256 sampled locations, not accuracy throughout the volume. |
| Volume selection and spatial metadata | The Spleen input is 512 × 512 × 31 at source and 31 × 512 × 512 in the recorded loader. Preserve geometry and selection. The histogram app defaults to Plane (2-D); the reviewed whole-volume example explicitly uses Volume (3-D), with no spleen-only mask. |
| Coordinate model and atom identity | For the 1CRN PDB example, the parser selects the first coordinate model and highest-occupancy alternate atom location, retaining the first encountered location on a tie. Preserve atom order, identities and angstrom units; this selection is not an ensemble average. |
| Nucleotide input and protein reference | Load the single-record ACTB transcript NM_001101.5, with 1,812 nucleotides. Keep the 375-residue NP_001092.1 protein reference separately. The ORF app uses the first sequence record; a multi-record FASTA does not request batch analysis. |
| Longitude, latitude and elevation signs | The prepared Baja table has 82,970 rows of longitude and latitude in degrees and negative elevation in metres. Its explicit longitude preparation and positive-up elevation import convention must travel with the file. These are spatial soundings, not regularly sampled waveform channels. |
| Signal orientation and selected recording scope | S001R01 EDF has 64 channels and 9,760 samples per channel at 160 Hz. Preserve channel order and source calibration. For the recorded scope, use 64 channels and 61 seconds; smaller limits select the beginning of the source, not an arbitrary channel set or later time window. |