REAME updated for dataset specification preprocessings

[Astarakee]

Description added in the README file, including:

• specific preprocessing steps required for each of the segmentation tasks.

[Copilot]

Pull request overview

Adds a new “Data Preprocessing Specifications” section to the README to document per-track preprocessing expectations (templates/spaces, masking/anonymization choices, and modality sets) so users can align their training/inference pipelines with the official BraTS preprocessing.

Changes:

• Documents a key constraint that voxel intensities are not globally normalized in released volumes.
• Adds an overview table summarizing modalities, reference spaces, and anonymization per challenge track.
• Adds per-task technical detail subsections for the main segmentation tracks.

Suppressed comments (1)

README.md:575

• Parentheses around defacing are unnecessary and make the sentence harder to read; remove them.

* **Spatial Alignment & Anonymization:** Provided in **Native Space** (no template registration) and processed using (**defacing**) rather than skull-stripping.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[neuronflow]

please adress copilot comments, also please refer to the preprocessing wrapper @MarcelRosier created and the settings that should be used ;)
https://github.com/BrainLesion/BraTS/blob/main/brats/preprocessing.py

[PaulaTBz]

The PR introduces additions to README that clearly define all details regarding preprocessing for every single subtask individually. It clearly distinguishes between 2024 or earlier and 2025+ where there have been changes over the years.

The only point that could be clarified:
Is "Cross-Tumor Generalizability (GLiNR)" the same as GoAT ?

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.

Suppressed comments (5)

README.md:545

• Capitalize the sentence in the Notes column.

| **Pediatric Brain Tumors (PED)** | 4 mpMRI (`T1c`, `T1n`, `T2f`, `T2w`) | SRI24 | Defaced | defaced rather than skull-stripped. |

README.md:588

• This SRI24-only claim contradicts README.md:416, which states that GoAT retains the atlas of each source dataset, as well as this table's MNI152 and native-space source cohorts. Users could therefore preprocess GoAT inputs into the wrong space; preserve the source-dependent reference space in this summary.

* **Spatial Alignment & Anonymization:** Standardized to **SRI24 atlas space**. All sub-datasets are skull-stripped, with the exception of the **Pediatric (PED)** cohort which remains defaced.

README.md:556

• This range conflicts with the existing preprocessing notes: README.md:163 limits SRI24 pre-treatment data to 2023 and earlier, while README.md:122 assigns 2024+ post-treatment data to MNI152. Labeling the pre-treatment SRI24 period through 2024 can direct readers to the wrong atlas; use the established 2023 cutoff.

  * **BraTS early challenges–2024:** Co-registered and normalized to the **SRI24 atlas space**, followed by rigid skull-stripping.

README.md:585

• The repository consistently names this task “Generalizability Across Tumors (BraTS-GoAT)” (README.md:380, brats/constants.py:216, and brats/data/meta/goat.yml:3). Introducing “GLiNR” here leaves readers unable to map the preprocessing guidance to GoATSegmenter; use the established task name unless this is a distinct task, in which case that distinction must be documented.

#### 8. Cross-Tumor Generalizability (GLiNR)

README.md:575

• The parentheses around “defacing” make this sentence grammatically malformed; state the processing method directly.

* **Spatial Alignment & Anonymization:** Provided in **Native Space** (no template registration) and processed using (**defacing**) rather than skull-stripping.