Correct MIMIC image availability and clarify derived data guidance

README.md

@@ -54,7 +54,7 @@ The repository's pre-existing license metadata is retained. It does not grant a
 
 - ODIR-5K: obtain the data and applicable terms from the [official ODIR challenge](https://odir2019.grand-challenge.org/dataset/).
 - IU-Xray: follow the [Open-i collection instructions](https://openi.nlm.nih.gov/faq#collection), including its request not to share the collection outside the research group/organization.
-- MIMIC-CXR: each researcher must obtain [PhysioNet credentialed access and sign the data use agreement](https://physionet.org/content/mimic-cxr/2.1.0/). Public redistribution is not permitted by the source agreement. Report-derived patient-level features and annotations should not be treated as unrestricted substitutes for the original records.
+- MIMIC-CXR: each researcher must obtain [PhysioNet credentialed access and sign the data use agreement](https://physionet.org/content/mimic-cxr/2.1.0/). The agreement restricts sharing access to the data, and [PhysioNet's guidance on derived resources](https://physionet.org/news/post/mimic-derived-datasets-models/) directs authors to share them through PhysioNet under the source agreement. Preprocessing alone does not establish permission for unrestricted redistribution.
 
 ## Citation
 

mimic-cxr/README.md

@@ -2,7 +2,7 @@
 
 Source: [MIMIC-CXR on PhysioNet](https://physionet.org/content/mimic-cxr/2.1.0/) and [MIMIC-CXR-JPG](https://physionet.org/content/mimic-cxr-jpg/2.1.0/).
 
-MIMIC-CXR is credentialed data. Each user must obtain access and sign the source data use agreement. The agreement prohibits sharing the data. This public repository contains no MIMIC images, reports, row-level labels, identifiers, prompt embeddings, or other patient-level records.
+MIMIC-CXR is credentialed data. Each user must obtain access and sign the [source data use agreement](https://physionet.org/content/mimic-cxr/view-license/2.1.0/). PhysioNet's [guidance for MIMIC-derived datasets](https://physionet.org/news/post/mimic-derived-datasets-models/) also directs authors to share derived resources through PhysioNet under the source agreement. Preprocessing does not by itself establish permission for unrestricted redistribution. This public repository currently contains preparation documentation, with no MIMIC images, reports, row-level labels, identifiers, or prompt embeddings.
 
 After obtaining the source files through PhysioNet, run the converter locally from the TriPAH code checkout:
 
@@ -13,7 +13,9 @@ python -m dataset.builders.convert_mimic_dataset \
 
 Inspect the converter's expected metadata, split, CheXpert and NegBio CSV files, record/study lists, and image/report directory layout before running. Complete instructions are in the [TriPAH code repository](https://github.com/Medical-Multi-Agent-AI/TriPAH#prepare-the-data).
 
-The current local metadata artifact contains 216,320 aligned rows, 14 labels, and float32 `[216320,77,512]` main prompt embeddings. Its local image location was not populated during this release audit, so it is not a complete runnable image dataset and is not presented as one here. Obtain and verify the required images through PhysioNet before training.
+The current prepared artifact contains 216,320 aligned rows, 14 labels, and float32 `[216320,77,512]` main prompt embeddings. All 216,320 referenced JPEGs were located and verified as nonempty files in the separately stored source image collection; together these indexed images occupy 6,810,528,693 bytes. This verifies file availability, not pixel decoding or a new de-identification review. The earlier statement that the images were unavailable resulted from checking an empty project-local placeholder before resolving the external image directory.
+
+For local use, `dataset/mimic-cxr/images/` must resolve to the extracted image root. Preserve the index-relative hierarchy below that root; do not flatten image filenames. A directory symlink can avoid duplicating the image collection. Check for an existing file or directory before creating a link.
 
 The stored source index includes official split information, but the current TriPAH loader uses its own seeded sample-wise permutation. Select and document the intended evaluation protocol; do not assume the official source split or patient-disjoint separation is automatically preserved.