DDx-TRACE is a EuroRad-derived neuroradiology benchmark for evaluating multimodal differential-diagnosis reasoning and evidence tracing.
data/eurorad_neuro_01_release.json: source nested manifest used by the benchmark code.data/cases.csv: one row per case.data/images.csv: one row per image/subfigure.data/diagnostic_steps.csv: one row per diagnostic trajectory step.data/imaging_examinations.csv: one row per imaging examination/evidence unit.data/DDx-TRACE-images.tar.zst: image archive; paths match images.csv:image_path and the JSON manifest images[].path fields.ATTRIBUTION.tsv: image-level attribution metadata.croissant.json: Croissant metadata with core and Responsible AI fields.This dataset is derived from public EuroRad cases and is released under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International. Per-case and per-image attribution metadata are provided in data/eurorad_neuro_01_release.json, data/images.csv, and ATTRIBUTION.tsv.
This dataset is intended for research evaluation of multimodal diagnostic reasoning, evidence tracing, rubric-based scoring, and dataset documentation/provenance workflows.
This dataset is not intended for clinical decision support, patient diagnosis, triage, treatment planning, or commercial medical product validation.
The dataset contains health-related medical images and public case narratives. It includes limited patient descriptors such as age, sex, and country where available. Although the source cases are public and intended to be de-identified, residual re-identification risk cannot be ruled out.
DDx-TRACE is not population-representative. It inherits case-report selection bias, publication bias toward educational or unusual cases, geographic/institutional skew from EuroRad contributors, modality availability bias, and annotation-schema bias from the DDx-TRACE construction process.
Citation information should be replaced with the NeurIPS/OpenReview citation after acceptance or camera-ready release.
DDx-TRACE is a EuroRad-derived neuroradiology benchmark for evaluating multimodal differential-diagnosis reasoning and evidence tracing.
data/eurorad_neuro_01_release.json: source nested manifest used by the benchmark code.data/cases.csv: one row per case.data/images.csv: one row per image/subfigure.data/diagnostic_steps.csv: one row per diagnostic trajectory step.data/imaging_examinations.csv: one row per imaging examination/evidence unit.data/DDx-TRACE-images.tar.zst: image archive; paths match images.csv:image_path and the JSON manifest images[].path fields.ATTRIBUTION.tsv: image-level attribution metadata.croissant.json: Croissant metadata with core and Responsible AI fields.This dataset is derived from public EuroRad cases and is released under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International. Per-case and per-image attribution metadata are provided in data/eurorad_neuro_01_release.json, data/images.csv, and ATTRIBUTION.tsv.
This dataset is intended for research evaluation of multimodal diagnostic reasoning, evidence tracing, rubric-based scoring, and dataset documentation/provenance workflows.
This dataset is not intended for clinical decision support, patient diagnosis, triage, treatment planning, or commercial medical product validation.
The dataset contains health-related medical images and public case narratives. It includes limited patient descriptors such as age, sex, and country where available. Although the source cases are public and intended to be de-identified, residual re-identification risk cannot be ruled out.
DDx-TRACE is not population-representative. It inherits case-report selection bias, publication bias toward educational or unusual cases, geographic/institutional skew from EuroRad contributors, modality availability bias, and annotation-schema bias from the DDx-TRACE construction process.
Citation information should be replaced with the NeurIPS/OpenReview citation after acceptance or camera-ready release.