Curated Thoracic Subset from MultiCaRe for Multi‑Label Chest X‑Ray Disease Classification

This dataset is a curated, thorax‑focused subset derived from the MultiCaRe clinical case dataset (https://zenodo.org/records/13936721). All images are chest X‑rays (radiology, x‑ray, thorax). The subset is intended for multi‑label classification of thoracic diseases. The curation process included:- Filtering the original MultiCaRe data to retain only X‑ray images of the thorax.- Extracting 16 thoracic pathology labels (binary) from case reports using a negation‑aware NLP pipeline.- Preserving full case texts, captions, and metadata for reproducibility. Labels (16 primary):pneumonia, pleural_effusion, atelectasis, pneumothorax, cardiomegaly, pulmonary_edema, lung_mass_or_cancer, interstitial_lung_disease, bronchitis, emphysema, tuberculosis, covid19, congestive_heart_failure, heart_failure, congenital_heart_disease, asthma. Additional labels extracted (e.g., anemia, seizure, stroke) are also included in some files but were not used as primary targets in the benchmarking paper. Files included:- case_images_filtered.parquet- cases_filtered.parquet- metadata_filtered.parquet- abstracts_filtered.parquet- captions_filtered.csv- merged_all_cleaned.parquet- merged_all_cleaned_Preserves_all_original_columns- merged_all_cleaned_scanned_per_row_labels- merged_all_cleaned_scanned_term_counts- merged_all_cleaned_scanned_ngram_candidates_unigrams- merged_all_cleaned_scanned_ngram_candidates_bigrams- merged_all_cleaned_scanned_ngram_candidates_trigrams- merged_all_cleaned_scanned_cooccurrence_matrix- duplicate_files_exact- duplicate_files- merged_all (original pre‑cleaning merge) All filtered files are derived from the original MultiCaRe dataset by retaining only X‑ray thoracic cases and images. The original MultiCaRe dataset is released under CC BY‑NC‑SA 4.0. This subset is distributed under the same license. Please cite this subset and the original MultiCaRe dataset when using it. For more details, see the README included in the ZIP file .