Resources

Early detection of heart failure is vital for improving outcomes. The dataset contains 71,589 CXRs paired with gold standard labels from echocardiograms to enable the training of models to detect pathologies indicative of early stage heart failure.

Chest x-rays in JPG format with structured labels derived from the associated radiology report.

RadCoref is a small subset of MIMIC-CXR with manually annotated coreference mentions and clusters. Based on the annotated data, we fine-tuned a deep neural model and used it to annotate the whole MIMIC-CXR dataset. Both data are available.

COVID Data for Shared Learning (CDSL) is a multimodal database comprising de-identified structured health data and radiological images from 4,479 patients with COVID-19, as a comprehensive toolkit for developing predictive models.

Chest radiographs in DICOM format with associated free-text reports.

The MS-CXR-T is a multimodal benchmark that enhances the MIMIC-CXR v2 dataset by including expert-verified annotations. Its goal is to evaluate biomedical visual-language processing models in terms of temporal semantics extracted from image and text.

This dataset was a collected using an eye tracking system while a radiologist interpreted and read 1,083 public CXR images. The dataset contains the following aligned modalities: image, transcribed report text, dictation audio and eye gaze data.

COVID Data for Shared Learning (CDSL) is a multimodal database comprising de-identified structured health data and radiological images from 4,479 patients with COVID-19, as a comprehensive toolkit for developing predictive models.

Chest radiographs in DICOM format with associated free-text reports.

CARMEN-I is a Spanish corpus of 2,000 clinical records from Hospital Clínic, Barcelona. It covers COVID-19 patients and comorbidities, serving as a resource for training clinical NLP models and researchers in NLP applied to clinical documents.