Resources

Code for generating the HAIM multimodal dataset of MIMIC-IV clinical data and x-rays

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.

CheXStruct is an automated pipeline that derives structured diagnostic reasoning steps from chest X-rays. CXReasonBench builds on this to evaluate whether models perform clinically grounded, multi-step reasoning beyond final diagnoses.

This dataset includes bounding box-statement pairs for chest X-ray images, derived from radiologists’ eye-tracking data (for explainability) and annotations, for local visual-language models.

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.

VinDr-CXR: An open dataset of chest X-rays with radiologist's annotations

CheXmask Database is a 657,566 uniformly annotated chest radiographs with segmentation masks. Images were segmented using HybridGNet, with automatic quality control indicated by RCA scores.

We present a large-scale CXR VQA dataset derived from MIMIC-CXR with 42M QA pairs, featuring hierarchical answers, bounding boxes, and structured tags. We generated QA-pairs using LLM-based extraction from radiology reports and localization models.

A multimodal clinical dataset consisting of CXRs, ECGs, and blood labs, designed to evaluate Symile, a simple contrastive loss that accommodates any number of modalities and allows any model to produce representations for each modality.

A chest x-rays segmentation dataset derived from MIMIC-CXR based on deep learning algorithm and human examination.