Shadow and Light: Digitally Reconstructed Radiographs for Disease Classification

TL;DR

This work presents DRR-RATE, a large synthetic chest X-ray dataset generated from the CT-RATE corpus to enable paired X-ray images, radiology reports, and 18 pathology labels, including lateral views. Using Siddon-Jacobs ray tracing, DRRs are produced at 512×512 resolution with a -100 HU threshold, and a CC BY-NC-SA license; the dataset contains 50,188 DRRs from 21,304 patients. CheXnet is evaluated on both CheXpert and DRR-RATE, showing strong performance for Cardiomegaly and Pleural Effusion on DRR-RATE, with domain-shift effects observed when models trained on real X-rays are tested on synthetic DRRs. The work demonstrates the viability of CT-derived, text-annotated DRR data for multimodal AI in radiology, while discussing limitations and societal considerations for AI deployment in healthcare.

Abstract

In this paper, we introduce DRR-RATE, a large-scale synthetic chest X-ray dataset derived from the recently released CT-RATE dataset. DRR-RATE comprises of 50,188 frontal Digitally Reconstructed Radiographs (DRRs) from 21,304 unique patients. Each image is paired with a corresponding radiology text report and binary labels for 18 pathology classes. Given the controllable nature of DRR generation, it facilitates the inclusion of lateral view images and images from any desired viewing position. This opens up avenues for research into new and novel multimodal applications involving paired CT, X-ray images from various views, text, and binary labels. We demonstrate the applicability of DRR-RATE alongside existing large-scale chest X-ray resources, notably the CheXpert dataset and CheXnet model. Experiments demonstrate that CheXnet, when trained and tested on the DRR-RATE dataset, achieves sufficient to high AUC scores for the six common pathologies cited in common literature: Atelectasis, Cardiomegaly, Consolidation, Lung Lesion, Lung Opacity, and Pleural Effusion. Additionally, CheXnet trained on the CheXpert dataset can accurately identify several pathologies, even when operating out of distribution. This confirms that the generated DRR images effectively capture the essential pathology features from CT images. The dataset and labels are publicly accessible at https://huggingface.co/datasets/farrell236/DRR-RATE.

Figures (6)

• Figure 1: DRR Synthesis Overview: (right) An idealized model of a projectional radiography imaging system. X-ray beams, $\mathbf{R}(t)$, originating from a fixed position, $\mathbf{s}$, and with constant energy diminish in intensity as they travel through the CT volume, $\mathbf{V}$. The attenuation, with respect to volume, is measured when the X-rays reach a point, $\mathbf{p}$, on the detector plane, resulting in the production of the DRR. (middle) Siddon's method of voxel interpolation. The pixel value at $\mathbf{p}$ represents the weighted average of the intensities of all voxels intersected by the ray. The weight corresponds to the length of the line segment passing through each voxel. $\frac{t_{m+1} - t_{m}}{2}$ denotes the midpoint, while $t_{m}$ and $t_{m+1}$ denote the boundary intersections of the voxel. (left) A rendered DRR from a chest CT volume.
• Figure 2: Histogram of patient age distribution and gender count
• Figure 3: Example DRR image created from . The binary labels are in the following order: Medical material, Arterial wall calcification, Cardiomegaly, Pericardial effusion, Coronary artery wall calcification, Hiatal hernia, Lymphadenopathy, Emphysema, Atelectasis, Lung nodule, Lung opacity, Pulmonary fibrotic sequela, Pleural effusion, Mosaic attenuation pattern, Peribronchial thickening, Consolidation, Bronchiectasis, and Interlobular septal thickening.
• Figure 4: ROC curve for CheXnet trained on CheXpert dataset. ROC scores = {Atelectasis: 0.80, Cardiomegaly: 0.79, Consolidation: 0.84, Lung Lesion: 0.80, Lung Opacity: 0.90, and Pleural Effusion: 0.95}.
• Figure 5: ROC curve for CheXnet trained on DRR-RATE dataset. ROC scores = {Atelectasis: 0.72, Cardiomegaly: 0.92, Consolidation: 0.74, Lung Nodule: 0.66, Lung Opacity: 0.67, and Pleural Effusion: 0.95}. Grey shaded region denotes standard deviations at thresholds.