Paper

Dark-Field X-Ray Imaging Significantly Improves Deep-Learning based Detection of Synthetic Early-Stage Lung Tumors in Preclinical Models

Taqi, Murtuza S, Meyer, Hunter C, Dey, Joyoni

2025

Overview

Low-dose computed tomography (LDCT) is the current standard for lung cancer screening, yet its adoption and accessibility remain limited. Many regions lack LDCT infrastructure, and even among those screened, early-stage cancer detection often yield false positives, as shown in the National Lung Screening Trial (NLST) with a sensitivity of 93.8 percent and a false-positive rate of 26.6 percent. We aim to investigate whether X-ray dark-field imaging (DFI) radiograph, a technique sensitive to small-angle scatter from alveolar microstructure and less susceptible to organ shadowing, can significantly improve early-stage lung tumor detection when coupled with deep-learning segmentation. Using paired attenuation (ATTN) and DFI radiograph images of euthanized mouse lungs, we generated realistic synthetic tumors with irregular boundaries and intensity profiles consistent with physical lung contrast. A U-Net segmentation network was trained on small patches using either ATTN, DFI, or a combination of ATTN and DFI channels. Results show that the DFI-only model achieved a true-positive detection rate of 83.7 percent, compared with 51 percent for ATTN-only, while maintaining comparable specificity (90.5 versus 92.9 percent). The combined ATTN and DFI input achieved 79.6 percent sensitivity and 97.6 percent specificity. In conclusion, DFI substantially improves early-tumor detectability in comparison to standard attenuation radiography and shows potential as an accessible, low-cost, low-dose alternative for pre-clinical or limited-resource screening where LDCT is unavailable.

• Publisher: Cornell University Library, arXiv.org
• Subject: Accessibility / Attenuation / Computed tomography / Deep learning / Medical imaging / Medical screening / Radiographs / Sensitivity / Tumors