Dark-Field X-Ray Imaging Significantly Improves Deep-Learning based Detection of Synthetic Early-Stage Lung Tumors in Preclinical Models
2510.27679v1
physics.med-ph, cs.CV, cs.LG, eess.IV, physics.optics
2025-11-04
Авторы:
Joyoni Dey, Hunter C. Meyer, Murtuza S. Taqi
Abstract
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.