Dual-energy x-ray imaging for AI-based lung tumor localization using foundation models: a phantom study

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Dual-energy (DE) X-ray imaging offers potential for improved tissue contrast.
• AI foundation models are increasingly used for medical image analysis.
• Accurate tumor tracking is crucial for effective treatment planning.

Purpose of the Study:

• To assess AI-based tumor tracking performance using DE X-ray versus high-energy (HE) X-ray.
• To evaluate the impact of DE imaging parameters on tumor localization.
• To compare AI model accuracy with manual ground truth segmentation.

Main Methods:

• DE planar imaging with fast kVp-switching was performed on a phantom.
• Weighted logarithmic subtraction generated soft-tissue-enhanced DE images.
• An AI foundation model was used for frame-by-frame tumor localization and compared to ground truth.

Main Results:

• Optimal weighting factors for bone suppression ranged from 0.57 to 0.79.
• Increased tube current improved DE contrast-to-noise ratio (CNR).
• AI model achieved high localization accuracy (Dice > 0.9, HD95 < 2-3 mm) on both DE and HE images.
• DE imaging significantly reduced the missed tumor fraction compared to HE images (e.g., <15% for DE vs. 50-55% for HE with 80/120 kVp).

Conclusions:

• Integrating DE imaging with AI foundation models enhances tumor localization accuracy.
• This approach is particularly beneficial in anatomically challenging regions with tumor-bone overlap.
• DE imaging shows superior performance in minimizing missed tumor areas compared to HE imaging.