Effects of tube voltage, radiation dose and adaptive statistical iterative reconstruction strength level on the detection and characterization of pulmonary nodules in ultra-low-dose chest CT

Yue Yao, Xuan Su, Lei Deng, JingBin Zhang, Zengmiao Xu, Jianying Li, Xiaohui LiCancer 2024 · DOI

TL;DR

This study investigates how varying CT tube voltage, radiation dose, and iterative reconstruction strength (ASiR-V) impact the detection and characterization of pulmonary nodules in ultra-low-dose CT (ULDCT). The findings suggest that while 70 kV offers better noise performance, 100 kV with higher ASiR-V levels better preserves nodule morphology, and doses above 0.33 mSv are crucial for sensitive nodule detection, particularly for ground-glass nodules.

What Problem Does This Paper Address?

Lung cancer screening using low-dose computed tomography (LDCT) has become increasingly common, but the cumulative radiation dose from repeated scans raises concerns about secondary cancer risks. Ultra-low-dose CT (ULDCT) offers a potential solution by significantly reducing radiation exposure, but this comes at the cost of image quality, potentially hindering nodule detection and characterization. This paper addresses the challenge of optimizing ULDCT protocols by exploring the interplay between key imaging parameters: tube voltage (kV), radiation dose, and the strength of adaptive statistical iterative reconstruction (ASiR-V). The goal is to find a balance that maximizes nodule detection and accurate morphological assessment while minimizing radiation dose, crucial for effective and safe lung cancer screening and follow-up.

Methods

The researchers utilized an anthropomorphic thorax phantom containing 12 simulated pulmonary nodules of varying sizes (5-12 mm) and CT values (-800 HU, -630 HU, 100 HU). These nodules were scanned using three ULDCT protocols with different dose levels (ranging from 0.10-0.11 mSv to 0.51-0.53 mSv) and two tube voltages (70 kV and 100 kV). Each scan was repeated five times to ensure statistical robustness.

Image reconstruction was performed using four different strength levels of ASiR-V (0% FBP, 30%, 50%, and 70%) with a slice thickness of 1.25 mm. A commercially available artificial intelligence (AI) software was employed to analyze the reconstructed images. This AI software was used to:

  • Calculate the nodule detection rate (DR).
  • Measure nodule diameters to compute the deformation coefficient (DC) and size measurement deviation percentage (SP), which quantify nodule morphology and size accuracy.

Statistical comparisons were made between different imaging groups to assess the impact of the tested parameters on nodule detection and characterization.

Datasets

The study utilized a single anthropomorphic thorax phantom (Lungman ph-1, Kyoto Kagaku Inc, Japan) containing 12 simulated spherical nodules. This phantom served as the sole dataset for evaluation. No new benchmarks were introduced; instead, the study focused on evaluating existing reconstruction techniques and parameters on a standardized phantom.

Key Insights and Ideas

  • Dose vs. Detection Trade-off: While ULDCT significantly reduces radiation dose, it's crucial to identify the minimum dose threshold that maintains acceptable nodule detection sensitivity. The study highlights that doses above 0.33 mSv are generally required for high sensitivity, especially for challenging ground-glass nodules.
  • Morphology Preservation: Higher ASiR-V strengths (≥50%) and 100 kV imaging appear to be more effective in preserving nodule morphology (lower DC and SP), which is critical for accurate diagnosis and follow-up. This suggests a trade-off between noise reduction and morphological accuracy at lower ASiR-V levels or lower kV.
  • Ground-Glass Nodules (GGNs) are Challenging: GGNs, particularly those with lower attenuation values (-800 HU), are harder to detect and characterize, especially at very low doses. This implies that specific protocols or higher doses might be necessary for reliable detection of these types of nodules.
  • AI Software's Role: The study demonstrates the utility of AI software in automating nodule detection and characterization, enabling large-scale analysis of imaging parameters. However, the performance of the AI is dependent on the quality of the input images.

Key Results

  • Detection Rate (DR): No significant differences in DR were observed between 70 kV and 100 kV at any dose level (p > 0.05). However, DR for GGNs increased significantly at dose levels of 0.32-0.34 mSv and higher.
  • Image Noise: Image noise decreased with increasing ASiR-V strength. 70 kV images generally showed slightly lower noise than 100 kV images under the same ASiR-V level.
  • Morphology (DC and SP): Higher ASiR-V strength levels (≥50%) and 100 kV generally resulted in better (lower) DC and SP, indicating improved nodule shape and size accuracy.
  • Nodule Size and Attenuation: Nodules smaller than 5 mm were frequently undetected, and nodules with very low CT attenuation values (-800 HU) were particularly difficult to detect, even at higher doses.

Limitations and Future Work

The authors acknowledge several limitations:

  • Small Sample Size: Only three types of CT attenuation values for nodules were analyzed.
  • Single AI Software: The study relied on a single commercial AI software, and its performance might vary with different algorithms or training data. The AI was not re-trained on the specific ULDCT data used in the study.
  • Phantom Study: The research was conducted on a phantom, and real-world patient data analysis is needed for validation.

Future work should involve:

  • Validation with multiple AI software packages.
  • Extension of the study to real patient image analysis.
  • Further investigation into optimal dose levels and reconstruction parameters for specific nodule types, particularly GGNs with low attenuation.
  • Exploring the impact of deep learning-based reconstruction algorithms, which have shown promise in ULDCT.

Summary updated

Abstract

Detection rates are similar between 70 kV and 100 kV scans. The 70 kV images have better noise performance under the same ASiR-V level, while images of 100 kV and higher ASiR-V levels are better in preserving the nodule morphology (lower DC and SP); the dose levels above 0.33mSv provide high sensitivity for nodules detection, especially the simulated ground glass nodules.
BibTeX
@inproceedings{yao2024effects,
  author = {Yue Yao and Xuan Su and Lei Deng and JingBin Zhang and Zengmiao Xu and Jianying Li and Xiaohui Li},
  title = {{Effects of tube voltage, radiation dose and adaptive statistical iterative reconstruction strength level on the detection and characterization of pulmonary nodules in ultra-low-dose chest CT}},
  booktitle = {Cancer},
  year = {2024},
  doi = {10.1186/s40644-024-00770-z},
}
Figure