Abstract: Objective: To evaluate the imaging performance of high-strength deep learning image reconstruction (DLIR-H) for ground-glass nodules (GGNs) on ultra-low-dose spectral computed tomography (CT)of the chest and to develop a alternative ultra-low-dose spectral CT protocol for lung GGN screening. Methods: A Gammex 472 thoracic phantom containing ground-glass-equivalent lesions was scanned using a 256-slice spectral CT scanner. Six radiation doses (30, 20, 15, 10, 7.5, 4.5 mGy) were used. The scanning parameters were fixed at 140 kVp with 74 eV monoenergetic imaging. Images were reconstructed using DLIR-H and adaptive statistical iterative reconstruction (ASIR-V). Quantitative metrics including CT number, image noise, signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), noise power spectrum (NPS), 50% threshold task transfer function (TTF50%), and lesion detectability index (d') were analyzed. Subjective visual assessment of images was performed simultaneously. Results: The Wilcoxon sign-rank test revealed statistically significant differences between the two reconstruction images at the same dose in terms of CT values, noise level, SNR, CNR, and TTF50%. DLIR-H significantly reduced image noise and improved the CNR and spatial resolution, whereas no statistically significant differences were observed between the two groups in NPS or d'. At a dose of 4.5mGy, DLIR-H exhibited only a mild decrease in spatial resolution (TTF50%) compared to the conventional 15 mGy dose using ASIR-V. The difference in the lesion detection index (d') was not statistically significant but showed a trend toward significance (P=0.076), whereas other parameters remained unchanged. Subjective visual evaluation indicated that the DLIR-H images demonstrated a uniform background, natural noise texture, and absence of significant reconstruction artifacts. Conclusion: DLIR-H can effectively alleviate image quality deterioration induced by ultra-low-dose CT scanning. This phantom study confirmed that for lung GGN screening, the 4.5mGy ultra-low-dose protocol combined with DLIR-H reconstruction achieved an overall image quality and lesion detectability equivalent to the standard 15 mGy ASIR-V protocol, with only a slight loss of spatial resolution.