Abstract: Objective: To explore the diagnostic efficacy of artificial intelligence (AI)-assisted Alberta Stroke Program Early CT Score (ASPECTS) based on dual-energy CT (DECT) in the evaluation of acute ischemic stroke (AIS) and to optimize reconstruction parameters. Methods: Fifty-one patients with suspected AIS (onset time ≤12 h) and positive diffusion-weighted imaging (DWI) findings were prospectively enrolled. A dual-energy scanning mode of 80/Sn150 kVp was used, and two sets of monoenergetic images (Qr 70 keV and Qr 75 keV) were reconstructed using Mono+ technology. The conventional linear fusion image generated by dual-energy scanning was defined as a non-contrast CT (NCCT), and two additional sets of images (Qr NCCT and Hr NCCT) were reconstructed using the Qr and Hr convolution kernels. Two AI models (Models 1 and 2) were applied to evaluate the four image sets (Qr 70 keV, Qr 75 keV, Qr NCCT, and Hr NCCT) using ASPECTS. Using DWI-ASPECTS as the gold standard, the diagnostic efficacy and consistency of different reconstruction methods and AI models were analyzed. Results: Model comparison showed that the accuracy of Model 1 was higher than that of Model 2 for all reconstruction methods and lesion locations. Specifically, the Model 1 + Qr 70 keV group achieved a higher overall accuracy (81.2% vs. 77.5%), sensitivity (72.3% vs. 58.6%), and kappa consistency with the gold standard (all P < 0.05). Model 2 outperformed Model 1 in terms of specificity and positive predictive value (PPV) (P < 0.05). Among the reconstruction methods, Qr 70 keV and Hr NCCT exhibited optimal diagnostic efficacy. Hr NCCT improved the model sensitivity compared to Qr NCCT, while Qr 70 keV was more sensitive for identifying small ischemic lesions. Conclusion: For AI-assisted ASPECTS scoring based on DECT in AIS evaluation, the use of images reconstructed using Qr 70 keV or Hr NCCT combined with AI Model 1 is recommended.