Objective To evaluate the ability of artificial intelligence-based non-electrocardiogram-gated global longitudinal strain (Non-ECG GLS) to identify the abnormalities of left ventricular ejection fraction (LVEF) and to explore its changing characteristics in populations with different cardiovascular disease burden. Methods A total of 181 patients who underwent transthoracic echocardiography in our hospital from March to November 2025 were prospectively enrolled. All subjects completed routine echocardiography and non-ECG GLS analysis. Using LVEF as the reference standard for left ventricular systolic function, Non-ECG GLS was compared between the presreved EF group (LVEF ≥ 54%) and the abnormal EF group (LVEF < 54%). Receiver operating characteristic (ROC) curve analysis was performed to evaluate the efficacy of Non-ECG GLS in identifying LVEF abnormalities. Based on an exploratory cardiovascular disease burden score, patients were divided into a low-burden (0~ 2 points) group, a moderate-burden (3~ 5 points) group and a high-burden (6~ 9 points) group. Differences in the Non-ECG GLS among the three groups were compared. A trend test was also conducted. Thirty subjects were randomly selected, and Non-ECG GLS was measured by two independent operators. Inter-operator reproducibility was assessed using Bland-Altman analysis. Results Among the 181 patients, 123 had normal LVEF and 58 had abnormal LVEF. The absolute value of Non-ECG GLS in the abnormal LVEF group was significantly lower than that in the normal LVEF group (P<0. 001). There were no statistically significant differences in baseline Non-ECG GLS between different gender groups and different age groups. ROC curve analysis showed that the area under the curve for Non-ECG GLS in identifying LVEF abnormalities was 0. 93 (95% CI: 0. 89~ 0. 96, P<0. 001). The optimal cutoff value was 16. 2% based on absolute value. The sensitivity was 89. 7%. The specificity was 84. 6%. There was a statistically significant difference in Non-ECG GLS among the different cardiovascular disease burden groups (F = 47. 97, P<0. 001). A decreasing trend in Non-ECG GLS was observed with increasing disease burden from the low-to the high-burden group (P<0. 001). Bland-Altman analysis revealed a mean difference of -0. 19% between the two operators. The 95% consensus threshold was -2. 99% to 2. 62%. Conclusions AI-based Non-ECG GLS demonstrates good discriminatory ability for identifying LVEF abnormalities. It also shows favorable inter-operator reproducibility. Its gradient variation across different cardiovascular disease burden populations suggests that Non-ECG GLS may serve as an auxiliary imaging parameter for assessing left ventricular systolic function.