This study aims to compare two AI methods in assessing decreased [18F]FDG metabolism for reflecting clinical neurological function in patients with ICVD, and to explore its value in research and clinical settings. The results demonstrated that the volume and percentage of decreased metabolism in the lobes on affected side, excluding the infarct area, attained through AI2 calculation consistently greater than those achieved through AI1. The correlation between the volume of decreased metabolism in the affected cerebral hemisphere, as calculated by AI1 and AI2, and the clinical score can be observed. The improved follow-up patients showed more pronounced metabolic improvement as assessed by AI1.

AI1 and AI2 methods are widely used in studies of ICVD, neurodegenerative diseases and epileptic disorders, mainly for the assessment of structural features within the brain, cerebral metabolism and asymmetry analysis of cerebral blood flow [22,23,24,25,26]. In this study of patients with ICVD, we found that the AI2 method consistently yielded larger estimates for the extent of hypometabolism on the affected side, in comparison to the AI1 method. It possible to interpret this in terms of the meaning expressed by the structure of the two formulas. Assuming that there is metabolic decrease around the infarction on the affected side, while the metabolism of the corresponding area on the unaffected side is normal, which indicates the absolute difference of metabolism between the two cerebral hemispheres. However, the denominator of the formula of the AI2 method is smaller because it is the average of the two sides, which leads to the amplification of the relative difference percentage, so that more voxels satisfy the condition of greater than 10% value. Therefore, the AI2 method may exhibit greater sensitivity to hypometabolism. Moreover, in regions with pronounced metabolic reduction on the affected side, such as the temporal and parietal lobes, the smaller denominator (mean value) in the AI2 formula will further magnify the relative difference in bilateral metabolism. Consequently, we hypothesize that the exaggerated effect of AI2 may be more pronounced in areas with more severe metabolic impairment.

Prior studies have established correlations between AI-assessed cerebral hypometabolism and clinical outcomes in ICVD. [27]. Cui et al. linked preoperative NIHSS scores to AI1 values using [18F]FDG PET [21], while Yu et al. demonstrated AI2's utility in tracking metabolic changes post-bypass surgery [28]. Sobesky et al. further associated the hypoperfusion volumes assessed by AI with NIHSS and mRS scores [29]. Our findings align with these observations, showing significant correlations between AI1/AI2-quantified hypometabolism and NIHSS/mRS scores, underscoring their clinical monitoring potential.

Notably, AI1 exhibited stronger correlations with neurological deficits in temporal/parietal regions compared to AI2. According to the cerebral arterial territories research, this distribution characteristic may be related to our subject selection, with cerebral ischemic attributed to the unilateral steno-occlusion attributed to the internal carotid artery or middle cerebral artery [30, 31]. The AI1 method employs the unaffected side as a benchmark, thereby quantifying the absolute metabolic loss on the affected side. This metric is directly correlated with clinical scores, including motor and language function. In contrast, AI2 may incorporate mild hypometabolic voxels into its calculations due to its smaller denominator. However, these voxels contribute mildly to clinical symptoms, leading to a diminished correlation. These suggest that the AI1 method may be more suitable for evaluating the metabolic changes around the infarct area and the clinical status of the unilateral internal carotid artery/middle cerebral artery steno-occlusive disease patients.

While focused on metabolism, our previous work validated AI-based asymmetry analysis for cerebral blood flow using PET/MRI [32], suggesting multimodal applications. Future studies should explore AI-based integration of hemodynamic and metabolic indices to enhance clinical stratification.

Sebök et al. [33] reported elevated AI value in ICVD with higher NIHSS/mRS scores, while Zhang H et al. [5] demonstrated that both the significant improvements in neurological function and improvements in brain metabolism shown by [18F]FDG PET imaging could be found after treatment. Similarly, our study found improved cerebral metabolism (assessed by AIs) and reduced NIHSS/mRS scores in fourteen follow-up patients, with AI1 showing greater sensitivity to metabolic changes than AI2. However, in five patients with improved NIHSS/mRS scores, hypometabolic areas did not consistently shrink, suggesting a need for closer monitoring and potential treatment adjustments, as neuronal activity correlates with glucose utilization [34, 35].

The quantification of the extent of decreased metabolism around infarct foci due to unilateral internal carotid artery or middle cerebral artery stenosis-occlusion based on AI the intuitive and objective assessment of the decreased metabolic state of the patient's brain tissue by neuroimagers and clinicians alike. For the most common type of ICVD, often caused by unilateral anterior circulation large vessel steno-occlusion, emphasizing metabolic reduction in affected brain tissue is crucial. On the one hand, given that hypometabolism assessed by the AI1 method correlates more strongly with clinical scores than the AI2 method in this study, we recommend using the AI1 method for evaluating cerebral hypometabolism in these patients in clinical practice. In addition, our findings demonstrate that the AI1 method can more accurately reflect metabolic changes before and after follow-up. The future research will focus on determining specific cutoff values based on the AI1 method to reflect improvements or deteriorations in neurological function, thereby enabling stratified management of ICVD patients in clinical applications. Specifically, This approach to imaging assessment can help guide timely clinical adjustments to further treatment regimens for those assessed to have a poor prognosis, guide close monitoring of the condition, and guide collaborative multidisciplinary clinical management to slow their disease progression, and improve functional outcomes. On another hand, we found that the AI2 method appears to be more frequently utilized in the analysis of diseases characterized by bilateral brain tissue lesions, such as ICVD with bilateral multiple cerebral artery steno-occlusion, Alzheimer’s disease and neuropsychiatric disorders through a comprehensive review of existing literature [24,25,26]. In accordance with varying research objectives and benchmark criteria, we recommend selecting a more suitable research methodology. In the present study, we found that cerebral metabolic evaluation based on AI2 approach can lead to an overestimation of the extent of hypometabolism in ICVD patients characterized by unilateral lesions. This overestimation may prompt clinicians to adopt aggressive therapeutic regimens, such as the implementation of cerebral revascularization surgery, frequent or high-intensity neurorehabilitation and prolonged pharmacological interventions. For patients, the unreasonable allocation of medical resources caused by it is likely to increase the psychological burden of patients and cause unnecessary anxiety. In addition, this extra demand for treatment may lead to an increase in the patient's medical costs.

Our study has a few limitations. First, the limited sample size and uniform participant population (unilateral steno-occlusive disease) may restrict broader applicability. Second, the single-center design, while ensuring consistency in imaging protocols and clinical assessments, may introduce selection bias due to regional healthcare variations. In addition, the retrospective nature of the study makes it susceptible to recall bias and information bias. And the rate of loss to follow-up may have impacted the completeness of our data. Future multicenter prospective studies are needed to validate these findings. Moreover, since the follow-up duration in this study was relatively short, the future study will improve the follow-up time (both number and duration) to capture a more comprehensive picture of disease progression and recovery. Additionally, the short follow-up duration should be extended in future studies for a more comprehensive view of disease progression. Lastly, this study employed a semi-quantitative approach. The future research might explore absolute quantitative aspects.