Altered dynamic and static brain activity and functional connectivity in COVID-19 patients: a preliminary study

This study aimed to investigate the effects of COVID-19 on brain functional activity through resting-state functional MRI (rs-fMRI). fMRI scans were conducted on a cohort of 42 confirmed COVID-19-positive patients and 46 healthy controls (HCs) to assess brain functional activity. A combination of dynamic and static amplitude of low-frequency fluctuations (dALFF/sALFF) and dynamic and static functional connectivity (dFC/sFC) was used for evaluation. Abnormal brain regions identified were then used as feature inputs in the model to evaluate support vector machine (SVM) capability in recognizing COVID-19 patients. Moreover, the random forest (RF) model was employed to verify the stability of SVM diagnoses for COVID-19 patients. Compared to HCs, COVID-19 patients exhibited a decrease in sALFF in the right lingual gyrus and the left medial occipital gyrus and an increase in dALFF in the right straight gyrus. Moreover, there was a decline in sFC between both lingual gyri and the right superior occipital gyrus and a reduction in dFC with the precentral gyrus. The dynamic and static combined ALFF and FC could distinguish between COVID-19 patients and the HCs with an accuracy of 0.885, a specificity of 0.818, a sensitivity of 0.933 and an area under the curve of 0.909. The combination of dynamic and static ALFF and FC can provide information for detecting brain functional abnormalities in COVID-19 patients.


Introduction
COVID-19 patients may exhibit cognitive and neurological symptoms, such as headaches, loss of smell and consciousness disorders [1,2].Moreover, long-term issues such as fatigue, lack of attention and neuropsychiatric illnesses such as depression have been observed in some patients [3][4][5][6].COVID-19 survivors frequently experience lingering neurological symptoms, including cognitive dysfunctions, memory problems and neuropsychiatric illnesses.The neuropathophysiology includes persistently impaired hippocampal neurogenesis, decreased oligodendrocytes and myelin loss together with elevated cerebrospinal fluid cytokines/chemokines [7].These studies suggest that the SARS-CoV-2 virus has negative effects on the human brain.However, early brain functional connectivity (FC) alterations can precede structural alterations and may better explain physical and cognitive impairments in these patients, our specific understanding of these brain FC alterations is quite limited.Hence, it is crucial to further investigate the influence of the novel coronavirus on the human brain, aiding our enhanced comprehension of viral dissemination and impact mechanisms.This will subsequently furnish scientific foundations for the treatment of neurological sequelae in the future.
Resting-state functional MRI (rs-fMRI) is a noninvasive neuroimaging technique that measures bloodoxygen-level-dependent (BOLD) signal changes in the brain during a state of quiet rest.Actually, rs-fMRI is not a true resting state measurement, as the scanner noise and uncomfortable environment can be annoying for many people.Clinical patients may exhibit specific physiological responses during scanning.Slight scanner movements may lead to sensorimotor stimulation.In addition, several papers have shown that this fMRI resting state situation is psychologically demanding for several people and partly associated with specific emotional and vegetative responses [8][9][10].Taken together, this is not a silent and resting state situation, it is rather a specific kind of regular stimulation.It allows us to gain insights into the fundamental functional state of the brain, neural information processing and the organizational principles of brain networks [11,12].The static amplitude of low-frequency fluctuations (sALFF) and seed-based static FC (sFC) are two commonly used metrics for rs-fMRI data analysis and have been extensively applied [13].However, it is important to note that ALFF and FC, as static metrics, are based on the assumption that brain activity remains constant throughout the acquisition of BOLD signals.
They can only provide approximate information about regional brain activity and connectivity without capturing detailed information about event-related or dynamically regulated processes.In fact, regional neural activity exhibits significant temporal fluctuations [14].To address this, sliding window techniques can be incorporated, allowing the variability of sALFF and sFC over time to be calculated, resulting in dynamic ALFF (dALFF) and dynamic FC (dFC), respectively.Dynamic metrics, compared to traditional static metrics, offer advantages such as enhanced sensitivity, reproducibility, adaptability and global representation.They are capable of capturing brain functional changes on shorter time scales and providing a comprehensive overview of activity across various brain regions.Furthermore, dynamic metrics objectively reflect the overall functioning state of the human brain and its neural system [15].To date, there have been no dynamic studies investigating regional neural activity in COVID-19 patients.In this study, we plan to utilize ALFF and seed-based FC to explore the alterations in static and dynamic brain function in COVID-19 patients.
Currently, machine learning has extensively been employed for disease classification and prediction based on neuroimaging data.Li et al. [16] employed a support vector machine (SVM) model to classify and predict obstructive sleep apnea patients and healthy volunteers using dReHo data results.The accuracy achieved was 81.60%.Lin et al. [17] utilized an SVM model to classify major depressive disorder and healthy controls (HCs) using degree centrality data.The accuracy obtained was 87.71%.Building on this, we aimed to differentiate between COVID-19 patients and healthy volunteers by employing machine learning techniques on multiscale neuroimaging biomarkers.Additionally, we sought to investigate whether dynamic indicators exhibit higher sensitivity than static indicators in the classification of COVID-19.
This study aimed to investigate whether dALFF and dFC measurements could offer additional information compared to sALFF and sFC measurements in identifying abnormal brain function in COVID-19 patients.The sFC and dFC methods were employed to identify interconnected brain regions.The seed regions were defined based on the areas exhibiting abnormal sALFF in local brain activity between patients with COVID-19 and HCs.

Participants
This study was approved by the Ethics Committee of Yantai Affiliated Hospital of Binzhou Medical University.
Written consent forms were obtained from all participants.From December 2022 to January 2023, a total of 42 COVID-positive patients and 46 healthy volunteers were recruited for the study.To reduce the false positive rate among COVID-19 patients, all COVID-19 patients are subjected to confirmatory retesting to ensure the accuracy of the results.Patient inclusion criteria were as follows: (1) positive for COVID-19 and was diagnosed within 1-3 days; (2) All cases were first-time diagnoses and had not received any treatment before; (3) right-handedness; (4) absence of contraindications for MRI and (5) no brain structural lesions or traumatic brain injury.Exclusion criteria included: (1) subjects with more than 2 mm translation and 2° rotation; (2) history of neurological or psychiatric disorders; (3) patients experiencing respiratory distress requiring assisted ventilation or oxygen therapy and (4) individuals with claustrophobia.Healthy volunteers had no history of COVID-19 positivity.Participants were aged between 18 and 70 years.

MRI data acquisition
All MRI data were obtained with a 3.0 Tesla MR scanner (Skyra, Siemens, Germany).To minimize noise and head movement during the scanning process, headphones and foam pads were used.Participants were instructed to close their eyes, remain quiet and try to keep their minds clear without dwelling on any particular thoughts.All participants underwent routine T2 structural imaging to exclude any brain structural lesions that could affect brain function.The rs-fMRI data were collected using an echo-planar imaging sequence with the following parameters: repetition time (TR) = 2000 msec, echo time (TE) = 30 msec, flip angle = 90°, field of view = 240 mm × 240 mm, matrix = 64 × 64, number of slices = 32, slice thickness = 4 mm and total volume number of 200 slices.

Data preprocessing
Figure 1 illustrates the study's workflow.Data preprocessing was performed using Sales Process Management 12 (SPM12) (https://www.fil.ion.ucl.ac.uk/spm/software/ spm12/) and Resting-State fMRI Data Analysis Toolkit (RESTplus) (http://restfmri.net/forum/RESTplus)V1.2 based on the MATLAB 2013b (https://www.mathworks.com/; MathWorks, Natick, Massachusetts, USA) platform [18].For each subject, we implemented the following preprocessing steps: (1) to establish signal equilibrium and support participants' acclimation to the scanning environment, the first 10 volumes were excluded.(2) Slice time and head motion correction were applied to the remaining 190 volumes for further refinement, subjects were excluded if they exhibited a maximum displacement of more than 3.0 mm in any direction during the fMRI scan or if the rotation angle on any axis exceeded 3.0°.(3) The functional images were then spatially normalized to the Montreal Neurological Institute template, and images were resampled to 3 × 3 × 3 mm 3 voxel size.(4) To minimize noise, the images were subjected to a smoothing process using a 6 mm full-width at half-maximum Gaussian kernel.( 5) Removal of linear drift generated by MRI or other factors.(6) The regression removed covariates such as the Friston 24 parameter model, white matter signal and cerebrospinal fluid signal [19].Three COVID-19 patients were excluded due to head movement of more than 3.0 mm.Eventually, a total of 39 COVID-19 patients and 42 healthy volunteers were included for further analysis.

Static amplitude of low-frequency fluctuations and dynamic amplitude of low-frequency fluctuations variance calculation
sALFF was calculated using RESTplus.The calculation steps are as follows (1) convert the time series of each voxel into the frequency domain using a method called fast Fourier transform.( 2) Analyze the power spectrum of each voxel and take the square root of the spectrum.(3) Average the square roots within the frequency range of 0.01-0.08Hz to obtain the sALFF value.(4) To normalize the results, divide the sALFF value of each voxel by the average sALFF value of all voxels within the defined brain region.( 5) sALFF was Z-score transformed [20].
For the calculation of dALFF, we used the dynamicBC toolkit [21].We utilized the integration of the sliding window technique and sALFF to gauge the stability of brain activity, shedding light on its inherent constancy.Choosing the appropriate window length is crucial.A longer window length can lead to the loss of dynamic changes in brain activity, while a shorter window length may introduce significant noise into the estimation of low-frequency components.Previous studies suggest that the minimum window length should be greater than 1/ f min , where f min represents the minimum frequency of the time series [22].Therefore, we chose a sliding window length of 50 TRs (100s) to capture the temporal signal of fMRI, and first selected 1 TR(2s) as the sliding step size [23].The time series for each subject was divided into 141 windows, which were subsequently concatenated to create the dALFF map.The variance of dALFF was calculated to assess the temporal variability of brain activity.Finally, the variance of dALFF was normalized using Z-scores to improve the normality of the data.

Static functional connectivity and dynamic functional connectivity calculation
The preprocessed functional images underwent bandpass filtering with a frequency range of 0.01-0.08Hz.Seed-based voxel-based correlation analysis is a method to explore FC networks by selecting a specific seed region and calculating the correlation between that region and other brain regions [24].We defined the brain regions with statistically significant differences in sALFF between the COVID group and HC group as seed points (bilateral lingual gyrus).sFC is derived from calculating the Pearson correlation between the average BOLD time series of seed points and each voxel in the other areas of the brain.Subsequently, we combined sFC with the sliding window technique to generate dFC maps and calculate the variability of dFC.The window length used is consistent with the analysis of dALFF.Finally, the Pearson correlation coefficients of sFC and the variability of dFC were transformed using Fisher's z-transform to improve data normality.

Statistical analyses
All demographic and clinical data analyses were performed using the IBM Statistical Package for the Social Sciences (SPSS) V 22.0 statistical software (Armonk, New York, USA).As the age variable did not follow a normal distribution, the Mann-Whitney U test was selected.The chi-square test was used for the statistical analysis of sex data.When the P value is less than 0.05, it indicates statistical significance.We used the statistical module of Data Processing & Analysis of Brain Imaging ( http://www.rfmri.org/dpabi)for statistical analysis [25].Group differences in imaging parameters (sALFF, dALFF, sFC and dFC) between the COVID-19 group under study and the HCs group were analyzed using a two-sample t-test.Age, sex and framewise displacement were included as covariates in the analysis.Two-sample t-tests between the two groups were all corrected by the Gaussian random field (GRF), and the thresholds for multiple comparisons were voxellevel P < 0.001 and cluster-level P < 0.05.

Classification analysis
The radial basis function kernel SVM model from the 'scikit-learn' package in Python was employed to identify whether the altered brain regions in the four neuroimaging indicators could differentiate COVID-19 patients from healthy volunteers, aiming to aid in clinical diagnosis.All subjects were randomly divided into a training set and a test set in a 7 : 3 ratio.The model was trained using the brain regions with changes in these four neuroimaging indicators as feature inputs.Due to the small sample size, we employed five-fold cross-validation to enhance the classifier's generalization ability and reliability [26].The performance of the SVM model was evaluated using metrics including accuracy, specificity, sensitivity and area under the curve (AUC).In addition, we sequentially inputted the same features into a random forest (RF) model to validate the reliability of the previous neuroimaging feature training results.To enhance our understanding and interpretation of the prediction outcomes, we employed the RF model to calculate the contribution of each differential brain region in the classification prediction.

Validation analyses
We utilized different window lengths, specifically 30 TRs (60s) and 70 TRs (140s), to eliminate the influence of parameter selection and showcase the stability of the dALFF and dFC results.

Demographic information and clinical characteristics
Demographic information and clinical characteristics of the 42 COVID-19 patients (27 females and 15 males, mean age 44.9 ± 10.5) and 46 HCs (28 females and 18 males, mean age 45.6 ± 10.6) were collected.There were no significant differences between the two groups in terms of age and sex.

Static amplitude of low-frequency fluctuations and dynamic amplitude of low-frequency fluctuations variance findings
Compared with HCs, the COVID-19 group had decreased sALFF in the right lingual gyrus and the left middle occipital gyrus (voxel-level P < 0.001, cluster-level P < 0.05, GRF corrected; Fig. 2 and Table 1).dALFF increased in the right rectus gyrus (voxel-level P < 0.001, cluster-level P < 0.05, GRF corrected; Fig. 2 and Table 1).

Static functional connectivity and dynamic functional connectivity findings
Compared with HCs, we observed reduced sFC intensity in the bilateral lingual gyrus and the right superior occipital gyrus among COVID-19 patients (voxel-level P < 0.001, cluster-level P < 0.05, GRF corrected; Fig. 3 and Table 1).Additionally, in the COVID-19 group, decreased dFC intensity was found in the bilateral lingual gyrus and the right precentral gyrus (voxel-level P < 0.001, cluster-level P < 0.05, GRF corrected; Fig. 3 and Table 1).

Classification analysis results
We selected the cluster values of the differences in sALFF and sFC as static features (three feature variables), and the cluster values of dALFF and dFC as dynamic features (two feature variables).Additionally, we combined the cluster values of sALFF, sFC, dALFF and dFC as combined static-dynamic features (five feature variables) to input into the classification model.With the combination of five brain regions (the lingual gyrus, middle occipital gyrus, gyrus rectus, superior occipital gyrus and precentral gyrus) that are altered in both dynamic and static measures, we were able to distinguish between COVID-19 patients and HCs with an accuracy of 0.885, a sensitivity of 0.818, a specificity of 0.933 and AUC of 0.909.The accuracy, specificity, sensitivity and AUC of each feature are summarized in Table 2 and depicted in Fig. 4. According to Fig. 4 and Table 2, we can observe that the classification results obtained by RF are similar to the SVM classification results.There are significant differences in the static amplitude of low-frequency fluctuations (sALFF) and dynamic amplitude of low-frequency fluctuations (dALFF) between COVID-19 patients and healthy controls.(a) Brain region indicates significant differences in sALFF between COVID-19 patients and healthy controls.(b) Brain region indicates significant differences in dALFF between COVID-19 patients and healthy controls.Using GRF correction, the threshold for multiple comparisons was set at a voxel-level of P < 0.001 and cluster level of P < 0.05.L, left; R, right.

Correlational analysis results
There was no correlation observed between the decline in sALFF, sFC, dFC and the increase in dALFF with clinical variables (i.e.maximum body temperature and duration of fever) in COVID-19 patients.

Validation results
The validation analysis revealed that the results obtained for 30 TRs and 70 TRs were largely consistent with the main findings derived from 50 TRs.This indicates that the results of our selected 50 TRs are reliable.There are significant differences in static functional connectivity (sFC) and dynamic functional connectivity (dFC) between COVID-19 patients and healthy controls.(a) Brain region indicates significant differences in sFC between COVID-19 patients and healthy controls.(b) Brain region indicates significant differences in dFC between COVID-19 patients and healthy controls.Using GRF correction, the threshold for multiple comparisons was set at a voxel level of P < 0.001 and a cluster level of P < 0.05.L, left; R,right.sALFF can reflect cerebral blood flow or the intensity of brain functional activity in that brain region, and a decrease in sALFF is generally associated with impaired brain activity [27].We observed a decrease in sALFF values in the fusiform gyrus and occipital middle gyrus of COVID-19 patients compared to healthy volunteers, suggesting potential disruption of local information processing within the visual network of COVID-19 patients.The fusiform gyrus is primarily involved in color perception and basic shape analysis, while the occipital middle gyrus contributes to motion, direction and size perception, as well as the construction of visual scenes and memory [28,29].Although the fusiform gyrus and occipital middle gyrus are located in different regions of the cerebral cortex, they both belong to the visual network that participates in processing visual information [30,31].

Discussion
In contrast to our findings, a longitudinal study by Li et al. [32] reported an increase in ALFF values in the occipital middle gyrus and inferior occipital gyrus of COVID-19 patients after 6 months of recovery, possibly due to compensatory effects as COVID-19 patients entered the recovery phase, leading to the observed ALFF increase during follow-up [32].Another study indicated a decrease in ReHo in the right superior temporal gyrus and left inferior temporal gyrus of COVID-19 patients, a phenomenon that draws attention due to the significant role of the temporal lobe in visual tasks such as object recognition, color identification and scene analysis [33,34].The aberrant performance of the fusiform gyrus and occipital middle gyrus within the visual network of COVID-19 patients supports our research findings.Therefore, in conjunction with our study, the visual network may play a significant role in the pathogenesis of COVID-19.
dALFF can analyze the temporal patterns of spontaneous brain neural activity, revealing spontaneous neural activity patterns within different periods of brain regions, which assists in understanding the spatiotemporal characteristics of the brain.In comparison to healthy volunteers, the COVID group exhibited an increase in the dALFF of the straight gyrus in the right inferior frontal gyrus, indicating abnormal temporal fluctuations in the brain activity of this region.Longitudinal studies have found an increase in sALFF in the frontal lobe, temporal lobe and parietal lobe of follow-up COVID patients 1 year after recovery compared to healthy volunteers [35].Furthermore, a reduction in gray matter thickness of the orbitofrontal cortex and parahippocampal gyrus, as well as a decrease in brain volume, has been observed in COVID-19 patients [36].The straight gyrus is located in the frontal lobe and is a significant brain region within the default mode network, which is associated with processes such as introspection, self-reflection, memory and emotion processing [37,38].Existing studies have indicated that many COVID-19 patients experience headaches or musculoskeletal pain, and some patients may also exhibit psychological symptoms such as anxiety or depression [4,39].This suggests that the increased variability in the straight gyrus may be related to physical pain, anxiety or depression in COVID-19 patients.
In this study, based on the differences in sALFF, we selected the bilateral lingual gyrus as seed sALFFnts for sFC and dFC analyses.sFC reflects the average FC strength between different brain regions over a certain period, revealing the brain's network structure, functional modules and baseline state, which aids in understanding the brain's normal functionality [40].In this study, we observed a decrease in the sFC connectivity strength between the bilateral lingual gyrus and the superior occipital gyrus in COVID-19 patients compared to HCs.Both the lingual gyrus and the superior occipital gyrus are essential components of the visual system, playing a crucial role in our perception and understanding of the surrounding visual world.This decrease in sFC strength could be attributed to issues in the lungs and respiratory system caused by infection with the novel coronavirus, resulting in patient hypoxia and subsequently reduced cerebral blood flow, thus affecting FC efficiency within the visual network.Reports have indicated that a COVID-19 patient experienced bilateral strength loss after suffering from an ischemic stroke in the bilateral occipital lobe, and another patient with COVID-19 developed quadrantanopia due to posterior cerebral artery infarction [41,42].Therefore, based on the aforementioned research findings, we hypothesize that the novel coronavirus may be prone to causing alterations of the visual system.However, the underlying mechanisms behind these results require further in-depth investigation.
dFC reflects the temporal changes in brain connectivity, revealing the temporal dynamics of brain connections, changes during cognitive tasks, neural regulatory mechanisms and alterations in disease and abnormal states.This assists in a deeper understanding of brain activity in different contexts [14].The precentral gyrus plays a critical role in the sensorimotor network, participating in various functions including motor control, planning, perception and execution control.This network encompasses multiple brain regions and supports complex sensory and motor tasks [43].We observed a decrease in the strength of dFC between the bilateral lingual gyrus and the precentral gyrus in COVID-19 patients, indicating disrupted information transfer between the visual network and the sensorimotor network.The visual network is interconnected with other cognitive neural networks such as the sensorimotor network and the cognitive control network, which collaborate to achieve complex cognitive and behavioral functions [44].Studies have found that patients with mild COVID-19 who experience fatigue and cognitive difficulties exhibit abnormalities in motor cortex neurophysiology, and neuropsychological examinations reveal deficits in executive attention even under mild influence [45].Another study discovered a negative correlation between the dFC of the visual network and the sensorimotor network in COVID-19 survivors from Wuhan, which aligns with our research findings [14]. We

Limitations
This experiment has some limitations.First, COVID-19 patients might be more scared, more sensitive and more worried due to their infection causing different mind-wandering states, which in turn might influence the hemodynamic fluctuations.Second, the study was carried out within 1-3 days after the onset of COVID-19, and specific skills and cognitive functions of the patients were not included.The missing information about cognitive functions and skills was a big problem.Third, in the dynamic analysis, a consensus on the optimal parameters for the sliding window was not reached.We conducted validation analyses using different window lengths, and the results indicated the reliability of our research findings without being affected by the sliding window parameters.Fourth, the number of participants in this experiment was relatively small; increasing the sample size in subsequent studies would enhance the persuasiveness of the results.Fifth, the machine learning classification performance for COVID-19 patients and healthy volunteers was not validated using external datasets, potentially leading to overfitting.Last, the experimental design was relatively simple, making it difficult to definitively attribute observed changes solely to novel coronavirus infection.Future studies should consider adding additional control and follow-up groups to gain a deeper understanding of the mechanisms and effects of novel coronavirus on brain function.

Conclusion
The study represents a pioneering effort in utilizing a combination of dynamic and static ALFF and FC methods to unravel the changes in brain function in COVID-19 patients.It offers a promising pathway toward developing effective diagnostic tools through the integration of machine learning techniques with neuroimaging markers.The findings of this study may provide the necessary evidence for a deeper understanding of the potential neuropathology of COVID-19 and offer possible neuroimaging markers for clinical diagnosis.

Fig. 3
Fig. 3 This study is the first to explore the functional changes in the brains of COVID-19 patients by combining static and dynamic rs-fMRI metrics.In this study, COVID-19 patients exhibited significant changes in brain functional activity, characterized by a decrease in sALFF values in the right fusiform gyrus and left occipital middle gyrus, as well as an increase in dALFF variance in the right precuneus.The sFC values of the bilateral fusiform gyrus and the right superior occipital gyrus decreased in COVID-19 patients, along with a decrease in dFC strength in the anterior cingulate cortex.Using the combination of abnormal brain regions based on sALFF, dALFF, sFC and dFC as features for classification prediction yielded an accuracy of 88.5%, a sensitivity of 81.8%, specificity of 93.3% and an AUC of 0.909, demonstrating the potential clinical application value of using machine learning for classifying COVID-19 patients.In conclusion, these findings provide evidence for the impact of the SARS-CoV-2 virus on the human brain and may aid in better understanding the pathological mechanisms of COVID-19.

Table 2
Performance of various resting-state functional MRI features in support vector machine and random forest models When writing this article, most studies were focused on follow-up research of COVID-19 patients, and there has not been any rs-fMRI monitoring study of acute brain function changes in COVID-19 patients after infection.This study used the rs-fMRI technique to capture functional connections and activity patterns between different regions of the brain in the acute phase of COVID-19 infection, filling the gap in rs-fMRI brain activity research during this period.In this rs-fMRI study, we found that COVID-19 patients in the acute phase of infection mainly exhibited neuroimaging alterations in the occipital gyrus, lingual gyrus and rectal gyrus.Both dALFF and dFC showed differences, indicating the presence of dynamic changes in the brains of COVID-19 patients.In terms of classification performance, dynamic indicators outperformed static indicators, suggesting that dynamic indicators can better reflect changes in brain function.Compared to static indicators, dynamic indicators are more time-sensitive and can more accurately reflect changes and adjustments in brain function.Therefore, in brain function research and classification, dynamic indicators may be more accurate and reliable.