Cerebral Perfusion Imaging Predicts Levodopa-Induced Dyskinesia in Parkinsonian Rat Model

Cerebral Perfusion Imaging Predicts Levodopa-Induced Dyskinesia in Parkinsonian Rat Model | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Cerebral Perfusion Imaging Predicts Levodopa-Induced Dyskinesia in Parkinsonian Rat Model Jarrad Perron, Sophia Krak, Samuel Booth, Dali Zhang, Ji Hyun Ko This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6658744/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 30 Sep, 2025 Read the published version in npj Parkinson's Disease → Version 1 posted 9 You are reading this latest preprint version Abstract Approximately half of Parkinson’s disease (PD) patients manifest motor complications related to treatment called levodopa-induced dyskinesia (LID). Preventing onset of LID is crucial to long-term management of PD, but the reasons why some patients develop LID are unclear, for instance, it is unknown if vascular or neuroinflammatory abnormalities exist prior to levodopa therapy or are a response to chronic exposure. The ability to prognosticate predisposition to LID would be valuable for the management of LID and for the investigation of strategies for its mitigation. Thirty rats received 6-hydroxydopamine to induce parkinsonism-like behaviors before treatment with levodopa (2 mg/kg) daily for 22 days. Fourteen developed LID-like behaviors. Fluorodeoxyglucose PET, T 2 -weighted MRI and cerebral perfusion imaging were collected before treatment. Support vector machines were trained to classify LID vs. non-LID animals. Volumetric perfusion imaging performed best overall with 86.16% area-under-curve, 86.67% accuracy, 92.86% sensitivity, 81.25% specificity for classifying animals with LID vs. non-LID from treatment-naïve baseline imaging in leave-one-out cross-validation. We have demonstrated proof-of-concept for imaging-based classification of a parkinsonian rat model. The ability to non-invasively identify a predisposition to LID would allow for more targeted investigations into the risk factors for LID and its prevention in the earliest stages. Health sciences/Diseases/Neurological disorders/Parkinsons disease Health sciences/Biomarkers/Predictive markers levodopa induced dyskinesia rodent perfusion neuroimaging machine learning Figures Figure 1 Figure 2 Figure 3 Introduction Parkinson's disease (PD) is the second most common neurodegenerative disorder, characterized by dopaminergic cell degeneration. Levodopa (dopamine precursor) is the gold standard for PD treatment (> 90% of PD patients are treated with levodopa) 1 . Levodopa-induced dyskinesia (LID) is a syndrome associated with involuntary movements such as dystonia, chorea and athetosis. The key mechanism of LID is associated with excessive synaptic dopamine release, yet several neurotransmitters and non-neuronal factors are also involved 2 – 4 . Nearly half of all PD patients treated with levodopa develop LID within 5 years, while > 90% of all patients eventually develop LID after 15 years 5 , 6 . It was previously recommended that levodopa treatment must be withheld as long as possible to delay the LID development, however, this was later been revised because it was suggested that the clinical benefit of levodopa is greater than the potential harm 7 , 8 . It is now settled that levodopa should not be withheld as an initial treatment option because epidemiological studies with drug-naïve patients confirmed that the disease progression is more important factor than the levodopa treatment duration for LID development 9 . LID is now regarded as one of the inevitable consequences in standard pharmacological PD management in many patients. There is currently no disease modifying therapy for PD and most patients eventually develop LID 1 . Concerted effort has been vested to find novel LID management strategies or alternative treatments. One major innovation is the usage of intestinal gel formulations that enable the slow and sustained release of levodopa over time (e.g. Duodopa©) 10 . This helps reduce so-called off-time and improves quality of life for people with PD, however, it does not solve the fundamental problem, since this approach focuses on refining dosage rather than addressing the underlying issues inherent to treatment with levodopa. In effect, any refinement of levodopa administration simply delays onset rather than treating underlying factors. Deep brain stimulation (DBS) is regarded as the most effective treatment options for LID, which allows reduction of levodopa dose, however, less than 10% of PD patients are eligible for DBS due to many contraindications (e.g. old age, cognitive impairment) 11 . It is also a necessarily invasive procedure and requires significantly greater human resources to pursue. Even with marked improvements in drug delivery and usage of brain stimulation technology, LID is still an unavoidable reality for many PD patients receiving standard pharmacological therapy. This is exacerbated by the fact that neurodegenerative disease is expected to proliferate dramatically by the mid-21st century 12 . Another alternative approach would be preventive treatment. Historically, less attention has been placed on those PD patients who do not develop LID (or very slow LID emergence; non-LID or NLID). Understanding what makes these patients resistant to LID may uncover more attainable, preventive therapeutic targets. Protective factors have previously been identified, but they are often limited to demographic variables (e.g., male sex, older age of disease onset and low levodopa dose) due to the lack of longitudinal data with deep phenotyping and lack of non-invasive investigation due to ethical reasons 13 . Animal models have greatly improved our understanding. For example, we have developed a novel low-dose levodopa treatment model that mimics the variability of LID emergence in humans which demonstrated that vasomotor response to dopamine is significantly altered in parkinsonian rodents that developed LID while the ones of non-LID animals was not 14 , 15 . This previous work in our group generated a neuroimaging dataset of 30 parkinsonian rodent models which were imaged with multiple modalities and time points, including a treatment-naive baseline 14 , 15 . In the present work, we detail the development and validation of a machine learning-based model to identify prospective LID in parkinsonian rats using pre-treatment baseline neuroimaging data. We investigate the utility of three separate neuroimaging modalities and their multimodal combinations for this purpose: structural magnetic resonance (MR) imaging, cerebral perfusion imaging (flow rate and volumetric data) and fluorodeoxyglucose emission tomography (FDG PET). The proposed animal model will provide a novel window of opportunity for researchers to investigate the brains of LID animals in the earliest possible stage as an avenue to discovery of protective factors against LID prior to chronic levodopa treatment. By identifying those factors that influence resilience or predisposition towards LID, we may identify a novel therapeutic target for the management or prevention of LID, which would dramatically improve quality of life for PD patients. Material and Methods Ethical Approval All experiments received approval from the University of Manitoba's Animal Care Committee and were conducted in compliance with the guidelines of the Canadian Council on Animal Care. Full details, including detailed ethical guidelines, can be found in Booth et al (under review). Data and Code Sharing Data, models and MATLAB scripts may be shared by reasonable request to the corresponding author. Levodopa Dosage, Behavioral Testing, Imaging Schedule and Acquisition Parameters Previous research in our group led to the development of a novel parkinsonian rodent model of LID that was used in the present work 14 . A validated rodent model of LID was developed using unilateral 6-hydroxydopamine (6-OHDA) lesions to induce dopaminergic denervation. This model was used to examine neurovascular and astrocytic calcium responses to dopamine in the dyskinetic state. Female Sprague-Dawley rats were housed under standard conditions (12-hour light/dark cycle, ad libitum food and water). 30 rats underwent unilateral 6-OHDA lesioning to the right medial forebrain bundle (12.5 µg in 2.5 µL of 0.2% ascorbic acid/saline, infused over 5 minutes). To minimize extrasynaptic degradation, pargyline hydrochloride (10 mg/kg, intraperitoneal) was administered 30 minutes prior to surgery. After a two-week recovery period, animals were screened for hemiparkinsonian-like behaviors using the cylinder test. Following lesion validation, animals received daily levodopa (2 mg/kg subcutaneous) co-administered with benserazide (15 mg/kg) for 22 days. A control group (n = 5) received isotonic saline. Dyskinesia severity was assessed on days 1, 11 and 22 using the abnormal involuntary movements (AIM) test, scoring axial, orolingual and limb movements on a 0–4 scale. Animals with AIM scores ≥2 in at least one category by day 22 were classified as LID-positive 16 . To confirm lesion integrity immunohistochemical staining was performed on striatal and substantia nigra (SN) sections. First, tyrosine hydroxylase immunostaining quantified dopaminergic cell loss in the SN. Animals with <90% dopaminergic depletion were excluded. Our previous study demonstrated that approximately 50% of the treated animals develop LID-like behaviors 14 . In the present study, 14 of 30 animals developed LID. Animals were scanned with PET-MR both OFF and ON levodopa at two timepoints: t 1 , when levodopa treatment first began and LID symptoms had not yet appeared (OFF 1 and ON 1 studies were performed on consecutive days; at OFF 1 , animals had never received levodopa treatment), and t 2 , when levodopa medication had been administered for 21 days (OFF 2 and ON 2 studies were also performed on consecutive days). On the day of OFF 2 , animals were administered levodopa immediately after scanning sessions. Dyskinesia was evaluated using three abnormal involuntary movement (AIMS) tests on days 1, 11 and 23 of levodopa usage. AIMS refers to uncoordinated, uncontrollable movements that occur without conscious effort and are reflected side effects of PD. The details of AIMS test scoring are described elsewhere 16 . The animals were sacrificed by perfusion fixation the day following the last AIMs test and their brain tissue was removed. Administration protocol differed by imaging sessions. For ON sessions, animals were given 2 mg/kg levodopa with 15 mg/kg benserazide. An equivalent volume of isotonic saline was given prior to scanning for OFF sessions. After twenty minutes, the animal was placed in an induction chamber with 5% isoflurane and 0.5 L/min oxygen flow to induce anesthesia. The animals were shifted to a nasal cone with an oxygen flow rate of 0.5 liters per minute and an isoflurane concentration of 1.5– 2%. To administer tracer during the scan, a 23-gauge needle was used to cannulate the tail vein before loading onto the scanning bed. Rats were placed face down in the anesthetic system's nasal cone. Throughout the imaging procedure, body temperature and respiration were recorded and kept constant at 37 ± 0.5 °C and 60-80 breaths per minute, respectively. Studies were acquired with an MR Solutions 7.0 Tesla drymag scanner with the clip-on PET module. The PET and MRI imaging were performed sequentially, and a rat head RF coil was used for all MR imaging. Structural MR, dynamic susceptibility contrast (DSC) MR (gadolinium contrast enhancement) and FDG PET studies were collected. Anatomical images were acquired using a fast spin echo T 2 -weighted pulse sequence (26 x 1 mm coronal slices, TR = 4000 ms, TE = 45 ms, matrix size = 250 x 256 and 3 averages) over an acquisition time of 12 minutes. DSC perfusion maps were acquired using a gradient-echo planar imaging sequence (TR = 333 ms, TE = 20 ms, field of view = 30 x 30 mm, matrix size = 64 x 64, slice gap = 0.2 mm, slice thickness = 1.5 mm, temporal resolution = 0.333 s) such that the entire rostral striatum and majority of caudal striatum was included in the acquisition. 180 acquisitions were taken over 60 seconds and 0.3 mmol/kg of Gadovist (gadobutrol-based contrast agent) in isotonic saline was injected through tail vein cannulation. From these, relative cerebral blood flow (CBF) images were generated using custom scripts in the MATLAB 2023a environment (MathWorks Inc., Natick, MA, USA) by voxel-wise calculation of the transverse relaxation rate curve maximum, where S(t) is signal intensity at time t, S 0 is the pre-contrast baseline intensity and TE is the echo time of the pulse sequence. Cerebral perfusion volume (CBV) images were generated by integrating over the perfusion flow volume. PET images were acquired immediately after DSC acquisition and the bed position of the subject moved to the PET coil for a 15-minute acquisition. PET images were reconstructed using filtered back projection. A summary of these methods is shown in Figure 1. Full experimental details may be found in Booth et al (under review). Neuroimaging Data Preprocessing Preprocessing operations were performed using the Statistical Parametric Mapping (SPM12) software package (http://www.fil.ion.ucl.ac.uk/spm/ Wellcome Department of Cognitive Neurology, London, UK) within the MATLAB R2023a environment. All images were resized by ×10, then manually re-oriented to align with the Schwartz rat T 2 template, which was also resized by ×10. The Schwartz template is based on 97 anatomical MR studies from Sprague Dawley rats, volumetrically reconstructed into Paxinos and Watson space 17, 18 . All functional images (PET, CBF, and CBV) were coregistered to T 2 -MRI using SPM12’s default coregistration function and were visually inspected. T 2 -weighted structural images were spatially normalized into template space using the "Old Normalize" function in SPM12, with no affine regularization, a nonlinear frequency cutoff of 25 and 16 iterations. The same deformations were applied to write the coregistered functional images into template space. Resulting images were generated with the template image's bounding box ([-95, -168, -138; 95, 70, 52]) and voxel size (2x2x2 mm). All functional images (FDG PET, CBF, and CBV) images were smoothed with a Gaussian kernel with full width at half maximum of 8x8x8 mm. Each image was proportionally scaled to the mean signal intensity of the entire volume. Comparison of Neuroimaging Modalities with a Support Vector Machine A support vector machine (SVM) is a very robust machine intelligence model that is commonly used for the purpose of binary classification. It functions on the principle of maximizing the margin between classes in an extremely high-dimensional feature space 19 . A series of SVMs (a priori defined hyperparameters: linear kernel, 5% outlier fraction and iterative single-data algorithm as optimizer) was trained to classify LID vs. NLID subjects, and modality-specific performance between models (T 2 -weighted MRI, CBV, CBF and FDG PET) were compared by area-under-curve (AUC), sensitivity (SEN), specificity (SPE), accuracy (ACC) and both positive and negative predictive values (PPV/NPV). Multimodal combinations of neuroimaging data were also investigated as a possible method of increasing performance through complementary information. Visualizing Model Decisions The hyperplane of the best performing model was visualized and underwent permutation testing. Study labels (LID vs. NLID) were permuted randomly and an SVM was trained on this permuted data. The original hyperplane weights were ranked in the permuted distribution of hyperplane weights at each voxel and p-value was determined. Weights were deemed significant if they were ranked within the top 500 of 10,000 iterations (p < 0.05). Only the significant voxel’s hyperplane region weight was visualized and carried forward for further anatomical analysis under the assumption that these voxels are meaningful for LID vs NLID classification. The original hyperplane was z-scored relative to the intracranial voxel weights and then masked by the statistically significant voxels of the permuted hyperplane. Clusters of voxels were then characterized using MarsBaR and a pre-determined size threshold of 50 voxels 20 . The anatomy was identified by volume-of-interest analysis with the Schwarz atlas found in the Small Animal Molecular Imaging Toolbox (SAMIT) 21 . Validation Due to the modest sample size of our rodent population, we employed leave-one-out cross-validation (LOOCV) to estimate the generalizability of our models. This also allowed us to infer the effect of overfitting upon base models trained on all available data. Results The purpose of the present study was limited to developing a machine learning model that classified LID vs. NLID animals using baseline neuroimaging data. More details on behavioral test results and overall outcomes will be published elsewhere (Booth et al., under review). Modality-based Performance Comparison Classification of LID vs. NLID status by SVM was evaluated on 4 separate neuroimaging modalities (anatomical T 2 , CBF, CBV and FDG PET) for our rodent model (n LID = 14 and n NLID = 16) and their multimodal combinations. The performance of each was estimated by examining ACC, AUC, SEN, SPE and predictive values in LOOCV over the entire subject set. All results are compiled in Table 1. SVM performance varied considerably depending on the neuroimaging modality and modality combinations used for classification of LID versus NLID subjects. Among individual modalities, models trained on perfusion data demonstrated very high performance. The SVM trained on CBV imaging achieved 86.67% ACC and 86.16% AUC. It also demonstrated strong performance with balanced SEN and SPE at 92.86% and 81.25% respectively. Similarly, the CBF model performed comparably well, with 88.84% AUC, 83.33% ACC, 62.49% SEN and 100% SPE. In contrast, models trained with structural and metabolic imaging performed less effectively. The model trained on T 2 -weighted MR images achieved 70.00% ACC and 60.27% AUC with a notably high 92.86% SEN (tied with CBV imaging) but a much lower 50.00% SPE. FDG PET imaging models performed similarly, with 70.00% ACC, 64.29% AUC, 64.29% SEN and 75.00% SPE. Multimodal combinations of imaging modalities were explored to potentially take advantage of complementary information and increase model performance. The combination of CBF and CBV outperformed all other multimodal with ACC of 86.67%, 90.62% AUC, 100% SEN and 75.00% SPE. On the other end of the spectrum, the lowest performing multimodal combination was FDG PET combined with CBV, which produced a model with 66.67%, 70.09% AUC, 64.29% SEN and 68.75% SPE. Other multimodal combinations allowed models to accurately classify rodents with AUCs ranging 80.36% to 83.93%. ACCs from 76.67% to 83.33% and reasonably balanced performance. Combining all modalities did not significantly improve performance beyond that of the best-performing unimodal models, with 83.33% ACC and 81.70% AUC, 85.71% SPE and 81.25% SPE. PPV and NPV also varied based on the modality. The CBV model had 78.79% PPV and 92.31% NPV, indicating strong overall prediction quality. The CBF model achieved 100% PPV and 76.19 % NPV. By contrast, T 2 -weighted imaging models had a lower performance with 61.90% PPV and 88.89% NPV. FDG PET models performed at a very similar level with 69.23% PPV and 70.59% NPV. Other multimodal combinations yielded PPVs ranging from 73.68% to 100% and NPVs between 76.19% and 92.31%. Though the combined cerebral perfusion model performed best in terms of AUC and tied with unimodal CBV on accuracy, further investigation was carried out on CBV-based modeling alone for its substantially better balance between positive and negative subjects. Table 1. Summary of model performance metrics. Bold font indicates best performance. An asterisk (*) indicates a tie. FDG = fluorodeoxyglucose PET, T2 = T 2 -weighted MRI, CBV = cerebrovascular volumetry, CBF = cerebrovascular flow. All values are in percentages. Predictive power was computed with observed sample prevalence of 45.16% or 14 out of 30 subjects. Modality ACC AUC SEN SPE PPV NPV T2 70.00 60.27 92.86 50.00 61.90 88.89 CBF 83.33 88.84 62.49 100 100 76.19 CBV 86.67* 86.16 92.86 81.25 78.79 92.31 FDG 70.00 64.29 64.29 75.00 69.23 70.59 T2, CBF 76,67 80.36 78.57 75.00 73.68 80.00 T2, CBV 83.33 83.93 92.86 75.00 76.47 92.31 T2, FDG 72.33 65.62 71.43 75.00 71.43 75.00 CBF, CBV 86.67* 90.62 100 75.00 77.78 100 CBF, FDG 76.67 82.14 71.43 81.25 76.92 76.47 CBV, FDG 66.67 70.09 64.29 68.75 64.29 68.75 T2, CBF, CBV 83.33 84.82 85.71 81.25 75.00 92.31 T2, CBF, FDG 76.67 77.23 71.43 81.25 76.92 76.47 T2, CBV, FDG 83.33 82.14 85.71 81.25 75.00 92.31 CBF, CBV, FDG 80.00 83.04 78.57 81.25 75.00 84.21 All 83.33 81.70 85.71 81.25 75.00 92.31 Hyperplane Visualization and Permutation Testing The hyperplane for the CBV-based SVM model was visualized upon the Schwarz T 2 intracranial template. Beta values were z-scored after extraction for ease of visual interpretation. Positive and negative values of the hyperplane (visualized in red and blue respectively) reflect those features (brain voxels) which are strongly associated with classification in the positive (LID) or negative (NLID) class. To discern only the most significant voxels, permutation testing was performed. Subject labels were randomly permuted over 10,000 iterations. The voxel weights from each iteration’s hyperplane were compared to those from the original model and used to compute a significance map of p-values over the brain data volume. A logical mask of all voxels where p < 0.05 was created and all other voxels were excluded in a refined hyperplane (Figure 2). Anatomical Overlap of Model Hyperplane After permutation testing over 10,000 iterations and cluster size thresholding (50 voxels or larger), there remained 6 significant clusters of voxels within the SVM hyperplane. These clusters are visualized in Figure 3. Positive contributions towards LID classification highlight the roles of the bilateral striatum (clusters 1 and 3) and right piriform cortex (cluster 4). Negative contributions are centered over the right globus pallidus, internal capsule and striatum (cluster 2), right insular cortex (cluster 5), left somatosensory cortex (cluster 6) and right piriform cortex (cluster 5). Note that the striatum showed mixed contributions on the right-hand side rostrocaudally, as did the piriform cortex mediolaterally. Table 2 contains a breakdown of the cluster properties and coordinates. Table 2. Summary of anatomy emphasized by statistically significant hyperplane clusters after size thresholding. Peak weight is z-scored relative to the intracranial voxels of the original hyperplane. Peak location is given in millimeters from the origin in real space. Relative overlap of anatomical regions in the atlas with clusters are shown in percentages. Anatomical coverage was considered significant if the cluster covered at least 20% of an anatomical region of interest. # Peak Location (mm: x, y, z) Peak Weight Volume (mm 3 ) Size (# voxels) Anatomical Coverage 1 -3.6, -1.2, -7.4 3.84 6.57 821 Striatum, left (74.0%) 2 4.0, -2.8, -6.6 -4.88 2.53 316 Globus pallidus, right (30.0%) Internal capsule, right (27.0%) Striatum right (20.4%) 3 3.8, 1.0, -6.8 3.23 2.38 297 Striatum, right (97.4%) 4 4.8, -1.2, -8.8 1.99 0.90 113 Piriform cortex, right (86.3%) 5 5.8, -0.8, -7.2 -5.80 0.58 73 Insular cortex, right (43.1%) Piriform cortex, right (42.3%) 6 -5.6, 2.4, -5.0 -4.83 0.42 53 Somatosensory cortex, left (81.3%) Discussion The present work shows proof of concept for the noninvasive and image-based ability to forecast response to levodopa therapy in a parkinsonian rodent model. Our results show that perfusion-based imaging is the most effective modality for this task. They also indicate that structural and metabolic modalities alone may be insufficient for distinguishing between LID and NLID rodents within the duration of the study. Though some combinations of different modalities improved classification performance, others provided minimal benefits or even harmed performance when compared to unimodal models. Our results align with growing evidence that abnormal neurovascular coupling is central to LID 22 . Previous work has shown that animals with LID exhibit dysregulated vessel responses to dopaminergic stimulation, resulting in increased blood flow to the striatum and midbrain structures at the peak levodopa dose 14, 23 . This vascular hyperreactivity appears to be mediated at least in part by astrocytes, which are critical regulators of local blood flow. Astrocytic calcium signaling can both promote and inhibit vasomotor responses under normal conditions; however, in the dyskinetic state, a shift toward vasodilation predominates, resulting in exaggerated, transient surges in blood flow and dopamine influx 14 . It is therefore plausible that the success of perfusion imaging in classifying future LID vs. non-LID status stems from capturing early, baseline differences in vascular regulation and astrocytic function. Even before LID emerges behaviorally, subtle shifts in cerebrovascular dynamics may be detectable through imaging metrics of CBV or CBF. In contrast, neither T 2 -weighted MR imaging nor FDG PET metabolic imaging revealed comparable predictive power at this stage. While metabolic and structural adaptations certainly evolve over time in PD, they may not reflect the acute or pre-symptomatic vasomotor predisposition that triggers dyskinesia once treatment with levodopa begins. Analysis of our model’s hyperplane pinpointed contributions from subcortical and cortical structures. The role of the striatum and pallidum in LID is well-documented, however, the apparent involvement of regions such as the somatosensory cortex is less well-known. Previous research performed by Alam et al reported through electrophysiological recording that a rodent model of LID had persistently reduce firing frequency and increased irregularity of inhibitory interneurons found within the somatosensory cortex 24 . Since these changes were persistent after levodopa exposure had ended, they concluded this was evidence of maladaptive neuroplasticity as a response to chronic levodopa exposure. Furthermore, Nelson et al also found that functional MRI-based measurement of afferent inhibition in the somatosensory cortex was reduced in a rat LID model 25 . Our model’s usage of these regions for prediction of LID may suggest that perfusion imaging is sensitive to broader neurovascular changes indicative of disruption in dopaminergic and astrocytic activity, suggesting that LID is not solely a nigrostriatal pathology but may also be influenced by more widespread anomalies in neurovascular signaling. The utility of such an image-guided predictive model is twofold: first, it enables investigators to noninvasively identify rodent subjects at heightened risk for LID before symptoms emerge, allowing for more targeted and timely interventions; second, it provides a mechanistic window into the neurovascular and astrocyte-mediated factors that distinguish “susceptible” from “resilient” subjects. With this capability, researchers can use experimental designs in which rodents identified as likely to develop LID can be specifically selected for therapeutic studies, potentially increasing statistical power and refining our understanding of how to delay or deny onset. By validating these findings across different time points and disease models, investigators can more confidently advance new clinical strategies to pre-empt or minimize LID. Translating these methods to human cohorts opens promising avenues for personalized care. Perfusion imaging could be performed in patients with prodromal or early PD to ascertain an individual’s predisposition to neurovascular dysfunction. Clinicians could then identify individuals who exhibit a neurovascular risk profile for LID and tailor levodopa dosages or adjunct therapies accordingly to minimize the long-term impact of dyskinesias. This approach could also guide the timing of alternative interventions (e.g. deep brain stimulation), so they are based on each patient’s personalized likelihood of LID and their progression towards symptomatic expressions. By moving beyond a one-size-fits-all paradigm, clinicians may practice predictive and precise care that not only manages PD more effectively but also preserves quality of life by mitigating dyskinesia risk. Our classification models nevertheless face some limitations. First, although our sample size is within a typical range for rodent imaging studies, any machine learning approach in a small dataset raises concerns of overfitting, even with robust validation procedures like LOOCV. A larger study is warranted to ensure generalizability of the method. Second, our low-dose levodopa regimen is designed to replicate variable LID onset in rats but might not mirror the diversity of clinical presentation in humans. Furthermore, it is possible that some of the NLID rodents would have eventually developed dyskinesia with extended observation periods and/or higher dosage of levodopa, implying that the NLID label in this work may reflect only a delayed onset rather than true resistance. All our animal subjects were females, making generalizability to male rodents is a priority for any future studies. Finally, while our results suggest baseline differences detectable by CBV imaging and SVM, these are not necessarily causative and may in fact be epiphenomena of another yet unknown physiological mechanism contributing to the eventually onset of LID. Conclusions The findings in the present work demonstrate that perfusion-based imaging is a strong predictor of dyskinesia risk in a parkinsonian rodent model, supporting the view that aberrant neurovascular coupling is a critical and potentially modifiable component of LID pathophysiology. If further validated, early detection of perfusion anomalies could guide preclinical research in testing strategies aimed at preventing or delaying LID, for example by targeting astrocyte–vascular interactions. As LID remains a major challenge in long-term levodopa therapy for PD, studies like ours underscore the need to move beyond dopamine-centric models and incorporate the vascular dimension into both mechanistic and therapeutic considerations. Declarations Acknowledgements The authors acknowledge the animal care staff at the Centre for Animal Care and the animals used in this study for their invaluable contributions to our research. Author Contributions Animal care, S.B. and D.Z.; animal evaluation, S.B. and D.Z.; animal tissue harvest, S.B.; conceptualization, J.P. and J.H.K.; conference presentation, J.P and S.K.; funding acquisition, J.P., S.K., S.B., D.Z. and J.H.K.; investigation, J.P., S.K.; methodology, J.P. and J.H.K.; resources, S.B. and D.Z.; supervision, J.H.K.; visualization, J.P.; writing-editing, J.P. and J.H.K.; writing-original draft preparation, J.P. and S.K. Financial Disclosures The authors make the following financial disclosures for the last 12 months. J.P. has none; S.K. was employed by J.K. through funding from a Mitacs Accelerate grant through Cubresa Inc., a manufacturer of imaging equipment for small animal research; S.B. has none; D.Z. has none; J.K. has none. References Turcano P, Mielke MM, Bower JH, et al. Levodopa-induced dyskinesia in Parkinson disease: A population-based cohort study. Neurology 2018;91(24):e2238-e2243. Pavese N, Evans AH, Tai YF, et al. Clinical correlates of levodopa-induced dopamine release in Parkinson disease: a PET study. Neurology 2006;67(9):1612–1617. Jenner P. Molecular mechanisms of L-DOPA-induced dyskinesia. Nat Rev Neurosci 2008;9(9):665–677. Ko JH, Lerner RP, Eidelberg D. Effects of levodopa on regional cerebral metabolism and blood flow. Mov Disord 2015;30(1):54–63. Ahlskog JE, Muenter MD. Frequency of levodopa-related dyskinesias and motor fluctuations as estimated from the cumulative literature. Mov Disord 2001;16(3):448–458. Hely MA, Morris JG, Reid WG, Trafficante R. Sydney Multicenter Study of Parkinson's disease: non-L-dopa-responsive problems dominate at 15 years. Mov Disord 2005;20(2):190–199. Fahn S. Is levodopa toxic? Neurology 1996;47(6 Suppl 3):S184-195. Fahn S. A new look at levodopa based on the ELLDOPA study. J Neural Transm Suppl 2006(70):419–426. Cilia R, Akpalu A, Sarfo FS, et al. The modern pre-levodopa era of Parkinson's disease: insights into motor complications from sub-Saharan Africa. Brain 2014;137(Pt 10):2731–2742. Nyholm D. Duodopa® treatment for advanced Parkinson's disease: a review of efficacy and safety. Parkinsonism Relat Disord 2012;18(8):916–929. Munhoz RP, Picillo M, Fox SH, et al. Eligibility Criteria for Deep Brain Stimulation in Parkinson’s Disease, Tremor, and Dystonia. Canadian Journal of Neurological Sciences / Journal Canadien des Sciences Neurologiques 2016;43(4):462–471. Nichols E, Steinmetz JD, Vollset SE, et al. Estimation of the global prevalence of dementia in 2019 and forecasted prevalence in 2050: an analysis for the Global Burden of Disease Study 2019. The Lancet Public Health 2022;7(2):e105-e125. Tran TN, Vo TNN, Frei K, Truong DD. Levodopa-induced dyskinesia: clinical features, incidence, and risk factors. J Neural Transm (Vienna) 2018;125(8):1109–1117. Booth S, Ramadan A, Zhang D, et al. The Vasomotor Response to Dopamine Is Altered in the Rat Model of l-dopa-Induced Dyskinesia. Mov Disord 2021;36(4):938–947. Booth S. Limitations in effective treatment of Parkinson’s Disease: neuroanatomical substrate of L-Dopa induced dyskinesia and cognitive impairment. 2024. Cenci MA, Lee CS, Björklund A. L-DOPA-induced dyskinesia in the rat is associated with striatal overexpression of prodynorphin- and glutamic acid decarboxylase mRNA. Eur J Neurosci 1998;10(8):2694–2706. Paxinos G. Paxinos & Watson the Rat Brain in Stereotaxic Coordinates: The New Coronal Set: Elsevier, 2005. Schwarz AJ, Danckaert A, Reese T, et al. A stereotaxic MRI template set for the rat brain with tissue class distribution maps and co-registered anatomical atlas: application to pharmacological MRI. Neuroimage 2006;32(2):538–550. Perron J, Ko JH. Review of Quantitative Methods for the Detection of Alzheimer’s Disease with Positron Emission Tomography. Applied sciences 2022;12(22):11463. Brett M, Anton J-L, Valabregue R, Poline J-B. Region of interest analysis using the MarsBar toolbox for SPM 99. Neuroimage 2002;16(2):S497. Vállez Garcia D, Casteels C, Schwarz AJ, Dierckx RAJO, Koole M, Doorduin J. A Standardized Method for the Construction of Tracer Specific PET and SPECT Rat Brain Templates: Validation and Implementation of a Toolbox. PLOS ONE 2015;10(3):e0122363. Cenci MA. Presynaptic Mechanisms of l-DOPA-Induced Dyskinesia: The Findings, the Debate, and the Therapeutic Implications. Front Neurol 2014;5:242. Ohlin KE, Sebastianutto I, Adkins CE, Lundblad C, Lockman PR, Cenci MA. Impact of L-DOPA treatment on regional cerebral blood flow and metabolism in the basal ganglia in a rat model of Parkinson's disease. Neuroimage 2012;61(1):228–239. Alam M, Rumpel R, Jin X, et al. Altered somatosensory cortex neuronal activity in a rat model of Parkinson's disease and levodopa-induced dyskinesias. Exp Neurol 2017;294:19–31. Nelson AJ, Hoque T, Gunraj C, Chen R. Altered somatosensory processing in Parkinson's disease and modulation by dopaminergic medications. Parkinsonism Relat Disord 2018;53:76–81. Additional Declarations Competing interest reported. The cosponsor of the Mitacs Accelerate grant received by J.K., Cubresa Inc., is a manufacturer of imaging devices for small animal studies. All sponsors had no role in the design and conduct of the study; in the collection, analysis and interpretation of data; in the preparation of the manuscript; nor in the review or approval of the manuscript. Cite Share Download PDF Status: Published Journal Publication published 30 Sep, 2025 Read the published version in npj Parkinson's Disease → Version 1 posted Editorial decision: Revision requested 28 Jul, 2025 Reviews received at journal 28 Jun, 2025 Reviews received at journal 15 Jun, 2025 Reviewers agreed at journal 04 Jun, 2025 Reviewers agreed at journal 04 Jun, 2025 Reviewers invited by journal 02 Jun, 2025 Editor assigned by journal 16 May, 2025 Submission checks completed at journal 16 May, 2025 First submitted to journal 13 May, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6658744","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":465374195,"identity":"e8f29afa-ca62-4e12-b58f-dd820223260e","order_by":0,"name":"Jarrad Perron","email":"","orcid":"","institution":"University of Manitoba","correspondingAuthor":false,"prefix":"","firstName":"Jarrad","middleName":"","lastName":"Perron","suffix":""},{"id":465374196,"identity":"5f1b37b9-2ca2-479e-b9fb-c3f8c06fdd8e","order_by":1,"name":"Sophia Krak","email":"","orcid":"","institution":"University of Manitoba","correspondingAuthor":false,"prefix":"","firstName":"Sophia","middleName":"","lastName":"Krak","suffix":""},{"id":465374197,"identity":"af239ada-e17a-4638-bf78-83a98e12b270","order_by":2,"name":"Samuel Booth","email":"","orcid":"","institution":"University of Manitoba","correspondingAuthor":false,"prefix":"","firstName":"Samuel","middleName":"","lastName":"Booth","suffix":""},{"id":465374198,"identity":"92718542-34bc-48c6-a628-dce643013129","order_by":3,"name":"Dali Zhang","email":"","orcid":"","institution":"University of Manitoba","correspondingAuthor":false,"prefix":"","firstName":"Dali","middleName":"","lastName":"Zhang","suffix":""},{"id":465374199,"identity":"5d5cf98c-90f9-4adb-97aa-6c098e57fe4f","order_by":4,"name":"Ji Hyun Ko","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAyElEQVRIiWNgGAWjYBACxmYGZgaGCjCzAYSI1XIGzGxsIEoLEDAzMLZBtBOnhbmdgdng47zD0fyzm9sffNxhx8DffoCwwxJnbjucO+POwcbGmWeSGSTOJBDWcpgXqKXhRmJjM28bM4MBA1Fa5hzOnQ/RUs9gwP+AsJZk3obDuRsgWg4zGEgQtIWx2XDGsfTcjUAtM2eeOc4jcYOALYb9hw9LfKixzp13I/3Bh487quX4+wnYYogeFzz41QOBPEEVo2AUjIJRMAoAHfNHXgWzpVwAAAAASUVORK5CYII=","orcid":"","institution":"University of Manitoba","correspondingAuthor":true,"prefix":"","firstName":"Ji","middleName":"Hyun","lastName":"Ko","suffix":""}],"badges":[],"createdAt":"2025-05-13 21:23:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6658744/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6658744/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41531-025-01133-6","type":"published","date":"2025-09-30T15:57:31+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":84186691,"identity":"e3631818-c4e3-4031-8c7f-1b459db38b0e","added_by":"auto","created_at":"2025-06-09 05:40:00","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":217908,"visible":true,"origin":"","legend":"\u003cp\u003eA graphical representation of the longitudinal data collection chronology (\u003cem\u003eleft\u003c/em\u003e) and the steps taken for each PET-MR scanning session (\u003cem\u003eright\u003c/em\u003e) is given for clarity of presentation. Only images acquired at t1 and in the OFF condition (pre-treatment baseline) were used for the present work. The timing of the imaging procedures is shown relative to levodopa injection (light orange bar) and to anesthesia through nosecone (light blue bar) Full details may be found in Booth \u003cem\u003eet al (under review)\u003c/em\u003e.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6658744/v1/0398b70dcdcf4a99bc87fa1b.png"},{"id":84186689,"identity":"2bfbd7ae-157e-46be-bcca-3fa68701a5e4","added_by":"auto","created_at":"2025-06-09 05:40:00","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":106094,"visible":true,"origin":"","legend":"\u003cp\u003eRepresentation visualization of the CBV hyperplane after permutation testing. Only statistically significant (p \u0026lt; 0.05) voxels from the original hyperplane remain. Results are visualized upon the Schwarz T2 intracranial rat template. Only slices 72 through 92 are shown in increments of 4 slices for clarity of presentation. No cluster size thresholding has been applied.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6658744/v1/c92cd5f6a5829453356c7949.png"},{"id":84186693,"identity":"65c4e13f-8230-4cb1-89d2-142350f9f494","added_by":"auto","created_at":"2025-06-09 05:40:00","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":221616,"visible":true,"origin":"","legend":"\u003cp\u003eVisualization of significant clusters. Clusters are visualized upon the Schwarz intracranial rat template MRI. Crosshairs indicate the center of mass for each cluster. Coordinates of the center of mass and extent over all three spatial dimensions are given in millimeters from the origin in real space. Clusters are shown in order of descending volume.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6658744/v1/a33ced0f269e77bb52a25e79.png"},{"id":92883742,"identity":"f943f993-939b-424a-bec7-8c5e1a26209b","added_by":"auto","created_at":"2025-10-06 16:08:51","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1107956,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6658744/v1/7d89a892-e988-463b-a493-851f85c85afd.pdf"}],"financialInterests":"Competing interest reported. The cosponsor of the Mitacs Accelerate grant received by J.K., Cubresa Inc., is a manufacturer of imaging devices for small animal studies. All sponsors had no role in the design and conduct of the study; in the collection, analysis and interpretation of data; in the preparation of the manuscript; nor in the review or approval of the manuscript.","formattedTitle":"Cerebral Perfusion Imaging Predicts Levodopa-Induced Dyskinesia in Parkinsonian Rat Model","fulltext":[{"header":"Introduction","content":"\u003cp\u003eParkinson's disease (PD) is the second most common neurodegenerative disorder, characterized by dopaminergic cell degeneration. Levodopa (dopamine precursor) is the gold standard for PD treatment (\u0026gt;\u0026thinsp;90% of PD patients are treated with levodopa)\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. Levodopa-induced dyskinesia (LID) is a syndrome associated with involuntary movements such as dystonia, chorea and athetosis. The key mechanism of LID is associated with excessive synaptic dopamine release, yet several neurotransmitters and non-neuronal factors are also involved\u003csup\u003e\u003cspan additionalcitationids=\"CR3\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. Nearly half of all PD patients treated with levodopa develop LID within 5 years, while\u0026thinsp;\u0026gt;\u0026thinsp;90% of all patients eventually develop LID after 15 years\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. It was previously recommended that levodopa treatment must be withheld as long as possible to delay the LID development, however, this was later been revised because it was suggested that the clinical benefit of levodopa is greater than the potential harm\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. It is now settled that levodopa should not be withheld as an initial treatment option because epidemiological studies with drug-na\u0026iuml;ve patients confirmed that the disease progression is more important factor than the levodopa treatment duration for LID development\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. LID is now regarded as one of the inevitable consequences in standard pharmacological PD management in many patients.\u003c/p\u003e \u003cp\u003eThere is currently no disease modifying therapy for PD and most patients eventually develop LID\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. Concerted effort has been vested to find novel LID management strategies or alternative treatments. One major innovation is the usage of intestinal gel formulations that enable the slow and sustained release of levodopa over time (e.g. Duodopa\u0026copy;)\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. This helps reduce so-called off-time and improves quality of life for people with PD, however, it does not solve the fundamental problem, since this approach focuses on refining dosage rather than addressing the underlying issues inherent to treatment with levodopa. In effect, any refinement of levodopa administration simply delays onset rather than treating underlying factors. Deep brain stimulation (DBS) is regarded as the most effective treatment options for LID, which allows reduction of levodopa dose, however, less than 10% of PD patients are eligible for DBS due to many contraindications (e.g. old age, cognitive impairment)\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e. It is also a necessarily invasive procedure and requires significantly greater human resources to pursue. Even with marked improvements in drug delivery and usage of brain stimulation technology, LID is still an unavoidable reality for many PD patients receiving standard pharmacological therapy. This is exacerbated by the fact that neurodegenerative disease is expected to proliferate dramatically by the mid-21st century\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eAnother alternative approach would be preventive treatment. Historically, less attention has been placed on those PD patients who do not develop LID (or very slow LID emergence; non-LID or NLID). Understanding what makes these patients resistant to LID may uncover more attainable, preventive therapeutic targets. Protective factors have previously been identified, but they are often limited to demographic variables (e.g., male sex, older age of disease onset and low levodopa dose) due to the lack of longitudinal data with deep phenotyping and lack of non-invasive investigation due to ethical reasons\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. Animal models have greatly improved our understanding. For example, we have developed a novel low-dose levodopa treatment model that mimics the variability of LID emergence in humans which demonstrated that vasomotor response to dopamine is significantly altered in parkinsonian rodents that developed LID while the ones of non-LID animals was not\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThis previous work in our group generated a neuroimaging dataset of 30 parkinsonian rodent models which were imaged with multiple modalities and time points, including a treatment-naive baseline\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. In the present work, we detail the development and validation of a machine learning-based model to identify prospective LID in parkinsonian rats using pre-treatment baseline neuroimaging data. We investigate the utility of three separate neuroimaging modalities and their multimodal combinations for this purpose: structural magnetic resonance (MR) imaging, cerebral perfusion imaging (flow rate and volumetric data) and fluorodeoxyglucose emission tomography (FDG PET). The proposed animal model will provide a novel window of opportunity for researchers to investigate the brains of LID animals in the earliest possible stage as an avenue to discovery of protective factors against LID prior to chronic levodopa treatment. By identifying those factors that influence resilience or predisposition towards LID, we may identify a novel therapeutic target for the management or prevention of LID, which would dramatically improve quality of life for PD patients.\u003c/p\u003e"},{"header":"Material and Methods","content":"\u003cp\u003e\u003cem\u003eEthical Approval\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eAll experiments received approval from the University of Manitoba\u0026apos;s Animal Care Committee and were conducted in compliance with the guidelines of the Canadian Council on Animal Care. Full details, including detailed ethical guidelines, can be found in Booth \u003cem\u003eet al\u003c/em\u003e (under review).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eData and Code Sharing\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eData, models and MATLAB scripts may be shared by reasonable request to the corresponding author.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eLevodopa Dosage, Behavioral Testing, Imaging Schedule and Acquisition Parameters\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003ePrevious research in our group led to the development of a novel parkinsonian rodent model of LID that was used in the present work\u003csup\u003e14\u003c/sup\u003e. A validated rodent model of LID was developed using unilateral 6-hydroxydopamine (6-OHDA) lesions to induce dopaminergic denervation. This model was used to examine neurovascular and astrocytic calcium responses to dopamine in the dyskinetic state.\u003c/p\u003e\n\u003cp\u003eFemale Sprague-Dawley rats were housed under standard conditions (12-hour light/dark cycle, ad libitum food and water). 30 rats underwent unilateral 6-OHDA lesioning to the right medial forebrain bundle (12.5 \u0026micro;g in 2.5 \u0026micro;L of 0.2% ascorbic acid/saline, infused over 5 minutes). To minimize extrasynaptic degradation, pargyline hydrochloride (10 mg/kg, intraperitoneal) was administered 30 minutes prior to surgery. After a two-week recovery period, animals were screened for hemiparkinsonian-like behaviors using the cylinder test.\u003c/p\u003e\n\u003cp\u003eFollowing lesion validation, animals received daily levodopa (2 mg/kg subcutaneous) co-administered with benserazide (15 mg/kg) for 22 days. A control group (n = 5) received isotonic saline. Dyskinesia severity was assessed on days 1, 11 and 22 using the abnormal involuntary movements (AIM) test, scoring axial, orolingual and limb movements on a 0\u0026ndash;4 scale. Animals with AIM scores \u0026ge;2 in at least one category by day 22 were classified as LID-positive\u003csup\u003e16\u003c/sup\u003e. \u0026nbsp; \u0026nbsp; \u0026nbsp;To confirm lesion integrity immunohistochemical staining was performed on striatal and substantia nigra (SN) sections. First, tyrosine hydroxylase immunostaining quantified dopaminergic cell loss in the SN. Animals with \u0026lt;90% dopaminergic depletion were excluded.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOur previous study demonstrated that approximately 50% of the treated animals develop LID-like behaviors\u003csup\u003e14\u003c/sup\u003e. \u0026nbsp;In the present study, 14 of 30 animals developed LID. Animals were scanned with PET-MR both OFF and ON levodopa at two timepoints: t\u003csub\u003e1\u003c/sub\u003e, when levodopa treatment first began and LID symptoms had not yet appeared (OFF\u003csub\u003e1\u003c/sub\u003e and ON\u003csub\u003e1\u003c/sub\u003e studies were performed on consecutive days; at OFF\u003csub\u003e1\u003c/sub\u003e, animals had never received levodopa treatment), and t\u003csub\u003e2\u003c/sub\u003e, when levodopa medication had been administered for 21 days (OFF\u003csub\u003e2\u003c/sub\u003e and ON\u003csub\u003e2\u003c/sub\u003e studies were also performed on consecutive days). On the day of OFF\u003csub\u003e2\u003c/sub\u003e, animals were administered levodopa immediately after scanning sessions. \u0026nbsp;Dyskinesia was evaluated using three abnormal involuntary movement (AIMS) tests on days 1, 11 and 23 of levodopa usage. AIMS refers to uncoordinated, uncontrollable movements that occur without conscious effort and are reflected side effects of PD. The details of AIMS test scoring are described elsewhere\u003csup\u003e16\u003c/sup\u003e. The animals were sacrificed by perfusion fixation the day following the last AIMs test and their brain tissue was removed.\u003c/p\u003e\n\u003cp\u003eAdministration protocol differed by imaging sessions. For ON sessions, animals were given 2 mg/kg levodopa with 15 mg/kg benserazide. An equivalent volume of isotonic saline was given prior to scanning for OFF sessions. After twenty minutes, the animal was placed in an induction chamber with 5% isoflurane and 0.5 L/min oxygen flow to induce anesthesia. The animals were shifted to a nasal cone with an oxygen flow rate of 0.5 liters per minute and an isoflurane concentration of 1.5\u0026ndash; 2%. To administer tracer during the scan, a 23-gauge needle was used to cannulate the tail vein before loading onto the scanning bed. Rats were placed face down in the anesthetic system\u0026apos;s nasal cone. Throughout the imaging procedure, body temperature and respiration were recorded and kept constant at 37 \u0026plusmn; 0.5 \u0026deg;C and 60-80 breaths per minute, respectively.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eStudies were acquired with an MR Solutions 7.0 Tesla drymag scanner with the clip-on PET module. The PET and MRI imaging were performed sequentially, and a rat head RF coil was used for all MR imaging.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eStructural MR, dynamic susceptibility contrast (DSC) MR (gadolinium contrast enhancement) and FDG PET studies were collected. Anatomical images were acquired using a fast spin echo T\u003csub\u003e2\u003c/sub\u003e-weighted pulse sequence (26 x 1 mm coronal slices, TR = 4000 ms, TE = 45 ms, matrix size = 250 x 256 and 3 averages) over an acquisition time of 12 minutes. DSC perfusion maps were acquired using a gradient-echo planar imaging sequence (TR = 333 ms, TE = 20 ms, field of view = 30 x 30 mm, matrix size = 64 x 64, slice gap = 0.2 mm, slice thickness = 1.5 mm, temporal resolution = 0.333 s) such that the entire rostral striatum and majority of caudal striatum was included in the acquisition. 180 acquisitions were taken over 60 seconds and 0.3 mmol/kg of Gadovist (gadobutrol-based contrast agent) in isotonic saline was injected through tail vein cannulation. From these, relative cerebral blood flow (CBF) images were generated using custom scripts in the MATLAB 2023a environment (MathWorks Inc., Natick, MA, USA) by voxel-wise calculation of the transverse relaxation rate curve maximum,\u003c/p\u003e\n\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\" height=\"66\" width=\"176\"\u003e\u003c/p\u003e\n\u003cp\u003ewhere S(t) is signal intensity at time t, S\u003csub\u003e0\u003c/sub\u003e is the pre-contrast baseline intensity and TE is the echo time of the pulse sequence. Cerebral perfusion volume (CBV) images were generated by integrating over the perfusion flow volume. PET images were acquired immediately after DSC acquisition and the bed position of the subject moved to the PET coil for a 15-minute acquisition. PET images were reconstructed using filtered back projection. A summary of these methods is shown in Figure 1. Full experimental details may be found in Booth \u003cem\u003eet al\u0026nbsp;\u003c/em\u003e(under review).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eNeuroimaging Data Preprocessing\u003c/em\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ePreprocessing operations were performed using the Statistical Parametric Mapping (SPM12) software package (http://www.fil.ion.ucl.ac.uk/spm/ Wellcome Department of Cognitive Neurology, London, UK) within the MATLAB R2023a environment. All images were resized by \u0026times;10, then manually re-oriented to align with the Schwartz rat T\u003csub\u003e2\u003c/sub\u003e template, which was also resized by \u0026times;10. The Schwartz template is based on 97 anatomical MR studies from Sprague Dawley rats, volumetrically reconstructed into Paxinos and Watson space \u003csup\u003e17, 18\u003c/sup\u003e. All functional images (PET, CBF, and CBV) were coregistered to T\u003csub\u003e2\u003c/sub\u003e-MRI using SPM12\u0026rsquo;s default coregistration function and were visually inspected. T\u003csub\u003e2\u003c/sub\u003e-weighted structural images were spatially normalized into template space using the \u0026quot;Old Normalize\u0026quot; function in SPM12, with no affine regularization, a nonlinear frequency cutoff of 25 and 16 iterations. The same deformations were applied to write the coregistered functional images into template space. Resulting images were generated with the template image\u0026apos;s bounding box ([-95, -168, -138; 95, 70, 52]) and voxel size (2x2x2 mm). All functional images (FDG PET, CBF, and CBV) images were smoothed with a Gaussian kernel with full width at half maximum of 8x8x8 mm. Each image was proportionally scaled to the mean signal intensity of the entire volume.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eComparison of Neuroimaging Modalities with a Support Vector Machine\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eA support vector machine (SVM) is a very robust machine intelligence model that is commonly used for the purpose of binary classification. It functions on the principle of maximizing the margin between classes in an extremely high-dimensional feature space\u003csup\u003e19\u003c/sup\u003e. A series of SVMs (a priori defined hyperparameters: linear kernel, 5% outlier fraction and iterative single-data algorithm as optimizer) was trained to classify LID vs. NLID subjects, and modality-specific performance between models (T\u003csub\u003e2\u003c/sub\u003e-weighted MRI, CBV, CBF and FDG PET) were compared by area-under-curve (AUC), sensitivity (SEN), specificity (SPE), accuracy (ACC) and both positive and negative predictive values (PPV/NPV). Multimodal combinations of neuroimaging data were also investigated as a possible method of increasing performance through complementary information.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eVisualizing Model Decisions\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe hyperplane of the best performing model was visualized and underwent permutation testing. Study labels (LID vs. NLID) were permuted randomly and an SVM was trained on this permuted data. The original hyperplane weights were ranked in the permuted distribution of hyperplane weights at each voxel and p-value was determined. Weights were deemed significant if they were ranked within the top 500 of 10,000 iterations (p \u0026lt; 0.05). Only the significant voxel\u0026rsquo;s hyperplane region weight was visualized and carried forward for further anatomical analysis under the assumption that these voxels are meaningful for LID vs NLID classification. The original hyperplane was z-scored relative to the intracranial voxel weights and then masked by the statistically significant voxels of the permuted hyperplane. Clusters of voxels were then characterized using MarsBaR and a pre-determined size threshold of 50 voxels\u003csup\u003e20\u003c/sup\u003e. The anatomy was identified by volume-of-interest analysis with\u0026nbsp;the Schwarz atlas found in the Small Animal Molecular Imaging Toolbox (SAMIT)\u003csup\u003e21\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eValidation\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eDue to the modest sample size of our rodent population, we employed leave-one-out cross-validation (LOOCV) to estimate the generalizability of our models. This also allowed us to infer the effect of overfitting upon base models trained on all available data.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eThe purpose of the present study was limited to developing a machine learning model that classified LID vs. NLID animals using baseline neuroimaging data. More details on behavioral test results and overall outcomes will be published elsewhere (Booth et al., under review).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eModality-based Performance Comparison\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eClassification of LID vs. NLID status by SVM was evaluated on 4 separate neuroimaging modalities (anatomical T\u003csub\u003e2\u003c/sub\u003e, CBF, CBV and FDG PET) for our rodent model (n\u003csub\u003eLID\u003c/sub\u003e = 14 and n\u003csub\u003eNLID\u003c/sub\u003e = 16) and their multimodal combinations. The performance of each was estimated by examining ACC, AUC, SEN, SPE and predictive values in LOOCV over the entire subject set. All results are compiled in Table 1.\u003c/p\u003e\n\u003cp\u003eSVM performance varied considerably depending on the neuroimaging modality and modality combinations used for classification of LID versus NLID subjects. Among individual modalities, models trained on perfusion data demonstrated very high performance. The SVM trained on CBV imaging achieved 86.67% ACC and 86.16% AUC. It also demonstrated strong performance with balanced SEN and SPE at 92.86% and 81.25% respectively. Similarly, the CBF model performed comparably well, with 88.84% AUC, 83.33% ACC, 62.49% SEN and 100% SPE.\u003c/p\u003e\n\u003cp\u003eIn contrast, models trained with structural and metabolic imaging performed less effectively. The model trained on T\u003csub\u003e2\u003c/sub\u003e-weighted MR images achieved 70.00% ACC and 60.27% AUC with a notably high 92.86% SEN (tied with CBV imaging) but a much lower 50.00% SPE. FDG PET imaging models performed similarly, with 70.00% ACC, 64.29% AUC, 64.29% SEN and 75.00% SPE.\u003c/p\u003e\n\u003cp\u003eMultimodal combinations of imaging modalities were explored to potentially take advantage of complementary information and increase model performance. The combination of CBF and CBV outperformed all other multimodal with ACC of 86.67%, 90.62% AUC, 100% SEN and 75.00% SPE. On the other end of the spectrum, the lowest performing multimodal combination was FDG PET combined with CBV, which produced a model with 66.67%, 70.09% AUC, 64.29% SEN and 68.75% SPE.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOther multimodal combinations allowed models to accurately classify rodents with AUCs ranging 80.36% to 83.93%. ACCs from 76.67% to 83.33% and reasonably balanced performance. Combining all modalities did not significantly improve performance beyond that of the best-performing unimodal models, with 83.33% ACC and 81.70% AUC, 85.71% SPE and 81.25% SPE.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ePPV and NPV also varied based on the modality. The CBV model had 78.79% PPV and 92.31% NPV, indicating strong overall prediction quality. The CBF model achieved 100% PPV and 76.19 % NPV. By contrast, T\u003csub\u003e2\u003c/sub\u003e-weighted imaging models had a lower performance with 61.90% PPV and 88.89% NPV. FDG PET models performed at a very similar level with 69.23% PPV and 70.59% NPV. Other multimodal combinations yielded PPVs ranging from 73.68% to 100% and NPVs between 76.19% and 92.31%.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThough the combined cerebral perfusion model performed best in terms of AUC and tied with unimodal CBV on accuracy, further investigation was carried out on CBV-based modeling alone for its substantially better balance between positive and negative subjects.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1.\u003c/strong\u003e Summary of model performance metrics. Bold font indicates best performance. An asterisk (*) indicates a tie. FDG = fluorodeoxyglucose PET, T2 = T\u003csub\u003e2\u003c/sub\u003e-weighted MRI, CBV = cerebrovascular volumetry, CBF = cerebrovascular flow. All values are in percentages. Predictive power was computed with observed sample prevalence of 45.16% or 14 out of 30 subjects.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.5962%;\"\u003e\n \u003cp\u003eModality\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003eACC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003eSEN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003eSPE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003ePPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003eNPV\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.5962%;\"\u003e\n \u003cp\u003eT2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e70.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e60.27\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e92.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e50.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e61.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e88.89\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.5962%;\"\u003e\n \u003cp\u003eCBF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e83.33\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e88.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e62.49\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e100\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e100\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e76.19\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.5962%;\"\u003e\n \u003cp\u003eCBV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e86.67*\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e86.16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e92.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e81.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e78.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e92.31\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.5962%;\"\u003e\n \u003cp\u003eFDG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e70.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e64.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e64.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e75.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e69.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e70.59\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.5962%;\"\u003e\n \u003cp\u003eT2, CBF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e76,67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e80.36\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e78.57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e75.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e73.68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e80.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.5962%;\"\u003e\n \u003cp\u003eT2, CBV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e83.33\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e83.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e92.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e75.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e76.47\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e92.31\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.5962%;\"\u003e\n \u003cp\u003eT2, FDG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e72.33\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e65.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e71.43\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e75.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e71.43\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e75.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.5962%;\"\u003e\n \u003cp\u003eCBF, CBV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e86.67*\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e90.62\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e100\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e75.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e77.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e100\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.5962%;\"\u003e\n \u003cp\u003eCBF, FDG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e76.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e82.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e71.43\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e81.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e76.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e76.47\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.5962%;\"\u003e\n \u003cp\u003eCBV, FDG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e66.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e70.09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e64.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e68.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e64.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e68.75\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.5962%;\"\u003e\n \u003cp\u003eT2, CBF, CBV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e83.33\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e84.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e85.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e81.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e75.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e92.31\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.5962%;\"\u003e\n \u003cp\u003eT2, CBF, FDG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e76.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e77.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e71.43\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e81.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e76.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e76.47\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.5962%;\"\u003e\n \u003cp\u003eT2, CBV, FDG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e83.33\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e82.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e85.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e81.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e75.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e92.31\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.5962%;\"\u003e\n \u003cp\u003eCBF, CBV, FDG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e80.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e83.04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e78.57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e81.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e75.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e84.21\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.5962%;\"\u003e\n \u003cp\u003eAll\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e83.33\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e81.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e85.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.1795%;\"\u003e\n \u003cp\u003e81.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e75.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6218%;\"\u003e\n \u003cp\u003e92.31\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eHyperplane Visualization and Permutation Testing\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe hyperplane for the CBV-based SVM model was visualized upon the Schwarz T\u003csub\u003e2\u003c/sub\u003e intracranial template. Beta values were z-scored after extraction for ease of visual interpretation. Positive and negative values of the hyperplane (visualized in red and blue respectively) reflect those features (brain voxels) which are strongly associated with classification in the positive (LID) or negative (NLID) class. To discern only the most significant voxels, permutation testing was performed. Subject labels were randomly permuted over 10,000 iterations. The voxel weights from each iteration\u0026rsquo;s hyperplane were compared to those from the original model and used to compute a significance map of p-values over the brain data volume. A logical mask of all voxels where p \u0026lt; 0.05 was created and all other voxels were excluded in a refined hyperplane (Figure 2).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAnatomical Overlap of Model Hyperplane\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eAfter permutation testing over 10,000 iterations and cluster size thresholding (50 voxels or larger), there remained 6 significant clusters of voxels within the SVM hyperplane. These clusters are visualized in Figure 3. Positive contributions towards LID classification highlight the roles of the bilateral striatum (clusters 1 and 3) and right piriform cortex (cluster 4). Negative contributions are centered over the right globus pallidus, internal capsule and striatum (cluster 2), right insular cortex (cluster 5), left somatosensory cortex (cluster 6) and right piriform cortex (cluster 5). Note that the striatum showed mixed contributions on the right-hand side rostrocaudally, as did the piriform cortex mediolaterally. Table 2 contains a breakdown of the cluster properties and coordinates.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2.\u003c/strong\u003e Summary of anatomy emphasized by statistically significant hyperplane clusters after size thresholding. Peak weight is z-scored relative to the intracranial voxels of the original hyperplane. Peak location is given in millimeters from the origin in real space. \u0026nbsp; Relative overlap of anatomical regions in the atlas with clusters are shown in percentages. Anatomical coverage was considered significant if the cluster covered at least 20% of an anatomical region of interest.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 4.55975%;\"\u003e\n \u003cp\u003e#\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 21.3836%;\"\u003e\n \u003cp\u003ePeak Location (mm: x, y, z)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.5346%;\"\u003e\n \u003cp\u003ePeak Weight\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.3774%;\"\u003e\n \u003cp\u003eVolume (mm\u003csup\u003e3\u003c/sup\u003e)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.4654%;\"\u003e\n \u003cp\u003eSize\u003c/p\u003e\n \u003cp\u003e(# voxels)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 38.6792%;\"\u003e\n \u003cp\u003eAnatomical Coverage\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 4.55975%;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 21.3836%;\"\u003e\n \u003cp\u003e-3.6, -1.2, -7.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.5346%;\"\u003e\n \u003cp\u003e3.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.3774%;\"\u003e\n \u003cp\u003e6.57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.4654%;\"\u003e\n \u003cp\u003e821\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 38.6792%;\"\u003e\n \u003cp\u003eStriatum, left (74.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 4.55975%;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 21.3836%;\"\u003e\n \u003cp\u003e4.0, -2.8, -6.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.5346%;\"\u003e\n \u003cp\u003e-4.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.3774%;\"\u003e\n \u003cp\u003e2.53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.4654%;\"\u003e\n \u003cp\u003e316\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 38.6792%;\"\u003e\n \u003cp\u003eGlobus pallidus, right (30.0%)\u003c/p\u003e\n \u003cp\u003eInternal capsule, right (27.0%)\u003c/p\u003e\n \u003cp\u003eStriatum right (20.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 4.55975%;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 21.3836%;\"\u003e\n \u003cp\u003e3.8, 1.0, -6.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.5346%;\"\u003e\n \u003cp\u003e3.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.3774%;\"\u003e\n \u003cp\u003e2.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.4654%;\"\u003e\n \u003cp\u003e297\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 38.6792%;\"\u003e\n \u003cp\u003eStriatum, right (97.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 4.55975%;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 21.3836%;\"\u003e\n \u003cp\u003e4.8, -1.2, -8.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.5346%;\"\u003e\n \u003cp\u003e1.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.3774%;\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.4654%;\"\u003e\n \u003cp\u003e113\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 38.6792%;\"\u003e\n \u003cp\u003ePiriform cortex, right (86.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 4.55975%;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 21.3836%;\"\u003e\n \u003cp\u003e5.8, -0.8, -7.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.5346%;\"\u003e\n \u003cp\u003e-5.80\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.3774%;\"\u003e\n \u003cp\u003e0.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.4654%;\"\u003e\n \u003cp\u003e73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 38.6792%;\"\u003e\n \u003cp\u003eInsular cortex, right (43.1%)\u003c/p\u003e\n \u003cp\u003ePiriform cortex, right (42.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 4.55975%;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 21.3836%;\"\u003e\n \u003cp\u003e-5.6, 2.4, -5.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.5346%;\"\u003e\n \u003cp\u003e-4.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.3774%;\"\u003e\n \u003cp\u003e0.42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.4654%;\"\u003e\n \u003cp\u003e53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 38.6792%;\"\u003e\n \u003cp\u003eSomatosensory cortex, left (81.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe present work shows proof of concept for the noninvasive and image-based ability to forecast response to levodopa therapy in a parkinsonian rodent model. Our results show that perfusion-based imaging is the most effective modality for this task. They also indicate that structural and metabolic modalities alone may be insufficient for distinguishing between LID and NLID rodents within the duration of the study. Though some combinations of different modalities improved classification performance, others provided minimal benefits or even harmed performance when compared to unimodal models.\u003c/p\u003e\n\u003cp\u003eOur results align with growing evidence that abnormal neurovascular coupling is central to LID\u003csup\u003e22\u003c/sup\u003e. Previous work has shown that animals with LID exhibit dysregulated vessel responses to dopaminergic stimulation, resulting in increased blood flow to the striatum and midbrain structures at the peak levodopa dose\u003csup\u003e14, 23\u003c/sup\u003e. This vascular hyperreactivity appears to be mediated at least in part by astrocytes, which are critical regulators of local blood flow. Astrocytic calcium signaling can both promote and inhibit vasomotor responses under normal conditions; however, in the dyskinetic state, a shift toward vasodilation predominates, resulting in exaggerated, transient surges in blood flow and dopamine influx\u003csup\u003e14\u003c/sup\u003e. It is therefore plausible that the success of perfusion imaging in classifying future LID vs. non-LID status stems from capturing early, baseline differences in vascular regulation and astrocytic function. Even before LID emerges behaviorally, subtle shifts in cerebrovascular dynamics may be detectable through imaging metrics of CBV or CBF. In contrast, neither T\u003csub\u003e2\u003c/sub\u003e-weighted MR imaging nor FDG PET metabolic imaging revealed comparable predictive power at this stage. While metabolic and structural adaptations certainly evolve over time in PD, they may not reflect the acute or pre-symptomatic vasomotor predisposition that triggers dyskinesia once treatment with levodopa begins.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAnalysis of our model\u0026rsquo;s hyperplane pinpointed contributions from subcortical and cortical structures. The role of the striatum and pallidum in LID is well-documented, however, the apparent involvement of regions such as the somatosensory cortex is less well-known. Previous research performed by Alam \u003cem\u003eet al\u003c/em\u003e reported through electrophysiological recording that a rodent model of LID had persistently reduce firing frequency and increased irregularity of inhibitory interneurons found within the somatosensory cortex\u003csup\u003e24\u003c/sup\u003e. Since these changes were persistent after levodopa exposure had ended, they concluded this was evidence of maladaptive neuroplasticity as a response to chronic levodopa exposure. Furthermore, Nelson \u003cem\u003eet al\u003c/em\u003e also found that functional MRI-based measurement of afferent inhibition in the somatosensory cortex was reduced in a rat LID model\u003csup\u003e25\u003c/sup\u003e. Our model\u0026rsquo;s usage of these regions for prediction of LID may suggest that perfusion imaging is sensitive to broader neurovascular changes indicative of disruption in dopaminergic and astrocytic activity, suggesting that LID is not solely a nigrostriatal pathology but may also be influenced by more widespread anomalies in neurovascular signaling.\u003c/p\u003e\n\u003cp\u003eThe utility of such an image-guided predictive model is twofold: first, it enables investigators to noninvasively identify rodent subjects at heightened risk for LID before symptoms emerge, allowing for more targeted and timely interventions; second, it provides a mechanistic window into the neurovascular and astrocyte-mediated factors that distinguish \u0026ldquo;susceptible\u0026rdquo; from \u0026ldquo;resilient\u0026rdquo; subjects. With this capability, researchers can use experimental designs in which rodents identified as likely to develop LID can be specifically selected for therapeutic studies, potentially increasing statistical power and refining our understanding of how to delay or deny onset. By validating these findings across different time points and disease models, investigators can more confidently advance new clinical strategies to pre-empt or minimize LID.\u003c/p\u003e\n\u003cp\u003eTranslating these methods to human cohorts opens promising avenues for personalized care. Perfusion imaging could be performed in patients with prodromal or early PD to ascertain an individual\u0026rsquo;s predisposition to neurovascular dysfunction. Clinicians could then identify individuals who exhibit a neurovascular risk profile for LID and tailor levodopa dosages or adjunct therapies accordingly to minimize the long-term impact of dyskinesias. This approach could also guide the timing of alternative interventions (e.g. deep brain stimulation), so they are based on each patient\u0026rsquo;s personalized likelihood of LID and their progression towards symptomatic expressions. By moving beyond a one-size-fits-all paradigm, clinicians may practice predictive and precise care that not only manages PD more effectively but also preserves quality of life by mitigating dyskinesia risk.\u003c/p\u003e\n\u003cp\u003eOur classification models nevertheless face some limitations. First, although our sample size is within a typical range for rodent imaging studies, any machine learning approach in a small dataset raises concerns of overfitting, even with robust validation procedures like LOOCV. A larger study is warranted to ensure generalizability of the method. Second, our low-dose levodopa regimen is designed to replicate variable LID onset in rats but might not mirror the diversity of clinical presentation in humans. Furthermore, it is possible that some of the NLID rodents would have eventually developed dyskinesia with extended observation periods and/or higher dosage of levodopa, implying that the NLID label in this work may reflect only a delayed onset rather than true resistance. All our animal subjects were females, making generalizability to male rodents is a priority for any future studies. Finally, while our results suggest baseline differences detectable by CBV imaging and SVM, these are not necessarily causative and may in fact be epiphenomena of another yet unknown physiological mechanism contributing to the eventually onset of LID.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eThe findings in the present work demonstrate that perfusion-based imaging is a strong predictor of dyskinesia risk in a parkinsonian rodent model, supporting the view that aberrant neurovascular coupling is a critical and potentially modifiable component of LID pathophysiology. If further validated, early detection of perfusion anomalies could guide preclinical research in testing strategies aimed at preventing or delaying LID, for example by targeting astrocyte\u0026ndash;vascular interactions. As LID remains a major challenge in long-term levodopa therapy for PD, studies like ours underscore the need to move beyond dopamine-centric models and incorporate the vascular dimension into both mechanistic and therapeutic considerations.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors acknowledge the animal care staff at the Centre for Animal Care and the animals used in this study for their invaluable contributions to our research.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAnimal care, S.B. and D.Z.; animal evaluation, S.B. and D.Z.; animal tissue harvest, S.B.; conceptualization, J.P. and J.H.K.; conference presentation, J.P and S.K.; funding acquisition, J.P., S.K., S.B., D.Z. and J.H.K.; investigation, J.P., S.K.; methodology, J.P. and J.H.K.; resources, S.B. and D.Z.; supervision, J.H.K.; visualization, J.P.; writing-editing, J.P. and J.H.K.; writing-original draft preparation, J.P. and S.K.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFinancial Disclosures\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors make the following financial disclosures for the last 12 months. J.P. has none; S.K. was employed by J.K. through funding from a Mitacs Accelerate grant through Cubresa Inc., a manufacturer of imaging equipment for small animal research; S.B. has none; D.Z. has none; J.K. has none.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eTurcano P, Mielke MM, Bower JH, et al. Levodopa-induced dyskinesia in Parkinson disease: A population-based cohort study. Neurology 2018;91(24):e2238-e2243.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePavese N, Evans AH, Tai YF, et al. Clinical correlates of levodopa-induced dopamine release in Parkinson disease: a PET study. Neurology 2006;67(9):1612\u0026ndash;1617.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJenner P. Molecular mechanisms of L-DOPA-induced dyskinesia. Nat Rev Neurosci 2008;9(9):665\u0026ndash;677.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKo JH, Lerner RP, Eidelberg D. Effects of levodopa on regional cerebral metabolism and blood flow. Mov Disord 2015;30(1):54\u0026ndash;63.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAhlskog JE, Muenter MD. Frequency of levodopa-related dyskinesias and motor fluctuations as estimated from the cumulative literature. Mov Disord 2001;16(3):448\u0026ndash;458.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHely MA, Morris JG, Reid WG, Trafficante R. Sydney Multicenter Study of Parkinson's disease: non-L-dopa-responsive problems dominate at 15 years. Mov Disord 2005;20(2):190\u0026ndash;199.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFahn S. Is levodopa toxic? Neurology 1996;47(6 Suppl 3):S184-195.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFahn S. A new look at levodopa based on the ELLDOPA study. J Neural Transm Suppl 2006(70):419\u0026ndash;426.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCilia R, Akpalu A, Sarfo FS, et al. The modern pre-levodopa era of Parkinson's disease: insights into motor complications from sub-Saharan Africa. Brain 2014;137(Pt 10):2731\u0026ndash;2742.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNyholm D. Duodopa\u0026reg; treatment for advanced Parkinson's disease: a review of efficacy and safety. Parkinsonism Relat Disord 2012;18(8):916\u0026ndash;929.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMunhoz RP, Picillo M, Fox SH, et al. Eligibility Criteria for Deep Brain Stimulation in Parkinson\u0026rsquo;s Disease, Tremor, and Dystonia. Canadian Journal of Neurological Sciences / Journal Canadien des Sciences Neurologiques 2016;43(4):462\u0026ndash;471.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNichols E, Steinmetz JD, Vollset SE, et al. Estimation of the global prevalence of dementia in 2019 and forecasted prevalence in 2050: an analysis for the Global Burden of Disease Study 2019. The Lancet Public Health 2022;7(2):e105-e125.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTran TN, Vo TNN, Frei K, Truong DD. Levodopa-induced dyskinesia: clinical features, incidence, and risk factors. J Neural Transm (Vienna) 2018;125(8):1109\u0026ndash;1117.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBooth S, Ramadan A, Zhang D, et al. The Vasomotor Response to Dopamine Is Altered in the Rat Model of l-dopa-Induced Dyskinesia. Mov Disord 2021;36(4):938\u0026ndash;947.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBooth S. Limitations in effective treatment of Parkinson\u0026rsquo;s Disease: neuroanatomical substrate of L-Dopa induced dyskinesia and cognitive impairment. 2024.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCenci MA, Lee CS, Bj\u0026ouml;rklund A. L-DOPA-induced dyskinesia in the rat is associated with striatal overexpression of prodynorphin- and glutamic acid decarboxylase mRNA. Eur J Neurosci 1998;10(8):2694\u0026ndash;2706.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePaxinos G. Paxinos \u0026amp; Watson the Rat Brain in Stereotaxic Coordinates: The New Coronal Set: Elsevier, 2005.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchwarz AJ, Danckaert A, Reese T, et al. A stereotaxic MRI template set for the rat brain with tissue class distribution maps and co-registered anatomical atlas: application to pharmacological MRI. Neuroimage 2006;32(2):538\u0026ndash;550.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePerron J, Ko JH. Review of Quantitative Methods for the Detection of Alzheimer\u0026rsquo;s Disease with Positron Emission Tomography. Applied sciences 2022;12(22):11463.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrett M, Anton J-L, Valabregue R, Poline J-B. Region of interest analysis using the MarsBar toolbox for SPM 99. Neuroimage 2002;16(2):S497.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eV\u0026aacute;llez Garcia D, Casteels C, Schwarz AJ, Dierckx RAJO, Koole M, Doorduin J. A Standardized Method for the Construction of Tracer Specific PET and SPECT Rat Brain Templates: Validation and Implementation of a Toolbox. PLOS ONE 2015;10(3):e0122363.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCenci MA. Presynaptic Mechanisms of l-DOPA-Induced Dyskinesia: The Findings, the Debate, and the Therapeutic Implications. Front Neurol 2014;5:242.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOhlin KE, Sebastianutto I, Adkins CE, Lundblad C, Lockman PR, Cenci MA. Impact of L-DOPA treatment on regional cerebral blood flow and metabolism in the basal ganglia in a rat model of Parkinson's disease. Neuroimage 2012;61(1):228\u0026ndash;239.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlam M, Rumpel R, Jin X, et al. Altered somatosensory cortex neuronal activity in a rat model of Parkinson's disease and levodopa-induced dyskinesias. Exp Neurol 2017;294:19\u0026ndash;31.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNelson AJ, Hoque T, Gunraj C, Chen R. Altered somatosensory processing in Parkinson's disease and modulation by dopaminergic medications. Parkinsonism Relat Disord 2018;53:76\u0026ndash;81.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"npj-parkinsons-disease","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"npjparkd","sideBox":"Learn more about [npj Parkinson's Disease](http://www.nature.com/npjparkd/)","snPcode":"41531","submissionUrl":"https://submission.springernature.com/new-submission/41531/3","title":"npj Parkinson's Disease","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"levodopa induced dyskinesia, rodent, perfusion, neuroimaging, machine learning","lastPublishedDoi":"10.21203/rs.3.rs-6658744/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6658744/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eApproximately half of Parkinson\u0026rsquo;s disease (PD) patients manifest motor complications related to treatment called levodopa-induced dyskinesia (LID). Preventing onset of LID is crucial to long-term management of PD, but the reasons why some patients develop LID are unclear, for instance, it is unknown if vascular or neuroinflammatory abnormalities exist prior to levodopa therapy or are a response to chronic exposure. The ability to prognosticate predisposition to LID would be valuable for the management of LID and for the investigation of strategies for its mitigation. Thirty rats received 6-hydroxydopamine to induce parkinsonism-like behaviors before treatment with levodopa (2 mg/kg) daily for 22 days. Fourteen developed LID-like behaviors. Fluorodeoxyglucose PET, T\u003csub\u003e2\u003c/sub\u003e-weighted MRI and cerebral perfusion imaging were collected before treatment. Support vector machines were trained to classify LID vs. non-LID animals. Volumetric perfusion imaging performed best overall with 86.16% area-under-curve, 86.67% accuracy, 92.86% sensitivity, 81.25% specificity for classifying animals with LID vs. non-LID from treatment-na\u0026iuml;ve baseline imaging in leave-one-out cross-validation. We have demonstrated proof-of-concept for imaging-based classification of a parkinsonian rat model. The ability to non-invasively identify a predisposition to LID would allow for more targeted investigations into the risk factors for LID and its prevention in the earliest stages.\u003c/p\u003e","manuscriptTitle":"Cerebral Perfusion Imaging Predicts Levodopa-Induced Dyskinesia in Parkinsonian Rat Model","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-09 05:39:56","doi":"10.21203/rs.3.rs-6658744/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-07-28T16:38:15+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-06-28T20:19:50+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-06-15T21:09:19+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"40809624948366598752645796012492063409","date":"2025-06-04T15:22:04+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"103662754674203876242885103641512360387","date":"2025-06-04T13:26:16+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-06-02T15:44:48+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-05-16T07:45:40+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-05-16T06:43:32+00:00","index":"","fulltext":""},{"type":"submitted","content":"npj Parkinson's Disease","date":"2025-05-13T21:21:36+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"npj-parkinsons-disease","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"npjparkd","sideBox":"Learn more about [npj Parkinson's Disease](http://www.nature.com/npjparkd/)","snPcode":"41531","submissionUrl":"https://submission.springernature.com/new-submission/41531/3","title":"npj Parkinson's Disease","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"a6037767-e9a4-45ff-a18a-1c455c3b58c3","owner":[],"postedDate":"June 9th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":49403355,"name":"Health sciences/Diseases/Neurological disorders/Parkinsons disease"},{"id":49403356,"name":"Health sciences/Biomarkers/Predictive markers"}],"tags":[],"updatedAt":"2025-10-06T16:02:00+00:00","versionOfRecord":{"articleIdentity":"rs-6658744","link":"https://doi.org/10.1038/s41531-025-01133-6","journal":{"identity":"npj-parkinsons-disease","isVorOnly":false,"title":"npj Parkinson's Disease"},"publishedOn":"2025-09-30 15:57:31","publishedOnDateReadable":"September 30th, 2025"},"versionCreatedAt":"2025-06-09 05:39:56","video":"","vorDoi":"10.1038/s41531-025-01133-6","vorDoiUrl":"https://doi.org/10.1038/s41531-025-01133-6","workflowStages":[]},"version":"v1","identity":"rs-6658744","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6658744","identity":"rs-6658744","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}