Objective: Predict PD patient’s clinical scores from brain activity. Inform on which assessments capture true changes in cortical dynamics caused by PD and explain those changes.

Background: PD patients suffer from a series of non-motor complications [1], and there is a need to identify which are most affected by the underlying cortical changes. While neuroimaging has correlated some symptoms to brain changes [2], it has not been successful in measuring brain activity and predicting the symptoms. We present a novel and precise way of predicting symptoms from cortical dynamics alone.

Method: 51 PD patients without dementia but with a varied clinical phenotype performed a battery of neuropsychological and neurological assessments, fMRI, and a sensorimotor robotic task to induce clinically relevant hallucinations [3]. We recovered brain activity from fMRI using ICA back-projection and created probabilistic representations of each patient’s brain-state dynamics using Bayesian methods (Hidden Markov Modelling) [4]. To predict each patient’s symptoms solely from these representations we resorted to Kernel Ridge Regression [5].

Results: Our models excel at predicting executive functioning, mental flexibility and interference-inhibition, lexical retrieval and production, as well as visuo-constructional ability and memory. Regarding hallucinations, questionnaires that captured their frequency and severity were well predicted, however, the models could not predict stratifications into structured-visual, minor or no hallucinations, suggesting that such approach does not sufficiently reflect the underlying changes in neural dynamics. Crucially, the sensitivity to an hallucination-inducing robotic task – previously linked to proneness to clinical hallucinations [6] – is significantly well predicted. In addition, non-motor experiences (MDS-UPDRS-I) and impulsivity are also well predicted. Finally, our models show the clinician’s assessment of neuropsychiatric symptoms to be more reflective of neural changes than the patient’s subjective assessment.

Conclusion: We developed an approach through which brain-dynamics predict PD patient’s phenotype across multiple clinical domains. With this we reveal the assessments that best capture the underlying neural dynamics changes. Future applications of this approach could see a new patient’s symptoms predicted from comparing their fMRI brain dynamics model to a pre-modelled large dataset of PD brain data.

Figure1: Schematic representation of the methods

Figure 2: Predicted symptoms from brain activity

References: 1. Postuma, R. B. & Berg, D. Advances in markers of prodromal Parkinson disease. Nat. Rev. Neurol. 12, 622–634 (2016).
2. Collerton, D. et al. Understanding visual hallucinations: A new synthesis. Neurosci. Biobehav. Rev. 150, 105208 (2023).
3. Bernasconi, F. et al. Neuroscience robotics for controlled induction and real-time assessment of hallucinations. Nat. Protoc. (2022) doi:10.1038/s41596-022-00737-z.
4. Vidaurre, D., Smith, S. M. & Woolrich, M. W. Brain network dynamics are hierarchically organized in time. Proc. Natl. Acad. Sci. 114, 12827–12832 (2017).
5. Vidaurre, D., Llera, A., Smith, S. M. & Woolrich, M. W. Behavioural relevance of spontaneous, transient brain network interactions in fMRI. NeuroImage 229, 117713 (2021).
6. Bernasconi, F. et al. Robot-induced hallucinations in Parkinson’s disease depend on altered sensorimotor processing in fronto-temporal network. Sci. Transl. Med. 13, eabc8362 (2021).

« Back to 2025 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/predicting-the-cognitive-psychiatric-phenotype-of-parkinsons-disease/