Objective: To systematically review the current literature on the application of artificial intelligence (AI) for imaging biomarkers in Parkinson’s disease (PD) and to identify translational gaps.

Background: A long asymptomatic and prodromal phase, as well as the considerable variation in symptoms and the symptomatic overlap with atypical parkinsonian syndromes (APS) make diagnosis, prognosis and disease monitoring of PD a challenge for the clinician. While imaging biomarkers have delivered much progress in this regard, artificial intelligence is a promising tool to further improve PD diagnosis, prognosis, and disease monitoring using imaging biomarkers.

Method: In this systematic review, we analyzed 156 out of 453 original English-language papers obtained from a PubMed search on the applications of AI to neuroimaging in PD after applying pre-specified inclusion criteria. We categorized the included papers into diagnosis, prognosis, and monitoring with further subcategories. Subsequently, we defined and applied a set of minimum quality criteria (MQC) to assess their methodological validity.

Results: Preliminary results show that most papers (n=142) were assigned to the diagnosis category. Here, the number of papers investigating binary classifications of PD against healthy controls (HC) was most prevalent, followed by differential diagnosis between (a particular) APS and PD. In the relatively sparsely populated prognosis category (n=15), biomarkers anticipated disease or symptom severity, including the development of dementia. Finally, seven papers in the monitoring category described the potential of AI in optimizing deep brain stimulation and drug efficacy. Based on the subset of papers assessed thus far, we anticipate that ~30% do not fulfill our MQC pertaining to reproducibility and interpretability.

Conclusion: In conclusion, few projects target the multiclass problem of differentiating various APS forms and PD, as well as disease prognosis and monitoring, while the classification of PD and HC has been sufficiently exploited. Future work should, therefore, focus on the less exploited but clinically more relevant categories. To do so, our proposed MQC may assist in the development of AI algorithms and contribute to bridging the translational gaps between engineers and clinicians and therefore boost the integration of AI tools in clinical practice.

« Back to 2023 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/artificial-intelligence-for-the-improvement-of-imaging-biomarkers-in-parkinsons-disease/