Objective: To identify a novel disease subtype of multiple system atrophy (MSA) by applying machine learning algorithms to a large cohort of autopsy-confirmed MSA.

Background: MSA shows various degrees of degeneration in striatonigral and olivopontocerebellar systems; however, its progression remains to be elucidated. Recently, probabilistic disease progression models have been developed, yet this approach has not been applied to MSA.

Method: We analyzed 167 autopsy-confirmed MSA patients. Neuronal loss in five regions (putamen, substantia nigra, pontine nucleus, inferior olivary nucleus, and cerebellar Purkinje cells) was semi-quantitatively assessed using a 4-point scale for input into an unsupervised learning algorithm (Subtype and Stage Inference – SuStaIn). Clinical information was collected from medical records. The immunoreactive area of α-synuclein immunohistochemistry (IHC) was quantitatively measured using a color deconvolution algorithm in the putamen, substantia nigra, pontine nucleus, inferior olivary nucleus, and cerebellar dentate nucleus. We examined correlations between SuStaIn subtyping results and both clinical information and pathological findings.

Results: The SuStaIn algorithm classified patients into three subtypes: subtype 1 (S1, n=86), subtype 2 (S2, n=44), and subtype 3 (S3, n=30). S1 showed initial neuronal loss in the striatonigral system, S2 in the olivopontocerebellar system, while S3 exhibited early neuronal loss in both systems. These patterns were validated by stained area analysis with S3 showing greater areas in the putamen compared to S2 (P = 0.02) and in the pontine and dentate nuclei compared to S1 (P = 0.001 and P = 0.016). Parkinsonism was predominant in 97% of S1 and 63% of S3, while cerebellar symptoms were predominant in 66% of S2 (P < 0.0001). S3 demonstrated shorter disease duration (6±2 years) compared to S1 and S2 (both 8±3 years; P < 0.05), more frequent rapid progression and early falls, and less frequent cognitive impairment. A positive correlation was observed between the SuStaIn stage and disease duration (r = 0.25, P = 0.0014).

Conclusion: This study identified a unique MSA subtype characterized by early involvement of both striatonigral and olivopontocerebellar pathology and poor prognosis. The SuStaIn algorithm shows promise for novel classifications of MSA subtypes.

« Back to 2025 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/novel-multiple-system-atrophy-subtypes-identified-using-unsupervised-machine-learning-algorithms/