Medical-AI trial: classifying parkinsonian syndromes from brain imaging (DaTSCAN SPECT) with machine learning — aiding a hard diagnosis (NCT05080296)

Parkinson disease (PD) is a neurological disease in which loss of dopamine nerves in the brain causes tremor, slowed movement, and stiffness. Diagnosis is made mainly by a clinician observing symptoms, but early disease, and telling PD apart from look-alike atypical parkinsonian syndromes, can be hard.

A useful clue is DaTSCAN, a nuclear-medicine imaging test (SPECT) that visualizes the distribution of dopamine transporters in the brain.

This study evaluates, as an observational study, whether machine learning (ML) applied to DaTSCAN images can classify. Per the registry summary, because PD diagnosis relies mainly on clinical observation and is sometimes challenging, ML is framed as potentially helping early diagnosis and differentiating idiopathic (true) PD from atypical parkinsonian syndromes.

Prior research provided a set of imaging features extracted from DaTSCAN (ioflupane, an iodine-123-labeled radiopharmaceutical) SPECT scans, used to discern healthy participants from those with PD. The primary outcome is algorithm accuracy, and the size is 1,664, at the completed stage.

Diagnosis by symptom observation varies with experience and disease stage, and confidence is hard to reach in early or hard-to-differentiate cases. If ML can aid classification from objective, quantified image features, it could support diagnostic consistency and early judgment. Distinguishing true PD from atypical syndromes, which differ in treatment response and prognosis, is clinically important.

That said, this study evaluates algorithm accuracy; it does not establish replacement of imaging diagnosis or specialist judgment.