Evaluation of the Use of Machine Learning Techniques to Classify Neurodegenerative PARKinsonian Syndromes (Artificial Intelligence)

Status

Completed

Phases

Unknown

Study type

Observational

Source

ClinicalTrials.gov

Registry ID

NCT05080296

Acronym

PARKIA

Enrollment

1664

Registered

2021-10-15

Start date

2021-12-20

Completion date

2024-09-01

Last updated

2026-06-25

For informational purposes only — not medical advice. Sourced from public registries and may not reflect the latest updates. Terms

Conditions

DaTSCAN SPECT Scans

Brief summary

The diagnosis of Parkinson's disease (PD) relies mainly on clinical observation of the patient, looking for the three characteristic symptoms and sometimes remains a real challenge. Machine Learning (ML) algorithms could help to diagnose PD early and differentiate idiopathic PD from atypical Parkinsonian syndromes. In this context, the work of Castillo-Barnes' team provided a set of imaging features based on morphological characteristics extracted from DaTSCAN® or Ioflupane (iodine-123-labeled radiopharmaceutical) single-photon emission computed tomography (SPECT) scans to discern healthy participants from participants with Parkinson's disease in a balanced set of SPECTs from the "Parkinson's Progression Markers Initiative" (PPMI) data base. The team of a study evaluated the classification performance of Parkinson's patients and normal controls when semi-quantitative indicators and shape features obtained on the dopamine transporter (DAT) by Ioflupane (123I-IP) single-photon emission computed tomography (SPECT) are combined as a machine learning (ML) feature. Artificial Intelligence (AI) based methods can improve diagnostic assessments. Several dopaminergic imaging studies using Artificial have reported accuracy of up to 90% for the diagnosis of PD. These automated approaches use machine learning methods, based on textural analyses, to (i) differentiate PD and healthy subjects, (ii) differentiate PD and vascular parkinsonism, and (iii) distinguish between different forms of atypical parkinsonism. A study conducted in 2 centers using a linear support vector machine (SVM) model discriminated patients with PD and healthy subjects with an accuracy of 82.5%.This performance is similar to visual assessment by nuclear physicians A linear SVM model based on voxel values of statistical parametric images was able to differentiate PD from vascular parkinsonism with an accuracy of 90.4%. The Nancy team has extensive experience in the detection of PD in SPECT and SPECT/CT scans with Ioflupane or DaTSCAN™

Interventions

None listed

Study design

Observational model

COHORT

Time perspective

RETROSPECTIVE

Eligibility

Sex/Gender

ALL

Age

18 Years to 85 Years

Inclusion criteria

* Patients who performed a DaTSCAN SPECT scan in the nuclear medicine department of the Nancy CHRU between 21/11/2011 and 01/09/2017. * Reviews that took place between 11/21/2011 and 9/1/2017 were repatriated from PACS to the processing consoles.

Design outcomes

Primary

Measure	Time frame	Description
Accuracy of the algorithm	2 months	Accuracy of the algorithm implemented for the new data in terms of predicting the type of atypical parkinsonian syndrome.

Secondary

Measure	Time frame	Description
Comparison of two networks	2 months	Comparison of the performance of the semi-supervised network with the supervised network, to recognize the importance of unlabeled data in learning
Analyze the robustness of the network	2 months	Analyze the robustness of the network to different data (data from different gamma camera models)

Countries

France

Outcome results