Deep Learning-Based Algorithm for Automatic Quantification of Nigrosome-1 and Parkinsonism Classification Using...

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medicine
• Neurodegenerative Diseases

Background:

• Nigrosome-1 abnormality on susceptibility map-weighted imaging (SMwI) is a potential biomarker for neurodegenerative parkinsonism.
• The diagnostic performance of deep learning models for simultaneous detection and quantification of nigrosome-1 abnormality remains unexplored.

Purpose of the Study:

• To develop and validate a deep learning-based automatic quantification model for nigral hyperintensity.
• To develop and validate a deep learning-based classification algorithm for neurodegenerative parkinsonism.

Main Methods:

• Retrospective collection of 450 participants (210 with idiopathic Parkinson disease [IPD], 240 controls) for training and 237 participants (168 IPD, 58 essential tremor, 11 drug-induced parkinsonism) for validation.
• SMwI data reconstructed from multiecho gradient echo.
• Diagnostic performance assessed using Heuron NI (quantification) and Heuron IPD (classification) models, with IPD diagnosis reference standard from DaTscan PET.

Main Results:

• Heuron NI achieved AUCs of 0.915 (left) and 0.928 (right) for nigral hyperintensity quantification.
• Heuron IPD achieved AUCs of 0.967 (left) and 0.976 (right) for classifying nigral hyperintensity abnormality.
• Smaller nigral hyperintensity volume was observed in patients with higher Hoehn and Yahr (H&Y) scores (≥3) compared to those with lower scores (1-2.5).

Conclusions:

• The developed deep learning model offers rapid and accurate automatic quantification of nigral hyperintensity.
• This facilitates idiopathic Parkinson disease diagnosis, prediction of symptom severity, and patient stratification for personalized therapy.
• Further validation across diverse clinical settings is recommended.