Gene-related Parkinson's disease diagnosis via feature-based multi-branch octave convolution network.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Parkinson's disease (PD) is a common neurodegenerative disorder diagnosed clinically, often after irreversible progression.
• Gene mutations linked to PD can affect asymptomatic individuals, necessitating early detection methods.
• Magnetic Resonance Imaging (MRI) offers rich brain tissue information but faces challenges with limited gene-related PD cohorts.

Purpose of the Study:

• To develop an effective diagnostic tool for gene-related Parkinson's disease using MRI data.
• To address the challenge of limited sample sizes in gene-related PD cohorts.
• To improve the accuracy of early PD diagnosis in individuals with genetic predispositions.

Main Methods:

• A joint learning framework, the feature-based multi-branch octave convolution network (FMOCNN), was developed.
• The framework incorporates cardinality constrained sample-feature selection (CCSFS) for identifying discriminative samples and features.
• A multi-branch octave convolution neural network (MBOCNN) was utilized for joint training of multiple feature inputs, leveraging high/low-frequency learning.

Main Results:

• The FMOCNN framework demonstrated promising classification performance on the Parkinson's Progression Markers Initiative (PPMI) dataset.
• The proposed method outperformed existing algorithms in diagnosing gene-related PD.
• Effective sample-feature selection and deep feature representation were achieved.

Conclusions:

• The FMOCNN framework offers a robust approach for diagnosing gene-related Parkinson's disease using MRI.
• The method shows potential for early detection and intervention in individuals at genetic risk for PD.
• This AI-driven approach can aid in distinguishing gene-related PD, even in asymptomatic carriers.