Multiclass Classification for the Differential Diagnosis on the ADHD Subtypes Using Recursive Feature Elimination and...

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Accurate diagnosis of brain diseases like ADHD is crucial.
• Neuroimaging data analysis is a key area in neuroscience and clinical research.
• Developing effective classification methods for neuroimaging is an ongoing challenge.

Purpose of the Study:

• To evaluate the performance of a hierarchical extreme learning machine (H-ELM) for multiclass classification of neuroimaging data.
• To compare H-ELM with support vector machine (SVM) and basic extreme learning machine (ELM) for classifying ADHD subtypes.
• To identify key neuroimaging features indicative of ADHD.

Main Methods:

• Utilized 159 structural MRI images from the ADHD-200 dataset, including typically developing children, ADHD-inattentive, and ADHD-combined subtypes.
• Employed a hierarchical extreme learning machine (H-ELM) classifier for multiclass classification.
• Applied SVM-based recursive feature elimination (RFE-SVM) for feature selection.

Main Results:

• Achieved a classification accuracy of 60.78% using RFE-SVM for feature selection.
• Demonstrated that the combination of RFE-SVM and H-ELM effectively enhances multiclass classification accuracy for neuroimaging data.
• Identified the surface area of the superior frontal lobe, and cortical thickness, volume, and mean surface area of the whole cortex as critical features.

Conclusions:

• The RFE-SVM feature selection method combined with H-ELM is effective for high-accuracy multiclass classification of structural neuroimaging data.
• This approach shows promise for the diagnosis and subtyping of conditions like ADHD.
• Specific cortical features are significant biomarkers for differentiating ADHD subtypes.