Deep learning classification of reading disability with regional brain volume features.

Area of Science:

• Neuroscience
• Machine Learning
• Developmental Disorders

Background:

• Developmental reading disability presents significant heterogeneity in mechanisms and phenotypes.
• Accurate neuroimaging-based classification is challenging due to data complexity and modest sample sizes.

Purpose of the Study:

• To develop and validate a deep learning model for classifying developmental reading disability using neuroimaging data.
• To identify specific brain regions associated with reading disability and control groups.

Main Methods:

• An unsupervised autoencoder reduced neuroimaging data dimensionality.
• A supervised convolutional neural network classified reading disability cases and controls.
• Noise-addition technique identified key brain regions for classification.

Main Results:

• The combined deep learning model achieved 77% accuracy, 0.75 precision, and 0.78 recall.
• Key regions for reading disability classification included the superior temporal sulcus, dorsal cingulate, and lateral occipital cortex.
• The supramarginal gyrus, orbitofrontal, and medial occipital cortex were important for control classification.

Conclusions:

• Deep learning offers an effective approach for neuroimaging-based classification of reading disability.
• The model identified distinct neural correlates differentiating reading disability from controls.
• This method provides insights into brain regions specifically affected in reading disability.