DTI based diagnostic prediction of a disease via pattern classification

Area of Science:

• Neuroimaging
• Machine Learning
• Medical Diagnostics

Background:

• Diffusion Tensor Imaging (DTI) provides insights into white matter integrity.
• Developing objective biomarkers for neurological and psychiatric disorders is crucial.
• Current diagnostic methods can be subjective and require improvement.

Purpose of the Study:

• To create abnormality classifiers using DTI data for schizophrenia (SCZ) and autism spectrum disorder (ASD).
• To evaluate the diagnostic and prognostic potential of classifier-generated abnormality scores.

Main Methods:

• Non-linear support vector machine (SVM) pattern classifier trained on DTI-derived features.
• Elastic registration of DT images followed by feature extraction (average anisotropy and diffusivity).
• Mutual information-based feature selection and sequential elimination.

Main Results:

• Achieved 90.62% accuracy for classifying SCZ patients versus controls.
• Obtained 89.58% classification accuracy for individuals with ASD versus controls.
• Abnormality scores effectively separated patient groups from control groups.

Conclusions:

• DTI-based abnormality classifiers demonstrate high accuracy in distinguishing SCZ and ASD.
• Classifier-generated abnormality scores show promise as diagnostic and prognostic markers.
• This approach offers a quantitative method to assess the degree of pathology.