Multiple instance learning for classification of dementia in brain MRI.

Area of Science:

• Neurology
• Computer Science
• Medical Imaging

Background:

• Machine learning aids neurological disease diagnosis, including dementia.
• Current methods analyzing image patches overlook patch relationships and face labeling ambiguity.
• Not all image patches from dementia patients are definitively disease-indicative.

Purpose of the Study:

• To develop a machine learning technique addressing patch labeling ambiguity in disease detection.
• To incorporate relationships among image patches for improved classification.
• To apply and evaluate the method for Alzheimer's disease (AD) detection.

Main Methods:

• Utilized a multiple instance learning method for patch label assignment.
• Constructed a graph for each image to capture relationships among patches.
• Applied the approach to baseline Magnetic Resonance Imaging (MRI) data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) study.

Main Results:

• Achieved 88.8% classification accuracy between Alzheimer's disease (AD) patients and healthy controls.
• Attained 69.6% accuracy in distinguishing between stable and progressive Mild Cognitive Impairment (MCI).
• Demonstrated performance comparable to or exceeding state-of-the-art classification methods.

Conclusions:

• The proposed multiple instance learning and graph-based method effectively addresses patch labeling challenges in neurological disease diagnosis.
• This approach enhances classification accuracy for Alzheimer's disease and Mild Cognitive Impairment progression.
• The findings suggest a promising direction for leveraging image patch relationships in neuroimaging analysis.