Classification of Aβ State From Brain Amyloid PET Images Using Machine Learning Algorithm.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Analyzing brain amyloid positron emission tomography (PET) images for β-amyloid (Aβ) deposition in Alzheimer's disease is time-consuming and subject to interpreter variability.
• Objective classification is needed to standardize the assessment of Aβ status from PET imaging.

Purpose of the Study:

• To develop and evaluate a machine learning model using a convolutional neural network (CNN) for objective classification of Aβ positive and Aβ negative status from brain amyloid PET images.
• To assess the clinical potential of the CNN model in screening amyloid PET scans.

Main Methods:

• Utilized 7,344 brain amyloid PET images from 144 subjects, administered with 18F-florbetaben PET.
• Classified images into Aβ positive and Aβ negative states based on physician-assessed brain amyloid plaque load scores (BAPL).
• Trained a CNN algorithm on batches of 51 images per subject directory.

Main Results:

• The CNN model achieved an average accuracy of 95.00±0.02% in classifying Aβ positivity and negativity on the test dataset.
• Demonstrated high performance with a sensitivity of 96.00±0.02% and specificity of 94.00±0.02%.
• Achieved an area under the curve (AUC) of 87.00±0.03 for binary classification.

Conclusions:

• The developed CNN model shows significant potential for clinical application in screening amyloid PET images.
• The model provides an objective and efficient method for assessing Aβ deposition, potentially improving Alzheimer's disease diagnosis.
• Further clinical validation could establish this AI tool as a valuable aid for physicians in interpreting PET scans.