Deep Learning-Based Image Classification in Differentiating Tufted Astrocytes, Astrocytic Plaques, and Neuritic...

Area of Science:

• Neuropathology
• Artificial Intelligence
• Medical Imaging

Background:

• Tauopathies, including Progressive Supranuclear Palsy (PSP), Corticobasal Degeneration (CBD), and Alzheimer's Disease (AD), are characterized by distinct tau protein aggregates.
• Accurate differentiation of these lesions is crucial for diagnosis and understanding disease mechanisms.
• Current diagnostic methods can be labor-intensive and require specialized expertise.

Purpose of the Study:

• To develop and validate a deep learning-based image classification model.
• To differentiate between tufted astrocytes (TA), astrocytic plaques (AP), and neuritic plaques (NP) using phospho-tau immunohistochemistry images.
• To assess the model's utility in assisting the differential diagnosis of tauopathies.

Main Methods:

• Utilized Google AutoML for automated deep learning model creation.
• Trained the model on 1332 phospho-tau stained tissue section images from PSP, CBD, and AD patients.
• Validated the model using 100 independent test images.

Main Results:

• Achieved high precision and recall in cross-validation: 100%/98.0% for TA, 98.5%/98.5% for AP, and 98.0%/100% for NP.
• In the test set, the model correctly identified all TA and NP images.
• Eleven AP images were misclassified as TA or NP, demonstrating high diagnostic accuracy.

Conclusions:

• Deep learning image classification effectively differentiates key tau lesions.
• The developed model shows significant potential as an assistive tool for diagnosing tauopathies.
• AI-driven analysis can enhance the accuracy and efficiency of neuropathological assessments.