T cell classification

T cell classification applies convolutional neural networks (CNNs) to classify T cell activation states from autofluorescence intensity images, using NAD(P)H-derived metabolic signals to assess antigen-induced T cell activity for immunotherapy research.

Key Features:

• Non-Destructive Imaging: Utilizes autofluorescence intensity imaging of endogenous metabolic co-enzymes such as NAD(P)H to distinguish T cell activity states without exogenous labels.
• Machine Learning Integration: Evaluates a range of classifiers from traditional models using pre-extracted image features to CNNs for classifying T cell activities across human donors.
• Advanced CNN Utilization: Adapts pre-trained CNNs originally trained on non-biological images, achieving substantially improved performance compared to traditional models and CNNs trained solely on autofluorescence images.
• Data Handling and Visualization: Processes a dataset of 8,260 cropped single-cell images from six donors and applies dimensionality-reduction techniques to visualize image representations and interpret CNN performance.

Scientific Applications:

• Immunotherapy research: Classifying antigen-induced T cell activation to assess functional capacity, evaluate treatment efficacy, and inform therapeutic interventions.

Methodology:

Processes autofluorescence intensity images derived from metabolic co-enzymes (e.g., NAD(P)H); evaluates various classifiers including traditional feature-based models and CNNs; adapts CNNs pre-trained on non-biological image datasets; analyzes 8,260 cropped single-cell images from six donors and employs dimensionality-reduction techniques for visualization.