Classifying T cell activity in autofluorescence intensity images with convolutional neural networks

Area of Science:

• Immunology
• Computational Biology
• Biotechnology

Background:

• T cells are crucial for immunotherapy efficacy.
• Assessing T cell activation is vital for therapeutic success.
• Autofluorescence imaging detects metabolic changes (NAD(P)H) to indicate T cell activity non-destructively.

Purpose of the Study:

• To develop and evaluate machine learning methods for classifying T cell activity.
• To leverage NAD(P)H autofluorescence imaging for objective T cell state assessment.
• To improve the computational analysis of T cell functional states.

Main Methods:

• Utilized 8260 single-cell images from six human donors.
• Compared traditional machine learning models with convolutional neural networks (CNNs).
• Evaluated CNNs pre-trained on general non-biological images against those trained solely on T cell data.

Main Results:

• Machine learning, particularly adapted pre-trained CNNs, accurately classified T cell activity from NAD(P)H images.
• Pre-trained CNNs significantly outperformed traditional models and simple CNNs trained on autofluorescence data alone.
• Dimensionality reduction visualization offered insights into CNNs' superior performance.

Conclusions:

• Adapted pre-trained CNNs offer a robust computational approach for classifying T cell activity using autofluorescence.
• This method enhances the objective assessment of T cell function in immunotherapy research.
• The developed software is publicly available for broader application.