Tumor-Infiltrating Lymphocyte Recognition in Primary Melanoma by Deep Learning Convolutional Neural Network

Area of Science:

• Digital pathology
• Artificial intelligence in oncology
• Melanoma research

Background:

• Tumor-infiltrating lymphocytes (TILs) are crucial prognostic indicators in primary melanoma (PM).
• Standardized assessment of TILs is challenging with conventional methods.
• AI in digital pathology offers potential for automated and standardized TIL evaluation.

Purpose of the Study:

• To develop and validate a novel convolution neural network (CNN) for automated TIL assessment in PM.
• To establish an AI-based TIL density index (AI-TIL) for prognostic evaluation.
• To compare the performance of AI-TIL with conventional TIL assessment methods.

Main Methods:

• A CNN was trained and validated on 307 primary melanoma whole slide images (WSIs).
• The CNN classified tumor patches based on the presence or absence of TILs, generating an AI-TIL index.
• Performance was evaluated on an independent testing set for specificity and sensitivity.

Main Results:

• The CNN demonstrated high performance, achieving 100% specificity and sensitivity in TIL recognition on the testing set.
• The AI-TIL index showed strong correlation with conventional TIL evaluation and patient clinical outcomes.
• The AI-TIL index was identified as an independent prognostic marker for favorable outcomes in PM.

Conclusions:

• A fully automated AI-TIL assessment is superior to conventional methods for differentiating PM clinical outcomes.
• The developed AI-TIL index serves as a robust and independent prognostic marker.
• Further research is needed to create user-friendly tools for clinical integration of AI-TIL assessment.