Automated Diagnosis of Lymphoma with Digital Pathology Images Using Deep Learning.

Area of Science:

• Computational pathology
• Digital pathology
• Machine learning in oncology

Background:

• Whole slide imaging (WSI) enables detailed analysis of tissue samples.
• Deep learning (DL) shows promise for malignancy detection in WSI.
• Current DL applications are often limited to binary classifications.

Purpose of the Study:

• To develop a DL-based diagnostic model for lymphoma.
• To classify four distinct categories: benign lymph node, diffuse large B-cell lymphoma, Burkitt lymphoma, and small lymphocytic lymphoma.
• To assess the diagnostic accuracy of the CNN model.

Main Methods:

• Utilized a convolutional neural network (CNN) algorithm implemented in Python.
• Trained the model on 2,560 digital WSI patches from 128 cases (32 per category).
• Images were H&E stained; 1,856 for training, 464 for validation, and 240 for testing.

Main Results:

• Achieved 95% accuracy in image-by-image predictions.
• Reached 100% accuracy in set-by-set predictions (combining predictions from 5 images).
• Demonstrated high diagnostic performance for the defined lymphoma categories.

Conclusions:

• The study provides a proof of concept for automated lymphoma diagnosis using DL.
• The developed model shows potential to enhance pathologist productivity in pathology workflows.
• Automated screening could improve efficiency and accuracy in lymphoma diagnostics.