A deep learning diagnostic platform for diffuse large B-cell lymphoma with high accuracy across multiple hospitals.

Area of Science:

• Pathology
• Hematology
• Artificial Intelligence
• Deep Learning

Background:

• Diagnostic histopathology is crucial for hematopoietic malignancies, but requires intensive labor and near-perfect accuracy.
• Current artificial intelligence (AI) in pathology reduces workload but lacks clinical-grade diagnostic accuracy.
• AI model development typically requires large datasets and robust handling of sample variations.

Purpose of the Study:

• To establish a highly accurate deep learning platform for classifying hematopoietic malignancy pathology images.
• To demonstrate the efficacy of AI models using smaller datasets.
• To achieve clinically practical diagnostic accuracy for diffuse large B-cell lymphoma (DLBCL) and other hematopoietic malignancies.

Main Methods:

• Developed a deep learning platform with multiple convolutional neural networks.
• Trained and analyzed AI models on human diffuse large B-cell lymphoma (DLBCL) and non-DLBCL pathology images from three separate hospitals.
• Assessed AI model performance considering technical variability from slide preparation and image collection.

Main Results:

• Achieved diagnostic accuracy close to 100% across three hospitals (100% for A, 99.71% for B, 100% for C).
• Technical variability reduced cross-hospital performance, but 100% accuracy was restored after elimination.
• Demonstrated the platform's ability to classify pathology images effectively using smaller datasets.

Conclusions:

• The developed deep learning platform offers a clinically practical solution for diagnosing DLBCL.
• AI models can achieve high diagnostic accuracy for hematopoietic malignancies, even with limited data.
• This approach has the potential to significantly improve the diagnosis of human hematopoietic malignancies.