Histological Image Classification Between Follicular Lymphoma and Reactive Lymphoid Tissue Using Deep Learning and...

Area of Science:

• Computational pathology
• Artificial intelligence in oncology
• Digital pathology

Background:

• Distinguishing between benign and malignant lymph node biopsies is critical for pathologists.
• Differentiating follicular lymphoma from reactive lymphoid tissue presents diagnostic challenges.
• Hematoxylin and eosin (H&E) staining is standard for histopathological analysis.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN) for classifying follicular lymphoma.
• To compare CNN performance against reactive lymphoid tissue using H&E-stained lymph node biopsies.
• To assess the interpretability of the AI model using explainable AI (XAI) methods.

Main Methods:

• A ResNet-based CNN was designed and trained on a dataset of 221 cases (177 follicular lymphoma, 44 reactive lymphoid tissue).
• The dataset comprised over 1.5 million image patches, partitioned into training, validation, and testing sets.
• Explainable AI techniques, including grad-CAM, image LIME, and occlusion sensitivity, were employed for model interpretability.

Main Results:

• The CNN achieved high accuracy (99.80%) in differentiating follicular lymphoma from reactive lymphoid tissue on the testing set.
• Excellent performance metrics were observed: precision (99.8%), recall (99.8%), specificity (99.7%), and F1 score (99.9%).
• Patient-level validation confirmed the model's robust classification performance, minimizing information leakage.

Conclusions:

• Task-specific artificial intelligence (AI) demonstrates efficacy in the differential diagnosis of follicular lymphoma.
• The trained ResNet CNN shows potential for transfer learning in broader lymphoma diagnostic applications.
• AI tools can assist pathologists, but operate within defined constraints for specific diagnostic tasks.