NHL Pathological Image Classification Based on Hierarchical Local Information and GoogLeNet-Based Representations.

Area of Science:

• Pathology
• Medical Imaging
• Machine Learning

Background:

• Accurate classification of non-Hodgkin lymphomas (NHL) is challenging due to complexity.
• Distinguishing between mantle cell lymphoma (MCL), follicular lymphoma (FL), and chronic lymphocytic leukemia (CLL) is critical for diagnosis.

Purpose of the Study:

• To propose an effective classification model for three types of NHL pathological images.
• To improve the accuracy of NHL pathological image classification.

Main Methods:

• Utilized hematoxylin and eosin (H&E) stained NHL images.
• Extracted hand-crafted features from blue ratio (BR) and Lab color spaces.
• Employed a pre-trained Google Inception Net (GoogLeNet) for high-level feature learning.
• Implemented patch-level and image-level classification strategies with feature fusion.

Main Results:

• Achieved an overall accuracy of 0.991 on the IICBU Malignant Lymphoma Dataset.
• Obtained an area under the receiver operating characteristic curve (AUC) of 0.998.
• Demonstrated significantly improved classification performance.

Conclusions:

• The proposed model is a suitable approach for classifying NHL pathological images.
• The model's effectiveness was validated through rigorous experimentation.