Improving Augmented Human Intelligence to Distinguish Burkitt Lymphoma From Diffuse Large B-Cell Lymphoma Cases

Area of Science:

• Computational pathology
• Artificial intelligence in medicine
• Hematopathology

Background:

• Distinguishing Burkitt lymphoma (BL) from diffuse large B-cell lymphoma (DLBCL) is critical for appropriate treatment.
• Hematopathologists face challenges in accurately classifying these similar-appearing B-cell lymphomas based on histologic images.

Purpose of the Study:

• To evaluate the effectiveness of a deep convolutional neural network (CNN) in assisting hematopathologists.
• To improve the diagnostic accuracy of differentiating BL from DLBCL using machine learning on histologic images.

Main Methods:

• A deep, densely connected CNN was trained and applied to a dataset of 10,818 histologic images from BL and DLBCL cases.
• Various network parameters, including image augmentation and network depth, were optimized to achieve the best performance.

Main Results:

• The best performing CNN achieved 94% accuracy in correctly classifying BL and DLBCL cases (17 out of 18).
• The optimal network utilized all training images, specific image augmentation techniques, and a depth of 22 layers.
• Receiver operating characteristic curve analysis showed an area under the curve of 0.92 for both lymphoma types.

Conclusions:

• Deep convolutional neural networks show significant potential as augmented intelligence tools for pathologists.
• CNNs can effectively assist in differentiating challenging cases of Burkitt lymphoma and diffuse large B-cell lymphoma.