[Paper Review] Automated Diagnosis of Lymphoma with Digital Pathology Images Using Deep Learning

Hanadi El Achi, Т. А. Белоусова+7 more|PubMed|Oct 30, 2018

AI in cancer detection48 citations

TL;DR

The paper presents a CNN-based system to diagnose lymphoma across four categories from digital pathology images, achieving high image-by-image accuracy and demonstrating concept feasibility in workflow augmentation.

ABSTRACT

Recent studies have shown promising results in using Deep Learning to detect malignancy in whole slide imaging, however, they were limited to just predicting a positive or negative finding for a specific neoplasm. We attempted to use Deep Learning with a convolutional neural network (CNN) algorithm to build a lymphoma diagnostic model for four diagnostic categories: (1) benign lymph node, (2) diffuse large B-cell lymphoma, (3) Burkitt lymphoma, and (4) small lymphocytic lymphoma. Our software was written in Python language. We obtained digital whole-slide images of Hematoxylin and Eosin stained slides of 128 cases including 32 cases for each diagnostic category. Four sets of 5 representative images, 40x40 pixels in dimension, were taken for each case. A total of 2,560 images were obtained from which 1,856 were used for training, 464 for validation, and 240 for testing. For each test set of 5 images, the predicted diagnosis was combined from the prediction of five images. The test results showed excellent diagnostic accuracy at 95% for image-by-image prediction and at 100% for set-by-set prediction. This preliminary study provided a proof of concept for incorporating automated lymphoma diagnostic screen into future pathology work-flow to augment the pathologists' productivity.

Motivation & Objective

• Motivate and assess deep learning for automated lymphoma diagnosis on whole slide imaging.
• Develop a convolutional neural network approach to classify four diagnostic categories.
• Evaluate performance on a labeled dataset of curated image patches extracted from WSIs.
• Provide proof-of-concept for integrating automated screening into pathology workflows.

Proposed method

• Use a CNN-based model implemented in Python to process image patches.
• Construct a dataset of 128 cases with four diagnostic categories (benign lymph node, DLBCL, Burkitt lymphoma, small lymphocytic lymphoma).
• Extract four sets of representative 40x40 px images (5 per case) totaling 2,560 patches.
• Split data into training (1,856), validation (464), and testing (240) patches.

Experimental results

Research questions

• RQ1 Can a CNN-based approach differentiate four lymphoid diagnostic categories from 40x40 pixel patches?
• RQ2 What is the diagnostic performance at the patch level versus the case/set level?
• RQ3 Is automated patch-based screening feasible as an augmentative tool in pathology workflows?

Key findings

• Image-by-image prediction accuracy of 95% on the test set.
• Set-by-set (5-image) prediction accuracy of 10% on the test set.
• Dataset comprises 128 cases with balanced category representation (32 per category).
• Demonstrates proof-of-concept for integrating automated lymphoma diagnosis into pathology workflows.
• The approach provides initial evidence that deep learning can assist pathologists in screening for lymphoma.