Machine Learning Based on Morphological Features Enables Classification of Primary Intestinal T-Cell Lymphomas

Area of Science:

• Oncology
• Computational Pathology
• Digital Pathology

Background:

• Primary intestinal T-cell lymphomas (PITLs) pose diagnostic challenges, particularly differentiating monomorphic epitheliotropic intestinal T-cell lymphoma (MEITL) from intestinal T-cell lymphoma, not otherwise specified (ITCL-NOS).
• Accurate classification is crucial for patient prognosis and treatment strategies.

Purpose of the Study:

• To evaluate the feasibility of machine learning (ML) utilizing morphological features for classifying PITL subtypes.
• To develop an accurate and interpretable ML model for differentiating MEITL from ITCL-NOS.

Main Methods:

• A dataset of 40 whole-slide images (WSIs) from surgically resected PITL cases was used.
• A deep neural network segmented lymphocyte nuclei, and quantitative nuclear morphometrics were extracted.
• An XGBoost classifier was trained on these morphometric features to distinguish between MEITL and ITCL-NOS.

Main Results:

• The deep neural network achieved an average precision of 0.881 for cell segmentation.
• The XGBoost model demonstrated high classification performance with an AUC of 0.966 for differentiating MEITL from ITCL-NOS.
• The study successfully reduced high-dimensional image data into interpretable features.

Conclusions:

• Machine learning, based on quantitative nuclear morphometrics, offers an accurate and interpretable method for classifying challenging PITL subtypes.
• This approach can aid pathologists in diagnosing MEITL versus ITCL-NOS.
• Further research into nuclear morphometric features may reveal insights into cellular phenotype-disease status relationships.