A Machine Learning Tool Using Digital Microscopy (Morphogo) for the Identification of Abnormal Lymphocytes in the...

Area of Science:

• Computational pathology
• Hematology
• Artificial Intelligence

Background:

• Manual bone marrow smear analysis is labor-intensive and prone to interobserver variability.
• Distinguishing abnormal from normal lymphocytes in bone marrow morphology is challenging.
• Digital pathology and machine learning offer potential for automated, accurate cell classification.

Purpose of the Study:

• To develop and validate a machine learning system (Morphogo) for discriminating abnormal from normal lymphocytes in digital bone marrow images.
• To assess the system's accuracy in identifying lymphoma cells in bone marrow samples.

Main Methods:

• Developed a machine learning algorithm incorporating color, texture, and geometrical features for lymphocyte image analysis.
• Trained the algorithm on 15,353 bone marrow cell images from 43 confirmed lymphoma cases.
• Validated the system on images from 10 additional cases, comparing results with pathologist diagnoses.

Main Results:

• The Morphogo system achieved a 99.04% positive predictive value for identifying reactive/normal versus abnormal lymphoid cells.
• The AI-based system demonstrated high accuracy in classifying lymphocyte categories based on digital image analysis.

Conclusions:

• Digital pathology combined with machine learning provides a feasible approach for automated lymphoma cell diagnosis in bone marrow.
• The Morphogo system shows promise as a robust and efficient screening tool for hematological diseases.
• Future improvements may enable further subclassification of specific lymphoma subtypes.