Machine Learning for Discriminating Microcytic Hypochromic Anemia Based on Erythrocyte Parameters.

Area of Science:

• Hematology
• Machine Learning
• Medical Diagnostics

Background:

• Thalassemia trait (TT) and iron deficiency anemia (IDA) are common causes of microcytic hypochromic anemia (MHA).
• Current diagnostic methods for differentiating TT and IDA have limitations.
• Accurate differentiation is crucial for appropriate patient management.

Purpose of the Study:

• To develop and evaluate machine learning (ML) algorithms for distinguishing between TT and IDA.
• To utilize erythrocyte parameters for accurate MHA classification.
• To identify key erythrocyte parameters for discriminant modeling.

Main Methods:

• Retrospective analysis of 193 MHA subjects (98 TT, 95 IDA).
• Data split into training (60%), validation (20%), and testing (20%) sets.
• Five ML algorithms (Random Forest, XGBoost, logistic regression, AdaBoost, LightGBM) applied to erythrocyte parameters from an automated hematology analyzer.
• Performance assessed using sensitivity, specificity, accuracy, AUC, NPV, PPV, F1 score, and Kappa coefficients.

Main Results:

• Random Forest and logistic regression models demonstrated excellent discriminant performance in the test set.
• Random Forest achieved an AUC of 0.977, sensitivity of 0.928, specificity of 0.953, and accuracy of 0.940.
• Logistic regression achieved an AUC of 0.978, sensitivity of 0.879, specificity of 0.979, and accuracy of 0.928.
• Eight vital erythrocyte parameters were identified: RBC, RDW, MCV, MCHC, RDWSD, HGB, MAF, and LHD.

Conclusions:

• Successfully developed ML-based discriminant models for rapid MHA identification.
• The models effectively differentiate between TT and IDA using erythrocyte parameters.
• This approach can assist in timely diagnosis and preventive measures for patients.