Enhancing Thalassemia Gene Carrier Identification in Non-anemic Populations Using AI Erythrocyte Morphology Analysis and Machine Learning

Title : Enhancing Thalassemia Gene Carrier Identification in Non-anemic Populations Using AI Erythrocyte Morphology Analysis and Machine Learning

Abstract:

Background:
Non-anemic thalassemia trait (TT) accounts for over 60% of TT cases in South China, yet remains challenging to identify due to absent anemia and the high cost of gold-standard genetic testing—especially in resource-limited regions. Traditional erythrocyte morphology analysis is subjective and time-consuming, limiting its use for screening.
Case presentation:
We conducted the first study leveraging AI for quantitative abnormal erythrocyte analysis to identify non-anemic TT carriers. Digital morphological data from 76 non-anemic TT carriers (69.7% α-TT) and 97 healthy controls were collected using the AI-powered Mindray MC-100i analyzer. Machine learning (ML) models were trained and validated, with external validation in 54 non-anemic TT carriers and 97 controls.
Results:
A Random Forest-based ML model (TT@Normal) was developed, with target cells, microcytes, and teardrop cells as the top three predictive features. TT@Normal exhibited exceptional performance: training/validation set metrics (AUC, sensitivity, specificity) all >94%, and external validation achieving AUC=97.65%, sensitivity=92.59%, and specificity=93.81%. It outperformed four conventional indexes (MI, EFI, GKI, RDW) in discriminative power. TT@Normal is freely accessible as an online tool (URL provided), enabling rapid screening even with rough estimation of abnormal erythrocyte percentages when automatic analyzers are unavailable.
Conclusion:
TT@Normal is the first AI/ML-driven tool for non-anemic TT carrier identification, addressing an unmet clinical need. Its high accuracy, reliance on routine erythrocyte morphology, and user-friendly online access make it a practical screening solution—particularly valuable for underdeveloped regions. Elevated target cells, microcytes, and teardrop cells warrant TT suspicion. This work advances thalassemia prevention by enabling efficient, low-cost carrier screening.
Keywords
Thalassemia; machine learning; artificial intelligence; erythrocyte; morphology

Biography:

Li JunXun has over 20 years of experience in blood cell morphology and hematologic disease diagnosis. In recent years, he has secured multiple research achievements in AI-based blood cell morphology recognition and hematologic disease diagnosis in Guangdong hematology field. His academic and professional positions include MHPE of Gulf University, Specialist in Medical Education and Associate Fellow of AMEE, and Standing Committee Member of Hematology Group, Laboratory Medicine Branch of Guangdong Medical Association.