Multiclass classification of thalassemia types using complete blood count and HPLC data with machine learning

Area of Science:

• Medical Diagnostics
• Computational Biology
• Genetics

Background:

• Thalassemia is a prevalent genetic blood disorder causing anemia, affecting over 100 countries.
• Challenges persist in diagnosing thalassemia, particularly in high-prevalence regions like Pakistan.
• Current diagnostic methods require enhancement for accuracy and accessibility.

Purpose of the Study:

• To evaluate machine learning models for detecting alpha and beta-thalassemia (minor and major types).
• To assess model performance using data from Complete Blood Count (CBC) and High-Performance Liquid Chromatography (HPLC).
• To investigate the efficacy of these models for Pakistani patient data.

Main Methods:

• Utilized K-nearest Neighbor (KNN), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost) algorithms.
• Trained and tested models on CBC and HPLC diagnostic data.
• Focused on discriminating between different thalassemia variants.

Main Results:

• XGBoost demonstrated superior performance, achieving approximately 99.5% accuracy on CBC data and 99.3% on HPLC data.
• SVM also showed strong performance, with a 99.4% testing accuracy on HPLC data.
• All tested models exhibited high accuracy in detecting thalassemia types using patient data.

Conclusions:

• Machine learning models, especially XGBoost, show high accuracy in detecting thalassemia using CBC and HPLC data.
• These computational approaches can significantly aid thalassemia diagnosis in resource-limited settings.
• This study pioneers the use of machine learning for predicting thalassemia forms from diagnostic reports.