Detection Of Thalassemia Using Blood Smear Images: A Machine Learning Approach

Abstract

Thalassemia is one of the most common genetic disorders worldwide, particularly prevalent in populations like Asian countries and African descent. In Pakistan thalassemia trait ranges from 5.0% to 7.0%, indicating the presence of more than 10 million carriers in the country. Moreover, the annual incidence of β-thal major (β-TM) in Pakistan is around 5000 children. Early detection of thalassemia carriers is crucial for effective management and prevention of severe forms of the disease. In this study, we propose a machine-learning approach for the detection of thalassemia carriers using blood smear images. Our methodology involves preprocessing the images to extract relevant features, including color, texture, and shape. We then employ deep learning models, including Convolutional Neural Networks (CNNs), to classify the images into thalassemia and non-thalassemia categories. The dataset consists of nearly 7108 blood images, including nine cell types associated with thalassemia. The highest accuracy in terms of machine learning models achieved from Random Forest of 91.1%. While on the other side, the highest accuracy of 90% was achieved from the MobileNetV2 model in applying deep learning algorithms. Our results show promising accuracy rates, with the potential for real-world application in thalassemia screening programs. This research contributes to the advancement of diagnostic methodologies for thalassemia and could lead to improved healthcare outcomes in the affected population.