Anemia detection and classification from blood samples using data analysis and deep learning*

Abstract

This study aims to examine the possibility and impact of utilizing data science on blood samples to rapidly and proactively identify underlying health issues. By utilizing effective algorithms, models will be constructed to address these problems and determine potential healthcare options based on geographical location. Once data is gathered, health officials will be notified of major diseases and individuals at risk or already affected. Authentic blood samples are used to ensure the credibility and validity of the proposed system. The data was collected during a volunteer-led hemoglobin blood test camp specifically for women residing in impoverished areas, resulting in a total of 551 samples. The effectiveness of this technique has been assessed through experimental results based on Hb, RDW%, MCV, RBC, and M-Index. The proposed data analysis and deep learning algorithm achieved average values of haemoglobin count 11.67 g/dL with a 1.33 standard deviation, RDW 14.59%, MCV 81.45, RBC 4.37 per microliter with a variance of 0.5, and M-Index 19.56. The experimental results achieved 97.60% accuracy, demonstrating the effectiveness of the proposed technique for classifying anemia and its subtypes. The experimental results indicate better overlap between the automated identification of anemia and manual detection by the experts.

Keywords

• Deep learning
• hemoglobin (Hb)
• Red Blood Cell Distribution Width (RDW)
• mean corpuscular volume (MCV)
• Red Blood Cell count (RBC)
• Mentzer Index (M-Index)