Document Type : Original Article

Authors

• Samira Jafari ¹
• Seyyed Mohammad Mousavi ^{2, 3}
• Amir Hossein Nabizadeh ⁴
• Mahdi Dadgarnia ⁵
• Sayyed Mostafa Moosavi Khaliji ⁶

¹ Modeling in Health Research Center Institute for Futures Studies in Health Kerman University of Medical Sciences, Kerman, Iran

² Health Information Sciences Department, Faculty of Management and Medical, Information Sciences, Kerman University of Medical Sciences, Kerman, Iran

³ Medical Informatics Research Center, Institute for Futures Studies in Health, Kerman University of Medical Sciences, Kerman, Iran

⁴ Medical Informatics Research Center, Institute for Futures Studies in Health, Kerman University of Medical Sciences, Kerman, Iran/INESC-ID, Lisbon, Portugal

⁵ Faculty of Biomedical Sciences, Kharazmi University, Tehran, Iran

⁶ Department of Research and Technology Activities support of Kerman Provincial Unit - University of Applied Science and Technology, Tehran, Iran

Abstract

Introduction: The correct classification of blood cells serves as a crucial basis for medical diagnoses to detect different blood disorders.
Methods: Traditional blood analysis techniques including manual microscopy and automated analyzers encounter speed and accuracy limitations and scalability problems which led to the adoption of deep learning approaches. In this research, we used ResNet101V2 through transfer learning to identify eight blood cell types by analyzing 17,092 images.
Results: Yielding an overall accuracy rate of 86%, the network showed exceptional ability in distinguishing specific cell types, reaching 98% and 93% for neutrophils and platelets, respectively. Our results showed challenges in maintaining class balance and extracting meaningful features. Specifically, erythroblasts and lymphocytes exhibited low precision and recall values, indicating difficulties in distinguishing these cell types accurately. To solve these problems, we suggested using data augmentation and class-weighted adjustments for better generalization.
Conclusion: In the future, we suggest researchers investigate deep learning, ensemble learning, and hybrid AI frameworks to enhance classification accuracy while broadening model applicability for pathological cases.

In the future, we suggest researchers investigate deep learning, ensemble learning, and hybrid AI frameworks to enhance classification accuracy while broadening model applicability for pathological cases.

Highlights

Samira Jafari (Google Scholar) (PubMed)

Sayyed Mostafa Moosavi Khaliji (Google Scholar) (PubMed)

Amir Hossein Nabizadeh (Google Scholar) (PubMed)

Mahdi Dadgarnia (Google Scholar) (PubMed)

Seyyed Mohammad Mousavi (Google Scholar) (PubMed)

Keywords

• Deep Learning
• Microscopy
• Neutrophils
• Lymphocytes
• Erythroblasts

Main Subjects

• Bioinformatics