International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 13 Issue: 02 | Feb 2026
p-ISSN: 2395-0072
www.irjet.net
Transfer Learning for Automated Anemia Diagnosis from Blood Smear Images: A Systematic Review Nehal Shivane1, Aditya Sontakke2, Piyush Salve3, Divya Pardeshi4 , Ratnamala Paswan5 1234Department of Computer Engineering, SCTR's Pune Institute of Computer Technology, Pune, Maharashtra,
India
5Professor, Department of Computer Engineering, SCTR's Pune Institute of Computer Technology, Pune,
Maharashtra, India ---------------------------------------------------------------------***--------------------------------------------------------------------Abstract - Imaging of Peripheral Blood Smears (PBS) helps experts identify and classify anemia subtypes optimally based on the structure of cells. But manual examination needs a plethora of time, is subjective, and susceptible to changes in staining and illumination across clinics. This review represents a comprehensive summary of recent advances in automatic anemia diagnosis using PBS. We have considered 21 studies selected according to PRISMA guidelines. We group existing methods into four categories including classical machine learning, CNN‑based classification, object detection, and hybrid approaches. Our results show that in the hybrid framework, combining CNN features with handcrafted features provides the highest accuracy at 91%. In comparison, pure CNN models reach 90.6% and pure ML models achieve 85.3%. We introduce the Clinical Alignment Score (CAS), which checks how well the model's focus matches with expert-annotated regions. Future directions include stain‑robust models, distributed learning, and multimodal integration to make anemia diagnosis accessible and reliable. Key Words: Peripheral Blood Smear, Anemia Classification, Transfer Learning, Deep Learning, Medical Image Analysis, Systematic Review, Explainable AI 1. INTRODUCTION 1.1 Motivation and Problem Statement Blood is the primary medium of transport for oxygen, nutrients, hormones, and metabolic wastes throughout the human body [6]. It constitutes red blood cells (RBCs), white blood cells (WBCs), platelets, and plasma, each of which plays a distinct, specialized role [6]. Anemia is a hematological condition that emanates from an inadequate concentration of RBCs or below optimum hemoglobin levels relative to the age and gender of the person [3]. It is the most prevalent blood disorder in the world that affects a large segment of the population including 42% of children under five and 40% of pregnant women [6], [9]. Anemia diminishes the oxygen carrying strength of the blood leading to tissue hypoxia, fatigue, and impeded cognitive and physical development [1], [6]. Peripheral Blood Smear (PBS) examination is a testing process where hematologists can visually inspect the structure of blood cells. It can be used to analyze vital parameters of cells including their size (anisocytosis), shape (poikilocytosis), central pallor and count. This helps identify subtypes and abnormalities [3], [6], [15]. But manual analysis takes a lot of time, is subjective, and can vary between observers [6], [12], [13]. Classical screening encounters issues like inconsistent staining, different levels of light, and insufficient image datasets [8], [15], [17], [19]. Costly laboratory tests can affect patient resources, healthcare systems and government budgets [9]. Although current blood analysers effectively perform a Complete Blood Count (CBC), they do not provide detailed examination of cell morphology at the pixel level [12]. Automated image analysis and machine learning techniques address these concerns by saving time and improving uniformity [1], [4], [6]. 1.2 Scope and Organization The scope of this survey looks into how deep learning is used in hematology. RBC morphology and classification of anemia subtypes are its major applications. It includes peer-reviewed contributions from 2020-2025, capturing how the field has transcended from classical machine learning to deep learning and models that use attention mechanisms. The scope concentrates on RBC classification and segmentation at the pixel level, but it does not cover general CBC automation. The paper is organized as follows: Section II establishes the review methodology. Section III provides clinical background and necessities. Section IV compares regular CNNs with hybrid and attention-driven fusion models. Section V presents comparison among the existing approaches. Section VI synthesizes challenges from the realm of research including data scarcity and inter-patient diversity. Section VII provides prospects. Section VIII wraps up with prime findings and future paths.
© 2026, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 647