Issuu

International Research Journal of Engineering and Technology (IRJET)

e-ISSN: 2395-0056

Volume: 12 Issue: 11 | Nov 2025

p-ISSN: 2395-0072

www.irjet.net

“Diagnosing blood cell classification and reticulocytes using AI and deep learning methods” Prof. Shobha S Biradar1, Mamta2 1

Professor, Master of Computer Application, VTU, Kalaburagi , Karnataka ,India 2 Student, Master of Computer Application, VTU, Kalaburagi , Karnataka ,India

----------------------------------------------------------------------------***--------------------------------------------------------------------------------

ABSTRACT- The accurate and timely diagnosis of blood-

2. PROBLEM STATEMENT

related disorders is a critical component of clinical hematology, as abnormalities in blood cells often serve as early indicators of disease progression. Manual examination of peripheral blood smears by pathologists, although considered the gold standard, is time-consuming, labor-intensive, and prone to inter-observer variability. In particular, the detection and classification of different blood cell types—including red blood cells (RBCs), white blood cells (WBCs), platelets, and immature cells such as reticulocytes - are essential for diagnosing conditions such as anemia, leukemia, infections, and bone marrow dysfunctions.

The primary challenge lies in designing a system that can accurately detect and classify multiple objects in a video stream while maintaining real-time performance. Achieving this requires balancing speed, accuracy, and computational efficiency. Furthermore, real-world environments often present additional difficulties such as variations in lighting, object occlusion, background clutter, and camera noise, which make detection tasks more complex.

3. OBJECTIVES

Keywords: A small, standardized microscopic-blood benchmark that provides tiny, labeled single-cell images for rapid prototyping and reproducible comparisons. Computer vision, object detection.

The primary objective of this study is to design and develop an intelligent deep learning–based framework for the automated classification of blood cells and the accurate detection of reticulocytes from microscopic smear images. The study aims to address the limitations of manual diagnostic methods by introducing a scalable, interpretable, and reliable system that can be deployed across diverse clinical environments. To achieve this overarching goal, the following specific objectives are outlined:

1. NTRODUCTION A small, standardized microscopic-blood benchmark that provides tiny, labeled single-cell images for rapid prototyping and reproducible comparisons. It’s ideal for sanity-checking preprocessing, augmentation, and baseline CNN architectures before moving to larger clinical sets. Because the images are normalized and compact, experiments run quickly and hyper parameter sweeps are cheap. Many teams use this dataset as a first-step to validate training pipelines and metric reporting conventions. Treat results on this benchmark as initial proofs-of-concept rather than definitive clinical evidence. Use it to iterate fast and to establish reproducible baselines. [1]

4. METHODOLOGY USED The methodology of this study is structured to systematically design, implement, and evaluate a deep learning–based framework for blood cell classification and reticulocyte detection. It consists of multiple stages, beginning with dataset collection and preprocessing, followed by model development, training, evaluation, and deployment. Each stage has been carefully chosen to ensure accuracy, robustness, and clinical applicability of the proposed system.

A large multi- center white-blood- cell collection with tens of thousands of images and multi- expert labels that supports both classification and segmentation work. Its scale and diversity across microscopes and labs make it well suited for training robust deep classifiers and segmentation backbones. Annotated masks enable experimentation with detection→ segmentation→ classification pipelines and with

The first step involves data collection and annotation. Highresolution microscopic images of peripheral blood smears are obtained from publicly available hematological datasets such as BCCD, HemaCell, and other institutional sources. These images contain different types of blood cells including red blood cells, white blood cells, platelets, and reticulocytes. Expert annotation is used to label each cell type, ensuring that the dataset is reliable and suitable for supervised learning.

Mask- based feature extraction. The dataset is particularly useful for transfer-learning experiments and domainadaptation research. Because it reflects real-world variability, it’s a strong choice for pushing models toward clinical resilience. Use it to train models expected to generalize across labs and devices. [2]

Impact Factor value: 8.315

ISO 9001:2008 Certified Journal

Page 1004