BLOOD CANCER IDENTIFICATION USING HYBRID ENSENBLE DEEP LEARNING

Abstract

Blood cancer is one of the most life-threatening
diseases that affects the production and functioning
of blood cells, making early diagnosis essential for
successful treatment and patient survival.
Traditional microscopic examination performed by
hematologists is time-consuming, subjective, and
prone to human error due to variations in staining,
overlapping cells, and morphological similarities
among cancer subtypes. To overcome these
limitations, this project proposes a hybrid ensemble
deep learning framework for automated blood
cancer identification using microscopic blood
smear images. The proposed system integrates
advanced Convolutional Neural Network (CNN)
architectures such as EfficientNet, ResNet, and
VGG to improve feature extraction and
classification accuracy. The system performs
preprocessing techniques including normalization,
resizing, and augmentation to enhance image
quality and dataset diversity. The ensemble model
automatically learns complex cellular patterns such
as nucleus irregularities, texture variations, and
abnormal white blood cell structures without
relying on manual feature engineering. The
developed framework classifies blood cancer cells
into different categories including Benign, Early
Pre-B, Pre-B, and Pro-B Acute Lymphoblastic
Leukemia (ALL). The implementation uses
TensorFlow Lite and Flask for efficient deployment
and real-time prediction. Experimental evaluation
demonstrates improved accuracy, robustness,
scalability, and reduced computational complexity
compared with traditional machine learning
methods. The proposed system assists
hematologists by providing fast and reliable
computer-aided diagnosis, thereby reducing
diagnostic delays and improving clinical decisionmaking.
Furthermore, the framework can be
extended to support large-scale healthcare
applications and future AI-driven medical
diagnostic systems