DESIGN OF A MULTI-MODAL DEEP LEARNING FRAMEWORK FOR EARLY LEUKEMIA DETECTION USING PERIPHERAL BLOOD SMEAR IMAGES AND CLINICAL DATA

ONYEBUCHI, DENIS IFEANYI (2026) DESIGN OF A MULTI-MODAL DEEP LEARNING FRAMEWORK FOR EARLY LEUKEMIA DETECTION USING PERIPHERAL BLOOD SMEAR IMAGES AND CLINICAL DATA. Other thesis, Godfrey Okoye University, Enugu Nigeria.

Text DOC-20260729-WA0032_ Download (86kB)

Abstract

Detection of leukaemia depends significantly on manually examining Peripheral Blood Smears (PBS), which is limited by inter-rater variation and the severe lack of specialist hematologists, especially in Nigeria and Sub-Saharan Africa. Current deep learning methods in automated leukaemia detection depend mostly on unimodal approaches that analyze PBS images alone without considering the structured clinical laboratory measurements typically considered alongside morphology by clinicians in the diagnostic process. This paper attempts to bridge these two gaps by developing a multi-modal deep learning system capable of integrating PBS images and clinical laboratory measurements. The proposed architecture includes a fine-tuned EfficientNetB3 for image feature extraction and a Multilayer Perceptron network for structured clinical data connected via a concatenation-attention fusion layer. Training and testing were carried out using the publicly available C-NMC 2019 benchmark dataset consisting of 10,661 PBS images. Synthetic feature vectors for the clinical laboratory data were obtained using class-conditional Gaussian distribution based on published clinical guidelines. On a test set of 1,600 samples, the multi-modal approach achieved 100% accuracy, precision, recall, F1 score, and AUC-ROC, performing better than state-of-the-art unimodal models in literature. The trained model was deployed as a full-stack web application using React.js and FastAPI, providing clinicians a real-time, browser-based decision support tool for leukaemia screening.

Item Type:	Thesis (Other)
Subjects:	Q Science > Q Science (General)
Divisions:	Faculty of Computing And Information Technology (FACIT)
Depositing User:	Cynthia Ugwuoti
Date Deposited:	29 Jul 2026 12:35
Last Modified:	29 Jul 2026 12:35
URI:	http://eprints.gouni.edu.ng/id/eprint/6073

Actions (login required)

View Item