Towards Explainable AI in Leukemia Diagnosis: Advancing Peripheral Blood Smear Analysis
Akaila, Dawlat Samer Bashir
Akaila, Dawlat Samer Bashir
Author
Supervisor
Department
Computer Vision
Embargo End Date
2025-05-30
Type
Thesis
Date
2025
License
Language
English
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
Leukemia is a lifethreatening blood cancer that disrupts white blood cell (WBC) function, and its diagnosis through peripheral blood smear (PBS) analysis remains labor intensive and prone to expert variability. To address these challenges, this thesis proposes an AI-driven approach that mimics the structured reasoning of hematopathologists. The method begins by detecting WBCs in PBS images, then extracts detailed morphological features using deep learning. A multitask model simultaneously classifies WBC types and predicts key attributes relevant to leukemia diagnosis. Finally, a large language model (LLM) analyzes these structured features to classify leukemia subtypes, enhancing transparency and clinical relevance. Evaluated on a public dataset, the approach outperforms existing methods in accuracy and explainability, offering a more consistent and explainable AI solution for leukemia diagnosis.
Citation
Dawlat Samer Bashir Akaila, “Towards Explainable AI in Leukemia Diagnosis: Advancing Peripheral Blood Smear Analysis,” Master of Science thesis, Computer Vision, MBZUAI, 2025.
Source
Conference
Keywords
Explainable AI, Cell Detection, Foundation Models, Leukemia Classification, Large Language Model (LLM), Multi-task Learning