Architecture-Dependent Synthetic Data Strategies for Leukemia Diagnosis with Explainable Deep Learning

Malik, Ibrahim and Agarwal, Bharat (2026) Architecture-Dependent Synthetic Data Strategies for Leukemia Diagnosis with Explainable Deep Learning. In: Artificial Intelligence and Cognitive Science. AICS 2025. Communications in Computer and Information Science (2950). Springer, Cham, Dublin, Ireland, pp. 1-14. ISBN 978-303225808-3

Full text not available from this repository.

Abstract

Deep learning for leukemia diagnosis via blood smear microscopy faces data scarcity and interpretability challenges. We demonstrate that optimal synthetic data generation is architecture-dependent: evaluating three GAN variants (DCGAN, WGAN, cGAN) across CNN, Vision Transformer, and Hybrid CNN-ViT architectures on the C-NMC dataset reveals distinct preferences. CNNs achieve best performance with cGAN augmentation (71% accuracy), ViTs excel with WGAN data (74% accuracy, 0.50 sensitivity), while Hybrid models achieve highest accuracy (80%) but only with DCGAN, exhibiting extreme sensitivity to GAN choice with 33% accuracy variation.

For clinical screenings that require prioritizing the reduction of false negatives, ViT-WGAN provides the best performance-robustness tradeoff at a sensitivity improvement of 167% (0.187 > 0.500). Multi-method XAI validation using Grad-CAM, SHAP, and LIME on identical test images confirms convergent attention towards clinically relevant morphological features, including cell boundaries and nuclear irregularities. This challenges conventional single-GAN-for-all approaches, demonstrating synthetic data strategies must be co-designed with target architectures for medical AI deployment.

Item Type:	Book Section
Uncontrolled Keywords:	Explainable AI; GANs; Leukemia Diagnosis; Medical AI; Vision Transformers
Subjects:	R Medicine > RC Internal medicine > RC0254 Neoplasms. Tumors. Oncology (including Cancer) T Technology > Biomedical engineering Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence > Computer vision Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence > Computer vision Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Staff Research and Publications
Depositing User:	Tamara Malone
Date Deposited:	18 Sep 2026 14:37
Last Modified:	18 Sep 2026 14:37
URI:	https://norma.ncirl.ie/id/eprint/9934

Actions (login required)

View Item