Abstract

Developing clinical image analysis models heavily relies on diverse and representative datasets. This is particularly important for malaria, a major public health challenge in Sub-Saharan Africa, a region concurrently facing a shortage of medical experts and diagnostic technology. Consequently, creating large, representative African datasets is a promising solution for training robust machine learning models, particularly for fine-tuning recently emerged vision transformer architectures. In this study, we propose a clinical description-guided malaria blood smear image synthesis algorithm based on Gaussian-regularized spatial masking and localized geometric perturbations. Our approach ensures realistic boundary transitions, enhances sample diversity, and preserves structural consistency. The comprehensive evaluations demonstrate that the generated images match the original data with high statistical consistency, capturing 80% of data variance within the first 10 components and reaching 95% similarity in deep feature distribution. The newly proposed algorithm generated up to 500 image variants in approximately one minute while requiring about 1 GB of RAM and operating under CPU-based execution, suggesting its potential applicability in resource-constrained environments.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_006.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/MIRASOL_006_supp.pdf

Link to Open Review

Open Review Page: https://openreview.net/forum?id=AxlVLQsvNq

BibTex

@InProceedings{NigGeb_Gaussian_MICCAISAT2026,
        author = { Nigusie, Gebregziabihier AND Mamaye, Yimer},
        title = { { Gaussian Regularized Spatial Masking for Enhanced Malaria Blood Smear Augmentation in Resource-constrained Settings } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17264},
        month = {pending},
        page = {pending}
}


back to top