Abstract
Malaria kills hundreds of thousands of people each year, most of them in sub-Saharan Africa, where expert diagnosis is scarce and computer-aided diagnosis systems consistently fail outside the lab. The standard explanation is model architecture. We think the real problem is annotation. Existing datasets use incompatible labeling schemes and no study has measured how that choice affects diagnostic performance or cross-dataset transfer. We harmonize three public datasets across three supervision regimes: NIH Malaria Cell Images (binary labels), Lacuna Malaria Blood Smear (parasite bounding boxes), and the Broad Bioimage Benchmark Collection dataset 041 (BBBC041, stage-aware bounding boxes), and train independent models under each. All three regimes produce usable models on their own data. When we test across datasets, a clear pattern emerges: richer annotation does not produce more transferable models. Detection models failed in both transfer directions tested. A binary classifier transferred when image framing was compatible and failed when it was not. These results point to acquisition style, rather than label granularity, as the primary barrier. These findings carry direct implications for how African malaria datasets should be designed and collected.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_045.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/MIRASOL_045_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=e3PPklIoFJ
BibTex
@InProceedings{AdeFra_From_MICCAISAT2026,
author = { Adelakun, Frances AND Adesina, Comfort AND Lawal, Olamide AND Akano, Olatokun Shamsudeen AND Adeniyi, Stephen Adeyinka AND Oyetunji, Abdulbasit},
title = { { From Binary Labels to Stage-Aware Bounding Boxes: Evaluating Annotation Depth in Malaria Microscopy } },
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}
}
