Skip to content
Emmanuel Idoko
Contact

Scholarship

Research & Publications

Research in medical imaging, multimodal learning, and applied AI systems. Also on Google Scholar.

2026

When Retrieval Hurts: Retrieval-Induced Hallucination in Medical Vision-Language Models

E. Idoko et al.

Africa in AI Workshop, NeurIPS 2026 (Springer LNCS) · Submitted · First author

Found that 95% of retrieval-augmented chest X-ray reports copy text verbatim from another patient's report (0% without retrieval). Retrieval doubles clinical accuracy when relevant (CheXbert F1 0.201 → 0.402) and collapses it to 0.043 when mismatched, over 100 MIMIC-CXR studies.

VAMAE: Vessel-Aware Masked Autoencoders for OCT Angiography

I. Abolade, P. Mireku, K. Chibundu, P. Ododo, E. Idoko, P. Omoigui, S. Odelola

ICPR 2026 — 28th Int'l Conference on Pattern Recognition · Accepted

PDFarXiv

2025

SharpXR: Structure-Aware Denoising for Pediatric Chest X-Rays

I. Abolade, E. Idoko, S. Odelola, P. Omoigui, A. Adebanwo, A. M. Iorumbur, U. Anazodo, A. Crimi, R. Confidence

MIRASOL Workshop, MICCAI 2025 · Published · pp. 83–92

Benchmarked seven denoising baselines; structure-preserving denoising raised downstream pneumonia-classification accuracy from 88.8% to 92.5%. Presented as a poster at MIRG-ICAIR 2025 and ACVSS 2026.

PDFarXivDOICodeAnnouncement

SharpXR Research Poster Presentation

E. Idoko

MIRG-ICAIR 2025 — Machine Intelligence Research Group Conference · Poster

In Progress

Cross-Modality Attention Fusion for Chest X-ray Diagnosis

E. Idoko et al.

Ongoing Research · Working Paper

View all on Google Scholar

Research activity

  • 2026

    African Computer Vision Summer School (ACVSS 2026) — poster & research hackathon

    Selected 1 of 31 from 312 applicants, with a full grant. Presented SharpXR at the poster session and competed in the research hackathon — building a task-aware super-resolution pipeline for satellite flood mapping — with keynotes from researchers at Oxford, EPFL, Amsterdam, Michigan, and Google DeepMind.

    About ACVSS
  • 2026

    “Attention Is All You Need” — paper walkthrough session

    A beginner-friendly walkthrough of the Transformer paper for early student researchers — self-attention, multi-head attention, and positional encoding.

    SlidesPhotos
  • 2025

    Presented the SharpXR poster at MIRG-ICAIR 2025

    Presented our pediatric chest X-ray denoising work to researchers at the MIRG-ICAIR conference, University of Lagos.

    PostLinkedIn
  • 2025

    SharpXR accepted to the MIRASOL Workshop at MICCAI 2025

    My first paper — structure-aware denoising for pediatric chest X-rays.

    Announcement