Skip to content

FAIBI 2nd Edition — Foundation AI models in Biomedical Imaging (Workshop)

Closing-day workshop on foundation models in biomedical imaging, co-organized by Jia Wu (MD Anderson Cancer Center) — direct cancer-center authorship signal among the workshop slate, with explicit oncology-imaging FM emphasis.

  • Speaker(s) / Authors: Co-organizers: Sharib Ali (Stirling), Rashid Qureshi (Salim Habib University), Islem Rekik (Imperial College London), Jia Wu (MD Anderson Cancer Center), Mohsin Bilal (Birmingham City)
  • Affiliation(s): Stirling · Salim Habib · Imperial · MD Anderson Cancer Center · Birmingham City
  • Anchor type: workshop / synthesis
  • Track / day: Sat Apr 11 PM (closing day, satellite track)
  • IEEE Xplore: TBD (workshop papers indexed separately from main proceedings; some FAIBI papers may appear as Xplore workshop volumes)
  • Status: post-meeting

What it does

FAIBI 2nd edition consolidates foundation-model methods for biomedical imaging across pathology, radiology, microscopy, and multimodal — the same author pool that populates MICCAI's COMPAY / DGM4MICCAI workshops and CVPR's FMV / MMFM-BIOMED. The cancer-relevance hook is Jia Wu's co-organizer slot: Wu's MD Anderson group focuses on radiomics, radiogenomics, and FM-based oncology imaging biomarkers, which biases the program toward cancer-axis FM applications relative to FAIBI 1st edition.

How it works / methodology

Workshop-format: invited talks + short oral / poster papers (typical FAIBI 2026 format ~10–20 papers across pathology / radiology / multimodal tracks). Methods center on (a) pretraining recipes for medical image encoders (SSL on H&E, CT, MR, US), (b) downstream task adaptation (segmentation, classification, generation), (c) evaluation protocols including fairness / robustness / OOD generalization, (d) clinical translation pathways.

Headline benchmarks

TBD — workshop-level synthesis; named-tool papers from the program will get individual tool pages where artifacts are released. Targeted: any oncology-imaging FM update, any pathology FM evaluation paper, any radiology FM with a cancer subgroup analysis.

How it works

Core idea. FAIBI is a curation venue, not a single methodology — but the workshop's structure operationalizes a shared methodological program across pathology, radiology, microscopy, and multimodal foundation models: SSL pretraining at scale, downstream task adaptation with minimal labels, and clinical-translation-aware evaluation (fairness, OOD robustness, regulatory pathway).

Inputs / outputs. Inputs (across the program): H&E whole-slide images, CT / MR / US volumes, microscopy stacks, and paired text / molecular labels. Outputs span the full medical-imaging task surface — classification, segmentation, generation, retrieval, captioning, agentic workflows — with a common emphasis on transfer to oncology-imaging endpoints under the Jia Wu co-organizer steering.

Key innovation. The cancer-center-organized workshop slot is the load-bearing methodological framing: FAIBI 2nd edition explicitly biases toward FM evaluations that include cancer-relevant cohorts and outcomes (lung CT screening cohorts, breast mammography reader studies, pathology survival prediction, radiogenomics) rather than generic medical-imaging benchmarks. The implicit hypothesis tested across the program is that domain-pretrained FMs + cancer-specific adaptation outperform generalist medical FMs on oncology tasks.

Parameters. Workshop-level — papers span small (single-modality, sub-10M parameter) adaptations to large (>1B-parameter foundation-model) updates. Typical FAIBI 2026 program likely contains ~10–20 short oral / poster papers across pathology / radiology / multimodal tracks.

Canonical example. A representative FAIBI 2nd-edition paper type: take a published pathology FM (UNI, CHIEF) or radiology FM (Med-PaLM-class, BiomedCLIP), apply minimal adaptation (linear probe, LoRA, few-shot prompting) to a cancer-axis downstream task (lung-CT nodule malignancy, breast-MRI lesion classification, pathology-based survival prediction), and report performance with fairness / OOD / radiologist-reader-study comparison. Named-tool releases from the workshop graduate to individual tool pages.

Where it fits in the corpus

  • AACR 2026: indirect via Jia Wu / MD Anderson radiomics line; pathology FM updates cross-link to CHIEF, UNI, Prov-GigaPath
  • MICCAI 2026: direct sibling — COMPAY (Sep 27–Oct 1) is the methodologically-nearest workshop; same author pool, fuller papers
  • CVPR 2026: FMV / MMFM-BIOMED workshops are the upstream architecture-feeder venues
  • RSNA 2026: if any radiology FM presented here reaches deployment, RSNA tool index gets the clinical-counterpart entry

Notes

FAIBI 2nd edition's MD Anderson co-organizer slot is the direct cancer-center authorship signal that elevates this workshop above the other ISBI 2026 satellite events for the cancer-relevance filter. Watch for the workshop's accepted-paper list (typically posted ~1 week post-meeting) and pick named-tool papers for individual tool pages.