README.md

---
license: apache-2.0
tags:
  - medical
  - chest-xray
  - federated-learning
---

# Ark+ fine-tuned chest X-ray classification checkpoint (FLIP tutorial)

`model.pt` is an Ark+ (Swin-Large, 768×768) checkpoint **federatedly fine-tuned** on the synthetic FLIP
tutorial splits
([`aicentreflip/tutorials-arkplus-cxr-classification`](https://huggingface.co/datasets/aicentreflip/tutorials-arkplus-cxr-classification))
with a fresh 5-class DECAF classification head (`Effusion`, `Consolidation`, `Infiltration`,
`Lung Nodule or Mass`, `Pneumothorax`).

It is consumed by the `arkplus_multimodel_classification_evaluation` tutorial in
[FLIP](https://github.com/londonaicentre/FLIP), which scores this fine-tuned model against the zero-shot
Ark+ foundation model on a shared hold-out cohort (per-lesion AUROC + pairwise DeLong).

## Not included: the Ark+ foundation model

This repo hosts **only** the FLIP fine-tuned checkpoint. The upstream Ark+ foundation model
(`Ark6_swinLarge768_ep50`) is **not** re-hosted here — request it from the original authors via
[this form](https://forms.gle/qkoDGXNiKRPTDdCe8) (see [`jlianglab/Ark`](https://github.com/jlianglab/Ark)).

Licensed Apache-2.0.