LAnA-MIMIC

LAnA-MIMIC

Layer-Wise Anatomical Attention model

Best current model in this collection: manu02/LAnA-Arxiv

Overview

LAnA is a medical report-generation project for chest X-ray images. The completed project is intended to generate radiology reports with a vision-language model guided by layer-wise anatomical attention built from predicted anatomical masks. This released checkpoint was trained on MIMIC-CXR only.

The architecture combines a DINOv3 vision encoder, lung and heart segmentation heads, and a GPT-2 decoder modified so each transformer layer receives a different anatomical attention bias derived from the segmentation mask.

How to Run

Standard AutoModel.from_pretrained(..., trust_remote_code=True) loading is currently blocked for this repo because the custom model constructor performs nested pretrained submodel loads. Use the verified manual load path below instead: download the HF repo snapshot, import the downloaded package, and load the exported model.safetensors directly. You must set an HF_TOKEN environment variable with permission to access the DINOv3 model repositories used by this project, otherwise the required vision backbones cannot be downloaded.

from pathlib import Path
import sys

import numpy as np
import torch
from PIL import Image
from huggingface_hub import snapshot_download
from safetensors.torch import load_file
from transformers import AutoTokenizer

repo_dir = Path(snapshot_download('manu02/LAnA-MIMIC'))
sys.path.insert(0, str(repo_dir))

from lana_radgen import LanaConfig, LanaForConditionalGeneration

config = LanaConfig.from_pretrained(repo_dir)
config.lung_segmenter_checkpoint = str(repo_dir / "segmenters" / "lung_segmenter_dinounet_finetuned.pth")
config.heart_segmenter_checkpoint = str(repo_dir / "segmenters" / "heart_segmenter_dinounet_best.pth")

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

model = LanaForConditionalGeneration(config)
state_dict = load_file(str(repo_dir / "model.safetensors"))
missing, unexpected = model.load_state_dict(state_dict, strict=True)
assert not missing and not unexpected

model.tokenizer = AutoTokenizer.from_pretrained(repo_dir, trust_remote_code=True)
model.move_non_quantized_modules(device)
model.eval()

image_path = Path("example.png")
image = Image.open(image_path).convert("RGB")
image = image.resize((512, 512), resample=Image.BICUBIC)
array = np.asarray(image, dtype=np.float32) / 255.0
pixel_values = torch.from_numpy(array).permute(2, 0, 1)
mean = torch.tensor([0.485, 0.456, 0.406]).view(3, 1, 1)
std = torch.tensor([0.229, 0.224, 0.225]).view(3, 1, 1)
pixel_values = ((pixel_values - mean) / std).unsqueeze(0).to(device)

with torch.no_grad():
    generated = model.generate(pixel_values=pixel_values, max_new_tokens=128)

report = model.tokenizer.batch_decode(generated, skip_special_tokens=True)[0]
print(report)

Intended Use

• Input: a chest X-ray image resized to 512x512 and normalized with ImageNet mean/std.
• Output: a generated radiology report.
• Best fit: research use, report-generation experiments, and anatomical-attention ablations.

MIMIC Test Results

These comparison tables are refreshed across the full LAnA collection whenever any collection model is evaluated.

Cross-Model Comparison: All Frontal Test Studies (3041 studies)

Cross-Model Comparison: Findings-Only Frontal Test Studies (2210 studies)

Data

• Full project datasets: CheXpert and MIMIC-CXR.
• Intended project scope: train on curated chest X-ray/report data from both datasets and evaluate on MIMIC-CXR test studies.
• Training data for this checkpoint: MIMIC-CXR only.
• Current released checkpoint datasets: MIMIC-CXR (findings-only) for training and MIMIC-CXR (findings-only) for validation.
• Current published evaluation: MIMIC-CXR test split, frontal-only (PA/AP) studies.

Evaluation

• Medical report metrics implemented in the repository include RadGraph F1 and CheXpert F1 (14-micro, 5-micro, 14-macro, 5-macro).

Experiment Model Descriptions

• LAnA-MIMIC-CHEXPERT: This variant was trained on a combined dataset of CheXpert and MIMIC-CXR using LoRA fine-tuning with the AdamW optimizer.
• LAnA-MIMIC: This model was trained on the MIMIC-CXR (findings-only) dataset using LoRA fine-tuning with the AdamW optimizer.
• LAnA: This model was trained on the MIMIC-CXR (findings-only) dataset using full-model optimization with AdamW instead of LoRA.
• LAnA-v2: This version keeps the same training setup as LAnA, but increases the effective global batch size from 16 to 128.
• LAnA-v3: This version keeps the same training setup as LAnA, including the effective global batch size of 16, but changes how EOS is handled so training and generation follow the same behavior. The model no longer uses the EOS token during training, and generation remained greedy without stopping when an EOS token was produced. In the previous setup, decoding was also greedy, stopped at EOS, and used a maximum of 128 new tokens.
• LAnA-v4: This version keeps the same decoding behavior as LAnA-v3, but increases the effective global batch size from 16 to 128.
• LAnA-v5: This version uses the training recipe from the original LAnA paper, while switching to the legacy CXR-Findings-AI generation behavior.
• LAnA-Arxiv: This model is the report-generation model created in the arXiv paper, packaged locally with its original legacy generation code.

Training Snapshot

• Run: LAnA-MIMIC
• This section describes the completed public training run.
• Method: lora_adamw
• Vision encoder: facebook/dinov3-vits16-pretrain-lvd1689m
• Text decoder: gpt2
• Segmentation encoder: facebook/dinov3-convnext-small-pretrain-lvd1689m
• Image size: 512
• Local batch size: 1
• Effective global batch size: 8
• Scheduler: cosine
• Warmup steps: 2636
• Weight decay: 0.01
• Steps completed: 52709
• Planned total steps: 52716
• Images seen: 421707
• Total training time: 7.8770 hours
• Hardware: NVIDIA GeForce RTX 5070
• Final train loss: 1.9587
• Validation loss: 1.7493

Status

• Project status: Training completed
• Release status: Completed training run
• Current checkpoint status: Final completed run
• Training completion toward planned run: 100.00% (3 / 3 epochs)
• Current published metrics correspond to the completed training run.

Notes

• Set HF_TOKEN with permission to access the DINOv3 repositories required by this model before downloading or running inference.
• segmenters/ contains the lung and heart segmentation checkpoints used to build anatomical attention masks.
• evaluations/mimic_test_metrics.json contains the latest saved MIMIC test metrics.