Bone Age Crop

🚀 Transformers for Bone Age Hand Radiograph Cropping

This model is designed to crop hand radiographs, aiming to standardize the image input for bone age models, thus enhancing the accuracy and consistency of bone age assessment.

🚀 Quick Start

To use the model, follow these steps:

import cv2
import torch
from transformers import AutoModel

model = AutoModel.from_pretrained("ianpan/bone-age-crop", trust_remote_code=True)
model = model.eval()
img = cv2.imread(..., 0)
img_shape = torch.tensor([img.shape[:2]])
x = model.preprocess(img)
x = torch.from_numpy(x).unsqueeze(0).unsqueeze(0)
x = x.float()

# if you do not provide img_shape
# model will return normalized coordinates
with torch.inference_mode():
  coords = model(x, img_shape)

# only 1 sample in batch
coords = coords[0].numpy()
x, y, w, h = coords
# coords already rescaled with img_shape
cropped_img = img[y: y + h, x: x + w]

If you have pydicom installed, you can also load a DICOM image directly:

img = model.load_image_from_dicom(path_to_dicom)

✨ Features

• Standardization: Crops hand radiographs to standardize the image input for bone age models.
• Lightweight Backbone: Utilizes a lightweight mobilenetv3_small_100 backbone for efficient processing.
• Normalized Coordinates: Predicts normalized xywh coordinates for cropping.

📦 Installation

The installation details are not provided in the original document. If you want to use this model, you may need to install the transformers library and other necessary dependencies like cv2, torch, and pydicom if not already installed. You can use the following commands:

pip install transformers
pip install opencv-python
pip install torch
pip install pydicom

💻 Usage Examples

Basic Usage

import cv2
import torch
from transformers import AutoModel

model = AutoModel.from_pretrained("ianpan/bone-age-crop", trust_remote_code=True)
model = model.eval()
img = cv2.imread(..., 0)
img_shape = torch.tensor([img.shape[:2]])
x = model.preprocess(img)
x = torch.from_numpy(x).unsqueeze(0).unsqueeze(0)
x = x.float()

# if you do not provide img_shape
# model will return normalized coordinates
with torch.inference_mode():
  coords = model(x, img_shape)

# only 1 sample in batch
coords = coords[0].numpy()
x, y, w, h = coords
# coords already rescaled with img_shape
cropped_img = img[y: y + h, x: x + w]

Advanced Usage

# Load a DICOM image directly if pydicom is installed
img = model.load_image_from_dicom(path_to_dicom)

📚 Documentation

The model was trained and validated using 12,592 pediatric hand radiographs from the RSNA Pediatric Bone Age Challenge using an 80%/20% split. On single-fold validation, the model achieved mean absolute errors (normalized coordinates) of:

x: 0.0152
y: 0.0121
w: 0.0261
h: 0.0213

🔧 Technical Details

This model uses a lightweight mobilenetv3_small_100 backbone to predict normalized xywh coordinates for cropping hand radiographs. The training data is sourced from the RSNA Pediatric Bone Age Challenge, with a clear 80%/20% split for training and validation. The validation results show relatively low mean absolute errors in normalized coordinates, indicating the model's effectiveness and accuracy.

📄 License

This model is licensed under the apache-2.0 license.

📋 Information Table