VPShuntDetector

VPShuntDetector is a Python project designed to detect various VPShunt valves from X-ray images. The tool supports several valve types including:

• Codman Certas
• Codman Hakim
• Codman Uni-Shunt
• paediGAV
• proGAV Gravitationseinheit
• proGAV
• proSA

Features

• Valve Detection: Identify multiple VPShunt valve types from X-ray images.
• Detailed Results: For each input image, the tool outputs:
  • A results.csv file containing the confidence and predicted label for each of the 5 folds, along with the aggregated final confidence and prediction.
  • An output image with a bounding box overlay showing the predicted valve model.
• Instruction Display: Optionally, if an instruction directory is provided, corresponding instruction images for each valve model will be displayed alongside the output image.

Installation

Clone the repository and install the package in editable mode:

git clone git@github.com:JannisStraus/VPShuntDetector.git
cd VPShuntDetector
pip install -e .

Usage

After installation, you can run the tool directly from the command line:

vpshuntdetector -i <input> -o <output>

If you have a directory with instruction images, include it using the --instructions flag (see Instruction Images for more details):

vpshuntdetector -i <input> -o <output> --instructions <instructions_directory>

Command Line Arguments

• -i, --input: (Required) Path to the input image or directory.
• -o, --output: (Required) Path where the output results will be saved.
• --instructions: (Optional) Directory with instruction images for valve models.
• -d, --device: (Optional) Device to perform inference on (e.g. cuda or cpu).

Instruction Images

When using the --instructions option, the provided folder should contain one PNG image for each supported valve type. The image files must be named exactly as the valve type with a .png or .jpg extension. For example, the folder should include the following files:

example_instructions/
├── Codman Certas.png
├── Codman Hakim.png
├── Codman Uni-Shunt.png
├── paediGAV.png
├── proGAV Gravitationseinheit.png
├── proGAV.png
└── proSA.png

Output Details

In the output directory specified by the -o or --output option, VPShuntDetector generates:

• results.csv: A CSV file containing, for every input image, the predicted label and its confidence score for each of the 5 folds. This file also includes the final aggregated confidence and prediction.
• Output Images: For every input image, an output image is saved with a bounding box overlay that highlights the predicted valve model.
• Instruction Display: If an instruction directory is provided via --instructions, the corresponding instruction image for the predicted valve model will be displayed alongside the output image.