Enhancing Testicular Ultrasound Image Classification Through Synthetic Data and Pretraining Strategies

This repository contains the official resources for the TesticulUS dataset and the relative paper, as presented at the International Conference on Image Analysis and Processing (ICIAP) 2025.

Synthetic Dataset Access

A synthetically generated version of the dataset, created to augment the original data and facilitate further research, is available for exploration and download at the following link:

Explore the Synthetic TesticulUS Dataset

Generating New Images

Follow these steps to generate new synthetic images using our pretrained models:

1. Clone the repository:

   git clone https://github.com/AImageLab-zip/TesticulUS
   cd TesticulUS

2. Download pretrained model weights:

   • Download the model weights from Google Drive.
   • Place the downloaded .pt file in your preferred directory.

3. Generate images:

   • Run the following command to generate images. You can set --logging_dir=<your_path> to specify where logs and generated images will be saved (or use the default directory):

   python guided-diffusion/scripts/image_sample.py \
     --in_channels 1 \
     --image_size 256 \
     --batch_size <your_batch_size> \
     --num_samples <num_of_generated_images> \
     --learn_sigma True \
     --model_path <path_to_downloaded_model.pt>

   • The generated output will be saved in the specified logging_dir as files named samples*.npz.

4. Filter generated images (optional but recommended):

   • To filter the generated images using our custom method, run:

   python guided-diffusion/filter_generation.py <path_to_samples.npz> --output_path=<output_directory>

This process allows you to generate new synthetic images and optionally filter them for higher quality using our provided scripts and models.

Citation

If you use the TesticulUS dataset or you find this code and paper useful for your research, please cite our work:

@inproceedings{2025ICIAP, 
    publisher={Springer},
    venue={Rome, Italy}, 
    month={Sep}, 
    year={2025}, 
    pages={1--12}, 
    booktitle={Image Analysis and Processing – ICIAP 2025}, 
    title={{Enhancing Testicular Ultrasound Image Classification Through Synthetic Data and Pretraining Strategies}}, 
    author={Morelli, Nicola and Marchesini, Kevin and Lumetti, Luca and Santi, Daniele and Grana, Costantino and Bolelli, Federico}}