Hologram-Project

'Diffraction-Informed Deep Learning for Molecular-Specific Holograms of Breast Cancer Cells '
Tzu-Hsi Song, Mengzhi Cao, Jouha Min, Hyungsoon Im, Hakho Lee, Kwonmoo Lee

Architecture

Lens-free digital in-line holography (LDIH) offers a large field of view at micrometer-scale resolution. However, the complex nature of diffraction images (holograms) produced by LDIH poses challenges for human interpretation and necessitates time-consuming computational processing for object image reconstruction. To address these challenges, we present HoloNet, a novel deep learning architecture specifically designed for the direct analysis of diffraction images for cellular diagnosis. HoloNet efficiently and accurately classifies breast cancer cell types and quantifies molecular marker intensities with high precision and interpretability. Additionally, the pretrained multi-tasked HoloNet has accurately classified breast cancer cell lines. The project shows that HoloNet offers a robust solution to the unique challenges of holographic data analysis, enhancing both the accuracy and interpretability of cellular diagnostics by seamlessly integrating computational imaging and deep learning.

Installation

To install necessary library packages, run the following command in your terminal:

pip install -r requirements.txt

Instructions

• Clone the repo to your project folder by using the following commend:

  git clone https://github.com/kleelab-bch/HoloNet

• Prepare the dataset as mat file and copy to the Data folder.

  • The training data has been uploaded. Please unzip the All_Data file to access the data.
  • If the user wants to use their own data, please put it in the Data folder.

• Follow the order of codes

  • Run main.py to get the model and prediction results for classification or regression.
    • Please read the instructions in main.py to swtich different models.

• The results will be printed on the terminal.

Note

• The image data consists of two channels (470 nm and 625 nm).
• The training data information (cell types and cell lines) is included in the All_Data file. The order related to cells is the same as in this article.
• The data is saved as mat format. If the user wnats to change it, the data collection function is in the Utilities.py
• The lib folder includes all dependencies required for the HoloNet and related models.
• All trained models are saved to the Model_Save folder.

License

This project is licensed under the MIT License.

Contact

If you have any question about the date or code, please contact [email protected]