plasma-cell-detection

This is the code repository for the paper Deep learning accurately quantifies plasma cell percentages on CD138-stained bone marrow samples (2022). It contains annotated training data (images and labels), a trained model, and a web application for using the model to evaluate plasma cell percentages in example microscopy images. See the paper (open-access) for more details.

Running the web application

To use the web interface, you will need to set up a Python environment with the required dependencies on a computer with sufficient disk space and CPU/GPU capabilities. The following instructions describe a method to do so by using the conda package manager on Windows.

1. Install the latest version of Miniconda at this link (this should by default install to a folder under C:\Users<your-user> so as to not require administrator permissions): https://docs.conda.io/en/latest/miniconda.html
2. Download the latest version of this repository in GitHub (click the green button and select "Download as ZIP") and extract its contents to a known location, e.g. C:\plasma-cell-detection for simplicity.
3. From the Start Menu, open the program Anaconda Prompt (Miniconda 3).
4. Type cd C:\plasma-cell-detection and press enter to navigate to that directory.
5. Create a new conda environment, run: conda create --name dlweb --file conda-env-win-<cpu/gpu>.txt, replacing <cpu/gpu> with gpu if you have a compatible NVIDIA GPU (CUDA >= 11.7), otherwise with cpu. This will install the required dependencies into a new conda environment called dlweb (if an environment with the same name already exists, it will prompt you to remove it first; you can also adjust this name as desired).
6. Type conda activate dlweb and press enter. The prompt should now start with (dlweb).
7. Type flask run and press enter to start the web app. Note on initial run, the dependent pre-trained PyTorch VGG models will be downloaded, which may take some time.
8. The web app should start in your browser, or you can access it at http://127.0.0.1:5000/ in a web browser. Follow the instructions on the app to proceed.

Example screenshot

This screenshot shows output from uploading the image in microscope/Image5 (138-029A).tif to the web interface.