https://github.com/dineshpinto/ml-droplet-recognition
Neural network for micro-fluidic droplet LLPS recognition
https://github.com/dineshpinto/ml-droplet-recognition
biology deep-learning keras machine-learning neural-network tensorflow
Last synced: 2 months ago
JSON representation
Neural network for micro-fluidic droplet LLPS recognition
- Host: GitHub
- URL: https://github.com/dineshpinto/ml-droplet-recognition
- Owner: dineshpinto
- License: mit
- Created: 2022-04-25T11:12:09.000Z (about 4 years ago)
- Default Branch: main
- Last Pushed: 2022-07-06T10:32:53.000Z (almost 4 years ago)
- Last Synced: 2025-10-24T01:56:37.042Z (8 months ago)
- Topics: biology, deep-learning, keras, machine-learning, neural-network, tensorflow
- Language: Jupyter Notebook
- Homepage:
- Size: 4.04 MB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Droplet Detection with Neural Networks
Use a Convolutional Neural Network built in TensorFlow and Keras to detect a droplet in an experimental data set.
## Test Data
The neural net consists of 4 layers, and for testing data shows reasonable results as shown below:
## Experimental Data
The goal is to apply it to a biological sample and detect droplet formation. The droplet as imaged by a microscope
looks like:
The model is trained on a subset of real data. The trained model is then used to find the droplet in the experimental
images.
An example of this is shown below:
**Note:** All biological droplet data is sourced from [@cfsb618](https://github.com/cfsb618)
## Model
## Installation
1. Create the conda environment from file (where `xx` is ether `win10` or `macm1`)
```shell
conda env create --file conda-env-xx.yml
```
3. Activate environment
```shell
conda activate ml_droplet
```
4. Add environment to Jupyter kernel
```shell
python -m ipykernel install --user --name=ml_droplet
```
### Performance
For optimal performance use a CUDA compatible NVIDIA GPU with the cuDNN libraries. The training times are around 3
orders of magnitude shorter.
A guide on how to do this is given in the [Tensorflow docs](https://www.tensorflow.org/install/gpu).
On Apple Silicon, use the `tensorflow-metal` plugin. A guide on how to do this is given in
the [Apple developer docs](https://developer.apple.com/metal/tensorflow-plugin/).
## Usage
1. Place your training data in `training_data/` and the corresponding droplet labels in `droplet_labels.py`
2. Train the model
```shell
python neural_network_training.py
```
4. This will train the neural network model and save the resulting model in `models/droplet_detection_model`
5. Test the model by running an automated test set
```shell
python plot_results.py
```
OR manually using the JupyterLab Notebook
```shell
jupyter lab DropletDetectionTesting.ipynb
```
## TODO
- [x] Add automated hyperparameter optimization.
- [ ] Add droplet feature detection, such as detecting the point of phase separation
- [ ] Add random sampling for training and testing data (example code below)
### Export conda environment
```shell
conda env export --no-builds | grep -v "^prefix: " > conda-env.yml
```