iammarco11 / devolearn

Accelerate data driven research on embryos with deep learning models :microscope:

Contents

• Example notebooks
• Segmenting the C. elegans embryo
• Generating synthetic images of embryos with a GAN
• Predicting populations of cells within the C. elegans embryo
• Contributing to DevoLearn
• Links to datasets
• Contact us

Installation

pip install devolearn

Example notebooks

• Extracting centroid maps and making 3d centroid models

Segmenting the Cell Membrane in C. elegans embryo

• Importing the model

from devolearn import cell_membrane_segmentor
segmentor = cell_membrane_segmentor()

• Running the model on an image and viewing the prediction

seg_pred = segmentor.predict(image_path = "sample_data/images/seg_sample.jpg")
plt.imshow(seg_pred)
plt.show()

• Running the model on a video and saving the predictions into a folder

filenames = segmentor.predict_from_video(video_path = "sample_data/videos/seg_sample.mov", centroid_mode = False, save_folder = "preds")

• Finding the centroids of the segmented features

seg_pred, centroids = segmentor.predict(image_path = "sample_data/images/seg_sample.jpg", centroid_mode = True)
plt.imshow(seg_pred)
plt.show()

• Saving the centroids from each frame into a CSV

df = segmentor.predict_from_video(video_path = "sample_data/videos/seg_sample.mov", centroid_mode = True, save_folder = "preds")
df.to_csv("centroids.csv")

Segmenting the Cell Nucleus in C. elegans embryo

• Importing the model

from devolearn import cell_nucleus_segmentor
segmentor = cell_nucleus_segmentor()

• Running the model on an image and viewing the prediction

seg_pred = segmentor.predict(image_path = "sample_data/images/nucleus_seg_sample.jpg")
plt.imshow(seg_pred)
plt.show()

Generating synthetic images of embryos with a Pre-trained GAN

• Importing the model

from devolearn import Generator, embryo_generator_model
generator = embryo_generator_model()

• Generating a picture and viewing it with matplotlib

gen_image = generator.generate()  
plt.imshow(gen_image)
plt.show()

• Generating n images and saving them into foldername with a custom size

generator.generate_n_images(n = 5, foldername= "generated_images", image_size= (700,500))

---

Predicting populations of cells within the C. elegans embryo

• Importing the population model for inferences

from devolearn import lineage_population_model

• Loading a model instance to be used to estimate lineage populations of embryos from videos/photos.

model = lineage_population_model(device = "cpu")

• Making a prediction from an image

print(model.predict(image_path = "sample_data/images/embryo_sample.png"))

• Making predictions from a video and saving the predictions into a CSV file

results = model.predict_from_video(video_path = "sample_data/videos/embryo_timelapse.mov", save_csv = True, csv_name = "video_preds.csv", ignore_first_n_frames= 10, ignore_last_n_frames= 10, postprocess = False)

• Plotting the model's predictions from a video

plot = model.create_population_plot_from_video(video_path = "sample_data/videos/embryo_timelapse.mov", save_plot= True, plot_name= "plot.png", ignore_last_n_frames= 0, postprocess = False)
plot.show()

Links to Datasets

Model	Data source
Segmenting the cell membrane in C. elegans embryo	3DMMS: robust 3D Membrane Morphological Segmentation of C. elegans embryo
Segmenting the nucleus in C. elegans embryo	C. elegans Cell-Tracking-Challenge dataset
Cell lineage population prediction + embryo GAN	EPIC dataset

Authors/maintainers:

• Mayukh Deb
• Ujjwal Singh
• Dr. Bradly Alicea

Feel free to join our Slack workspace!