search

coffee19850519 / single_cell_spatial_image

A deep-learning framework for characterizing and visualizing tissue architecture from spatially resolved transcriptomics
MIT License
12 stars 6 forks source link

Pre-trained image segmentation model #10

Open ciciQ777 opened 2 years ago

ciciQ777 commented 2 years ago

Hi RESEPT authors,

Thank you for developing such a wonderful tool for spatial resolved transcriptomics! I'm so impressed by the high ARI in the spatial LIBD data. I would like to try this great tool on my own data and I have the following questions, and hope I can get some suggestions or help from you:

  1. I was wondering how can I get the pre-trained image segmentation deep-learning model by myself?
  2. And also does the example pre-trained model you provided in the Github page was trained by the 12 H&E staining images (sample 151507-151676 from http://spatial.libd.org/spatialLIBD/) with their manual annotations?
  3. Was the high ARI provided in your paper for the S6-S17 samples calculated based on the results using the pre-trained image segmentation model while the image-segmentation model is the one mentioned above?
  4. After I ran the tutorial on the sample 151669, it seems that the "Demo_result_tissue_architecture" folder only contain the images with tissue structures and top5_MI_value.csv, I was wondering how can I extract the cluster/tissue structure number like BayesSpace output?

Thank you very much and I appreciate your time! This will help me a lot!

coffee19850519 commented 2 years ago

Hello, Thanks for your interest in our work. Here are our responses to your inquiry:

  1. Please refer to this for building your own segmentation model with your own labeled spatial data.
  2. Yes, The example model was trained on the 11 samples (12 H&E samples except for sample 151669), where we utilized a leave-one-out strategy to validate our architecture.
  3. The ARIs of S6-S17 were achieved by the models from their leave-one-out validation.
  4. That is a good question. We updated this type of output to this function. You may find such results in a csv file named 'predicted_tissue_architecture.csv'.

Feel free to let us know if you have any further questions.