Empty Output Files After Successful Processing of H&E Slides for ccRCC

[alisongonpereira]

Hello Dr. Mart and team,

I am using the pathology-hooknet-tls tool to identify TLSs and GCs in H&E-stained slides of diagnosed ccRCC samples. Despite following the installation and preprocessing instructions meticulously, the output files remain empty, and I am unable to see the expected results.

Environment Details:

Docker container setup as per the documentation.
Input images preprocessed to convert .svs to .tif with spacings of 0.5µm for the image and 2.0µm for the mask.
Weights file (weights.h5) downloaded from Zenodo and verified inside the Docker container.
Run configurations and paths verified, and the program accesses the correct files without any path-related errors.

Execution Process:

The preprocessing and setup steps complete without any errors.
The program successfully processes the input files (both .tif and mask), writes the output files, and finishes the run as expected.

Problem:

The outputs, specifically the JSON and XML files intended to contain annotations and post-processed results, are effectively empty (showing only {} or minimal metadata without actual data).
The .tif files for TLS and GC heatmaps and predictions are generated but do not display the expected biological features.

Tree Output of Files (for reference):

css └── [4.0K] images ├── [111M] 11514_22_hooknettls_heat1.tif ├── [119M] 11514_22_hooknettls_heat2.tif ├── [1.4M] 11514_22_hooknettls.tif ├── [157M] 12400_22_hooknettls_heat1.tif ├── [162M] 12400_22_hooknettls_heat2.tif ├── [1.0M] 12400_22_hooknettls.tif ├── [138M] 2675_22_hooknettls_heat1.tif ├── [142M] 2675_22_hooknettls_heat2.tif ├── [1.5M] 2675_22_hooknettls.tif ├── [120M] 3737_22_hooknettls_heat1.tif ├── [126M] 3737_22_hooknettls_heat2.tif ├── [1.2M] 3737_22_hooknettls.tif ├── [4.0K] filtered │ ├── [ 2] 11514_22_hooknettls_gc_filtered.json │ ├── [ 2] 11514_22_hooknettls_tls_filtered.json │ ├── [ 2] 12400_22_hooknettls_gc_filtered.json │ ├── [ 2] 12400_22_hooknettls_tls_filtered.json │ ├── [ 2] 2675_22_hooknettls_gc_filtered.json │ ├── [ 2] 2675_22_hooknettls_tls_filtered.json │ ├── [ 2] 3737_22_hooknettls_gc_filtered.json │ └── [ 2] 3737_22_hooknettls_tls_filtered.json ├── [4.0K] post-processed │ ├── [ 105] 11514_22_hooknettls_post_processed.xml │ ├── [ 105] 12400_22_hooknettls_post_processed.xml │ ├── [ 105] 2675_22_hooknettls_post_processed.xml │ └── [ 105] 3737_22_hooknettls_post_processed.xml └── [4.0K] raw ├── [ 2] 11514_22_hooknettls_with_confidence.json ├── [ 105] 11514_22_hooknettls.xml ├── [ 2] 12400_22_hooknettls_with_confidence.json ├── [ 105] 12400_22_hooknettls.xml ├── [ 2] 2675_22_hooknettls_with_confidence.json ├── [ 105] 2675_22_hooknettls.xml ├── [ 2] 3737_22_hooknettls_with_confidence.json └── [ 105] 3737_22_hooknettls.xml

Sample of the console outputs is in the attached file, for two samples.

Could you please assist in identifying what might be causing these empty outputs? Any guidance on additional configurations or adjustments to the processing parameters that may resolve this issue would be greatly appreciated.

Thank you for your support and for developing such a valuable tool for the research community.

Best regards, Alison Pereira outputs_2samples.txt

[martvanrijthoven]

Dear Alison,

Thank you for your interest in HookNet-TLS! And thanks for following the instructions. It seems indeed that everything went alright but no output is generated. Would it be possible to share the original image with me (mart.vanrijthoven@gmail.com)? Then I will check what is going wrong.

Best wishes, Mart

[alisongonpereira]

Thank you for your prompt response!

I have uploaded the input files used for our analysis to Google Drive. There are two .zip files, one for each sample. These zips contain the original .svs files and the 'converted_0-5' directory which includes the .tif files generated following your preprocessing.ipynb tutorial. Despite the directory's name suggesting a spacing of 0.5µm, I assure you that the tutorial was followed precisely, with spacings of 0.5µm for the sample and 2.0µm for the mask.

Thank you once again for your support.