zonechen1994
commented
3 months ago Hello, thank you for your attention.
I have some suggestions:
You can adjust the overlap ratio and calculate the union of predictive areas for the final prediction result. I conducted this experiment, and the results were significantly better in certain scenarios.
Another approach is to utilize the other model, like Hooknet-TLS, for predictions. By combining the union of predictive results from our model and others, we can potentially achieve improved outcomes.
Additionally, you can vary the lymphocyte count from 80 to other values. This will help determine the optimal lymphocyte parameter for your specific situation.
Our goal is to enhance our training distribution comprehensively. However, false positives, similar to the irregular results you mentioned, still persist. Addressing this is a focus of my future work. Thank you!
Hello, thank you very much for the code you published on GitHub. It is a pioneering contribution. I am currently encountering some issues. I used your get_patch_tls.py code to segment TLS in my patch data. However, it seems that a single TLS might be divided into multiple parts. After segmentation, I stitched them back into a complete image similar to the size of a WSI based on the coordinates of each patch. But the TLS appears very irregular. Do you know what aspects I should consider to address this issue? Because I need to quantify the WSI for a patient subsequently. Looking forward to your reply.