Feature Selection Differences Across Different Tasks

[bl175]

Hello everyone,

I have a question regarding the features used in different tasks. Do the features differ between tasks, such as detection versus tumor origin prediction? If so, what are the key differences between the features used for each task, and how can we determine these differences in the code?

I appreciate your help and insights. Thank you!

[813BoZai]

I have same question about this.

[Dadatata-JZ]

@bl175
CHIEF provides WSI-level features, which serve as the "foundational" component. For each downstream task, a specific classifier (e.g., linear probe) may need to be trained using these features along with the corresponding labels (e.g., tumor origins, molecular markers). This follows a standard pipeline in computational pathology.