[FEAT EXTRACTION] Handling Multiple CT Slices in Survival Analysis for Radiomics Feature Extraction

[tadongguk]

I am working with CT scan data to perform survival analysis. For each patient, I have several CT slices (multiple images) from their scan. I used Radiomics to extract features like texture and shape from each of these slices. However, I am not sure how to use this data for survival analysis because:

Each patient has many slices (for example, 50 images from a single scan).
For each patient, I have survival data (e.g., how long they lived after treatment or when a disease reoccurred).
The problem is:
For every patient, I have multiple feature sets (one from each CT slice), but only one survival time and event (such as survival time or time to disease recurrence).

Example:
Let's say I have a patient with 50 CT slices. I extract features from all 50 slices, which gives me 50 sets of features. But I only have one survival time (e.g., the patient lived for 3 years) and one event status (e.g., the patient passed away). The question is: how do I combine these multiple slices into one result that I can use for survival analysis?

Should I:

Average the features from all slices?
Pick one slice (e.g., the most central slice)?
Use advanced techniques (e.g., deep learning or multi-instance learning) that can handle multiple slices at once?
I need advice on the best approach to handle multiple slices for survival analysis.