Configurable Segmentation & CSV Mask Naming Conventions

[dmandache]

Problem : framework currently hardcoded for liver segmentation

The framework lacks a single, configurable way to define what is being segmented (e.g., which organ, tumor vs. organ, merged masks) and how segmentation mask paths are represented in CSV outputs. This leads to ambiguity and inconsistent downstream use.

Requested Enhancements

1. Configurable segmentation scope

• Allow configuration of which organ(s) to segment.
• Support explicit tumor segmentation options (tumor-only, organ-only, both).
• Support merged masks (e.g., organ + tumor) as a configurable output.

2. Global naming convention for CSV columns

• Define a consistent, documented naming scheme for columns that store segmentation mask paths.
• Apply the naming convention across the entire framework (training, inference, evaluation, export).
• Ensure naming scales to multiple organs and mask types (e.g., liver_mask_path, liver_tumor_mask_path, liver_merged_mask_path).

Acceptance Criteria

• A single config entry (or section) specifies target organ(s) and mask types.
• CSV outputs use standardized column names per the agreed convention.
• Existing pipelines are updated to read/write the new columns consistently