FlexiCT-2D

FlexiCT-2D is the slice-level CT encoder in the FlexiCT family.

Input and preprocessing

Default processor output shape is [B, 1, 512, 512].

Preprocessing clips HU to [-1000, 1000], z-score normalizes, selects a slice when given a volume, center-pads the slice to square, and bilinear-resizes to 512 x 512. Path inputs are loaded in LPS orientation when SimpleITK is available.

from transformers import AutoImageProcessor, AutoModel

processor = AutoImageProcessor.from_pretrained("ricklisz/FlexiCT-2D", trust_remote_code=True)
model = AutoModel.from_pretrained("ricklisz/FlexiCT-2D", trust_remote_code=True)

inputs = processor("/path/to/ct.nii.gz", return_tensors="pt", slice_index=80)
outputs = model(**inputs)
cls_token = outputs.cls_token
patch_tokens = outputs.patch_tokens

Outputs

cls_token is a global slice embedding. patch_tokens are spatial patch embeddings.

Limitations

This model is a feature extractor, not a diagnostic system. It expects CT-like HU inputs and has not been validated for clinical decision-making.