ctp-skill

Run the ctp.py compute mode with a motion-corrected 4D HU NIfTI series, acquisition metadata JSON, brain mask, and AIF mask. It performs baseline subtraction and block-circulant truncated SVD deconvolution, writing the four maps plus ROI summaries and QC files.

Use motion_correct.py with the 4D NIfTI series and metadata containing baseline frame indices. It rigidly registers each frame to the first precontrast frame using SimpleITK mutual information, then outputs the corrected series, coverage mask, and per-frame displacement metrics for manual review.

No. The skill is explicitly scoped to CT perfusion with HU-based delta-signal curves and does not apply the logarithmic MR transform. For ASL or DSC perfusion MRI, use a modality-appropriate processing pipeline instead.

Inputs must be 4D NIfTI in mm spatial units with HU values, plus a JSON sidecar containing frame_times_s in seconds, at least two initial baseline_frames, and a motion_corrected flag. Masks must be binary 0/1 NIfTI on the exact same grid and affine.

The deconvolution requires uniform time spacing for the convolution matrix. If frame times are irregular, explicitly pass --resample-dt in seconds after reviewing sampling; the script interpolates without extrapolating and rejects non-uniform grids otherwise.

No. The maps are research reference implementations without clinical validation, leakage correction, or diagnostic thresholds. They must not be used to label ischemic core, penumbra, treatment eligibility, or patient outcomes.