Data Format¶

KinePET fits flattened TAC batches.

Required arrays¶

Array	Shape	Meaning
`tacs` / `tacs.npy`	`(T, N)`	TAC matrix. Each column is one TAC.
`time` / `time.npy`	`(T,)`, `(T, 1)`, or `(T, N)`	Time vector or per-TAC time matrix.
`aif` / `aif.npy`	`(T,)`, `(T, 1)`, or `(T, N)`	Arterial input function.

Optional arrays¶

Array	Shape	Meaning
`weights` / `weights.npy`	`(T,)`, `(T, 1)`, or `(T, N)`	Residual weights.

Shared and per-TAC semantics¶

For time, aif, and weights:

(T,) means one shared vector for every TAC.
(T, 1) also means one shared vector for every TAC.
(T, N) means one column per TAC.

When aif is (T, N), time must also be (T, N). Column i is fitted as time[:, i], aif[:, i], and tacs[:, i].

Time units¶

Time values are interpreted as minutes. If max(time) > 60, KinePET assumes the input is in seconds and converts to minutes internally.

Outputs¶

Python returns dataclass wrappers. The CLI writes .npy files. In both cases outputs are flat per-TAC values with shape (N,).

Main outputs:

Output	Availability
`K1`	Always
`k2`	Always
`k3`	Always
`k4`	`rev` only
`vB`	Always
`delay`	When fitted or fixed to a nonzero value
`dispersion`	When fitted or fixed to a nonzero value
`Ki`	Always
`rmse`	Always
`weighted_cost`	Python API
`iterations`	Always
`status`	Always

KinePET does not produce image maps, masks, rescue overlays, rawfit arrays, or post-filtered arrays. Image reconstruction and preprocessing should happen outside the core solver.