Discrepancy in Long Trajectory Extraction: Sorting by Displacement Instead of Traverse Length

[studentK2004]

While reviewing the pseudotime estimation implementation in pseudo_time.py (as of the latest version), I noticed a potential discrepancy between the described methodology in the CellDancer paper and the current code behavior regarding the selection of long trajectories.
According to the paper (Step 5: Long Trajectory Extraction and Pseudotime Assignment), the traverse length of trajectories should be computed as the accumulated distance $ \sum_t || \xi (t + \Delta t) - \xi (t) || $, and long trajectories $ {L_k(t)}_{k=1,\ldots,m} $ should be selected based on this metric, with the longest trajectories iteratively chosen and similar ones eliminated within a cutoff.
However, in the current implementation:

The compute_trajectory_displacement function (line 38) calculates the straight-line (Euclidean) distance from the start to the end point of each trajectory.
Trajectories are sorted using this displacement metric (line 595: order = np.argsort(traj_displacement)[::-1]), rather than the cumulative traverse length provided by compute_trajectory_length (line 41).
The extract_long_trajectories function (line 607) then proceeds with this sorted order, potentially prioritizing straighter paths over those with significant detours or loops.