Jump to content

CryoNAV Tilt Image Exclusion

From CryoNAVwiki
Revision as of 22:57, 20 May 2026 by Admin (talk | contribs) (Bot edit via cryonavedit.py: add reference pages and screenshots)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

In cryo-electron tomography, individual tilt images may be unusable due to ice contamination, beam-induced drift, tracking failures, or unreliable CTF fits. CryoNAV supports per-step frame exclusion: bad views are flagged (never physically removed) and propagated through the pipeline so that alignment, CTF estimation, and reconstruction skip them consistently.

Core principle: monotonic superset exclusion

Each processing step stores its own excluded_views parameter. Every downstream step's exclusion set must be a superset of the upstream step's -- views can be added but never un-excluded. The full stack is preserved on disk; exclusion is implemented via IMOD flags only.

To un-exclude a view, the user must re-run from the step where it was first dropped, invalidating all downstream products.

Per-step exclusion points

CryoNAV maps each pipeline step to its native IMOD flag:

  • coarse_alignment (tiltxcorr) -- flag -SkipViews. First exclusion point; typically used for obvious ice contamination or drifted views spotted in the pre-aligned stack.
  • patch_tracking (tiltxcorr xcorr_pt) -- flag -SkipViews. Add views where patch tracks fail.
  • bead_tracking (beadtrack) -- flag -SkipViews. Add views where fiducial beads are lost.
  • seed_finding (autofidseed) -- no flag available. Exclusion is handled downstream by beadtrack.
  • fine_alignment (tiltalign) -- flag -ExcludeList. Drop views with large residuals.
  • ctf_estimation (ctfplotter) -- flag -ViewsToSkip. Drop views with unreliable defocus fits. Optional linear step.
  • ctf_correction (ctfphaseflip) -- no exclusion (processes views independently).
  • tomogram reconstruction (tilt) -- flag -EXCLUDELIST2. Final opportunity, e.g. to drop extreme tilts.
  • denoising half-reconstructions -- flag -EXCLUDELIST2. Inherits independently per endpoint.

CTF as an optional linear step

CTF estimation sits between fine alignment and the endpoints per-endpoint. The upstream constraint for each endpoint is resolved as:

  • Tomogram -- if CTF is in its path, inherits from ctf_estimation; otherwise from fine_alignment.
  • Denoising -- same rule, evaluated independently.

Most common configuration: tomogram uses CTF (inherits CTF exclusions), denoising skips CTF (inherits only from fine alignment). A view dropped purely for a bad CTF fit need not be excluded from a denoised half that bypasses CTF.

Why exclusion matters downstream

  • Alignment quality -- leaving in drifted or contaminated views inflates residuals and biases the tilt-axis solution.
  • CTF reliability -- views with poor Thon-ring fits, if not skipped, contaminate the per-view defocus model.
  • Reconstruction artifacts -- including bad views adds streak/ray artifacts in the back-projected volume.
  • Reproducibility -- because exclusion is flag-based and stored per step, every reconstruction can be regenerated exactly from the recorded excluded_views plus the original stack.

UI convention

In the frame browser, each view is shown as: green (included), grey/locked (excluded upstream, inherited), or red (excluded at the current step). Submission is rejected if the user's set is not a superset of the upstream set.

See also