Listing standards on inspecting components and reclassification

[handwerkerd]

Summary

At the Nov 2022 dev call #897 we brought up the idea of adding instructions to our documentation on how to visually inspect components and consider reclassification.

[dowdlelt]

This paper will probably end up being relevant: https://www.sciencedirect.com/science/article/pii/S1053811916307583

[Shotgunosine]

Related to this. If the interactive figure used in the report would support it, some keyboard shortcuts for classifying components and saving those classifications would be fantastic as well.

Edit: duh, the power spectra are already there, my mistake. Keyboard shortcuts would be great. Edit 2: Ah, I see there's a rica package for doing what I was thinking about for classification of components: https://github.com/ME-ICA/rica

[eurunuela]

Glad you found it @Shotgunosine ☺️

By the way, we welcome contributions to Rica, so feel free to share feedback or open a pull request.

[tsalo]

Some features that are useful for manual classification that we may want to account for in tedana or rica:

• Overlay thresholded components on a high-resolution anatomical image.
• Time series plots showing the components' time series and confound time series (esp. motion parameters).

Here is Table 1 from Griffanti et al. (2017), for reference:

Features	S-IC characteristic	N-IC characteristic
Spatial
Number and dimension of clusters	Low number of large clusters	Large number of small clusters
Overlap with GM	Clusters’ peaks in GM and overall good overlap of the clusters with GM.	Indiscriminate overlap with non-GM tissues, or clusters’ peaks in WM/CSF
Overlap with WM, CSF, blood vessels	Very low or absent overlap with WM, CSF, blood vessels	High overlap with one or more of WM, CSF, blood vessels
Overlap with brain boundaries or areas close to the edges of the FOV.	Very low or absent overlap with brain boundaries. Clusters follow known anatomical (e.g. structural/ histological) boundaries.	Ring-like or crescent shape or stripes near the edges of the field-ofview
Location near area of susceptibility induced signal loss (e.g. orbitofrontal)	Generally located away from these areas	Located within the region of signal loss (e.g. areas of air-tissue interface)
Non-biological, acquisition-related patterns	Patterns have no relation to acquisition parameters	Often show banding patterns in slice direction or streaks along the phase encoding direction, accelerated sequences may have centrally located artefacts
Temporal (and spectral) features
Overall aspect of the time series	Fairly regular/oscillatory time course	Large jumps and/or sudden change of oscillation pattern.
Distribution of power in frequency domain	Predominantly low frequency (at least one strong peak within 0.01 – 0.1 Hz)	Predominantly high frequency, very low frequency, or pan frequency