Fix bandpass filter

[koenhelwegen]

This PR modifies the filtering behavior in two ways:

• it removes the matching of the initial and final timepoints of the unfiltered signals, as these are often outliers from the perspective of the filtered signal;
• it pads the signal using the the time-mirrored timeseries, not the inverted time-mirrored timeseries.

closes #27

[SiemondeLange]

Looks good! I will test this implementation on a few datasets and write a unit-test to further validate the implementation. If those are successful I think we can merge this branch.

[SiemondeLange]

After the additional discussion provided in Issue #27, I think that we are almost ready to merge this branch into the main branch. I will do some additional benchmarking to investigate the impact of this change in the functional network reconstruction step. If that is satisfactory, I will merge this branch.

[SiemondeLange]

I compared the functional connectomes obtained using CATO version 3.2.0 and the new version incorporating these code changes and found that they were very similar. The average correlation coefficient was 0.93 (SD: 0.04) for participants in the HCP test-retest dataset (aparc atlas, default parameters). Additionally, I conducted an evaluation of the impact of these code changes on the functional connectivity reconstruction reliability by following the same HCP data test-retest approach described in CATO manuscript (bioRxiv). The test-retest reliability results after implementing these code change are highly comparable to the results presented in the CATO manuscript, which were obtained using CATO version 3.2.0 (as seen the two tables below). Given the similarity of the results, I am confident in moving forward with the merging of this branch.

Table 1. Functional test-retest reliability across atlases (top 20%-proportional thresholded).	Reconstruction method	Median ICC	IQR ICC	% Poor	% Fair	% Good or excellent
aparc	0.49	0.42-0.58	22.8	56.8	20.4
lausanne120	0.53	0.45-0.61	15.7	56.5	27.8
lausanne250	0.55	0.46-0.64	13.4	51.0	35.5
economo	0.51	0.42-0.58	20.5	57.5	22.0

Table 2. Functional test-retest reliability of all connections across atlases.	Reconstruction method	Median ICC	IQR ICC	% Poor	% Fair	% Good or excellent
aparc	0.45	0.36-0.54	34.9	53.0	12.1
lausanne120	0.47	0.37-0.55	32.9	52.4	14.7
lausanne250	0.47	0.37-0.57	31.9	50.8	17.3
economo	0.44	0.32-0.53	40.9	46.5	12.6