|
| 1 | +# HiP-CT data |
| 2 | + |
| 3 | +HiP-CT stands for hierarchical phase-contrast tomography. |
| 4 | +For more information on the modality see the [HiP-CT website](https://mecheng.ucl.ac.uk/hip-ct/). |
| 5 | +Most of the time HiP-CT consists of a single 'low' (but still quite high!) resolution scan of a whole sample, and then one or more 'high' resolution scans of particular volumes of interest within the same sample. |
| 6 | + |
| 7 | +## Data acquisition and naming |
| 8 | + |
| 9 | +There are HiP-CT scans of two donors, S45 and I58. |
| 10 | +The data are part of the [Pilot human brains dataset](https://lincbrain.org/dandiset/000005) on lincbrain. |
| 11 | + |
| 12 | +Datasets have the name `sub-<donor>_sample-<sample name>_chunk<chunk number>_PC.ome.zarr`. |
| 13 | +The chunk number is set to `yyxx`, where yy is the resolution of the scan in micrometers, and xx is a scan number that is unique between datasets. |
| 14 | +For example, a chunk number of `402` is the second 4um scan. |
| 15 | +The lowest resolution dataset (equivalently, highest chunk number) is the overview dataset that contains an image of the whole sample. |
| 16 | +There are then one or more datasets at higher resolution. |
| 17 | + |
| 18 | +## Registration |
| 19 | + |
| 20 | +Each high resolution dataset is registered to the overview dataset using [`hipct-reg`](https://github.com/hipCTProject/hipct-reg). |
| 21 | +Each transform maps the pixel coordinates of the high resolution zoom datasets to the overview full-sample dataset. |
| 22 | +The transform is of the form: |
| 23 | + |
| 24 | +$$ |
| 25 | +T(\mathbf{x}) = s \left ( \mathbf{R} \cdot\mathbf{x} \right ) + \mathbf{t} |
| 26 | +$$ |
| 27 | + |
| 28 | +Where |
| 29 | + |
| 30 | +- $s$ is a scaling factor |
| 31 | +- $\mathbf{R}$ is a rotation matrix that rotates about the z-axis. |
| 32 | +- $\mathbf{t}$ is a translation vector |
| 33 | + |
| 34 | +The rotation matrix takes the form |
| 35 | + |
| 36 | +$$ |
| 37 | +\mathbf{R} = \begin{pmatrix} |
| 38 | +\cos(-\theta) & \sin(-\theta) & 0 \\ |
| 39 | +-\sin(-\theta) & \cos(-\theta) & 0 \\ |
| 40 | +0 & 0 & 1 \\ |
| 41 | +\end{pmatrix} |
| 42 | +$$ |
| 43 | + |
| 44 | +The transform has 5 parameters (1 scaling, 1 rotation, 3 translation). |
| 45 | + |
| 46 | +### Registration accuracy |
| 47 | + |
| 48 | +The HiP-CT registration pipeline is designed to be accurate enough to approximately visually align two datasets. **It is not designed for pixel perfect registration.** We recommend anyone doing analysis relying on the registration between two datasets performs a more accurate registration themselves, using our registration as an initial starting point. |
| 49 | + |
| 50 | +### Registration data |
| 51 | + |
| 52 | + |
| 53 | +| Dataset | Overview dataset | $\mathbf{t}$ | $s$ | $\theta$ (degrees) | |
| 54 | +| -------- | -------- | -------- | -------- | -------- | |
| 55 | +| `sub-I58_sample-blockIC2_chunk-101_PC` | `sub-I58_sample-blockIC2_chunk-401_PC` | (3746.4467746089194, 3660.0610535229666, 1127.3616712249263) | 0.2016106611944697 | -6.496401244495998 | |
| 56 | +| `sub-I58_sample-blockIC2_chunk-201_PC` | `sub-I58_sample-blockIC2_chunk-401_PC` | (2119.695989239648, 2321.6572801771194, 971.0041825720597) | 0.5181803632402424 | 0.5004912711693281 | |
0 commit comments