Parameter sweep
eSRRF provides an explorative reconstruction parameter search by scanning through a range of reconstruction parameters. The parameters that can be scanned are the radius, the sensitivity and the number of frames. The range can be set using the main graphical user interface (GUI), as shown below.
Upon completion of the sweep, it will return a stack with all the reconstructions tested. This can be easily inspected. Additionally, the output of the sweep can provide maps that represent how the Resolution-Scaled Pearson (RSP) coefficient and the Fourier Ring Correlation (FRC)- based resolution vary across the parameter range, respectively representing the image fidelity and the image resolution of the eSRRF output. You can choose which output to visualise in the Advanced GUI:
The sweep maps are then displayed as images where the horizontal axis represents the radius range and the vertical axis the sensitivity range (check image calibration for what each pixel represents). The number of frame tested varies between the slices of the sweep map stacks.
The FRC is a quantitative metric that allows to determine the image resolution of an eSRRF reconstruction. The Resolution-scaled Pearson correlation coefficient (RSP) serves as a metric for structural discrepancies between the reference and super-resolution images, which is also referred to as image fidelity.
The FRC and RSP sweep map output can be used to calculate the Quality and Resolution (QnR) metric map, combining the two image quality metrics to find a compromise between image fidelity and resolution.
The intention here is to provide quantitative metrics to guide the user towards an informed decision while keeping the human-in-the-loop. It is very important to consider as a user that, while the QnR map can directly highlight optimal reconstruction settings by indicating the maximum QnR parameter, local variations in background level, emitter density, and sample structure across the field of view can cause different reconstruction requirements and non-linearities in the QnR-maps.
Therefore, a critical analysis of the reconstruction results of the sweep range by the user is mandatory.
By guiding users through quantitative optimization, eSRRF aims to reduce bias relative to manual parameter tuning. Keep in mind that FRC, RPC, and QnR are supporting you in selecting parameters that balance resolution and artifacts, however depending on the data and question at hand you might want to prioritise specific subregions of the image. We recommend saving the sweep results and reporting the sweep range and the chosen reconstruction parameters.
You can run the plugin on corresponding sweep maps and see what parameter range is most appropriate for a particular dataset. The QnR score is normalised between 0 and 1 and the highest value represents the most balanced compromise between fidelity and FRC resolution.
For the calculation of the QnR map, the FRC resolution map needs to be normalised. For this, the plugin uses a logistic conversion of the map using the FRC resolution min and max from the user interface. The automatic setting of min and max here will use the global minimum and maximum from the whole stack of FRC maps.
If the pixel size was set in the raw dataset used for the reconstructions, then the FRC resolution map is expressed in micrometer. Otherwise, it will be in number of pixels. The rescaled (normalised) FRC map can be displayed as well as a logistic conversion curve used for the conversion.
