This is the compiled form of the SignalFinder Extension for QuPath.
The GUI is based off of the Instaseg GUI (https://github.com/qupath/qupath-extension-instanseg/tree/main)
Adaptive thresholding uses an extremely fast KD-Trees K-Nearest Neighbors algorithm developed in Java by "Rednaxela" (https://gitlab.com/agschultz/robocode-knn-benchmark/-/tree/master/ags/utils/dataStructures/trees/thirdGenKD)
As the name implies, SignalFinder is a tool for finding signal (thresholding) in immunoflourescense images with respect to background.SignalFinder works through the use of image texture analysis theory, relying patterns in the statistical variation of small regions of image in common between background regions of different IF images. Notably, while mean, median, and standard deviation may vary in background regions between images, coeffecient of variation (CV) is consistent.
The SignalFinder Extension is based off of the SFT algorithm (doi.org/10.1021/acs.analchem.5b03159)
Unlike the original SFT algorithm, however, the SignalFinder extension does not using a sliding window analysis to sample the image statistics. Rather, it uses the cell measurements(mean, sd, median) from cell compartments (Nucleus, Cytoplasm, and Membrane) to sample the image statistics. This approach has been validated to give results with a high degree of agreement with the original algorithm. Also compared to the standalone SignalFinder software, this extension runs much faster, taking less than a minute to threshold 30 channels for a million-cell dataset and less than hour if calculating cell measurements for the same.
Image thresholding has long been a tool in the image analyst's toolbelt, though in many cases thresholds are determined manually (by eye). Automated thresholding options such as Otsu's method and the Triangle method are popular but are known to fail when the distribution of signal fails to meet the assumptions of the method. SignalFinder, on the other hand, has more robust assumptions about signal, only requiring some amount of background to be present in the subject of the image.This extension in particular, allows you to measure cell fractions for each marker - that is the area of the cell positive for signal. This allows for interrogation of signal density which is only indirectly measured with continuous measures of signal (mean, median, sd, max, and min). In this way it is possible to distinguish diffuse, moderate signal from strong, punctate signal which would be mathematically identical for continuous measures.
Installation is easy, simply download the .jar file and then drag+drop from your file explorer into QuPath. There are no other dependencies. The SignalFinder extension assumes that you have cell detections with mean, median, and standard deviation measurements in the format of "Compartment: Marker: Measurement" like those generated by InstanSeg.To run simply select "Extensions">"SignalFinder">"Run SignalFinder"
This is the pre-release form of SignalFinder, we're actively working on publication. In the meantime, cite the original SFT algorithm and this GitHub (doi.org/10.1021/acs.analchem.5b03159).CC BY-NC-4.0 Contact Van Andel Intitute for commercial licensing options.