What is QuPath?
QuPath is a cross-platform software application designed for bioimage analysis - and specifically to meet the needs of whole slide image analysis and digital pathology.
QuPath aims to be user-friendly without compromising on power and flexibility, so that it can handle tasks at all levels of complexity: from manually drawing regions and counting structures, to automatically detecting and interactively classifying hundreds of thousands of cells in huge images many gigabytes in size.
QuPath is short for Quantitative Pathology.
The QU also stands for Queen's University Belfast, where QuPath was first developed.
QuPath is designed to meet the diverse needs for a range of users, including both researchers in biology or pathology (who need the analysis results) and algorithm developers (who want a platform on which to develop, run, test and share their algorithms).
QuPath's primary use to date has been in high throughput biomarker analysis in immunohistochemically stained Tissue Microarrays (TMAs) for cancer research - incorporating fast cell detection with powerful tumor recognition algorithms to greatly speed up the evaluation of thousands of tissue samples. It even incorporates survival analysis tools to link this back to clinical data to help quickly uncover the prognostic and predictive roles for each biomarker.
However, QuPath has also been used in a range of other studies: from tumor recognition in H&E tissue sections, to manual and automated cell counting for both brightfield and fluorescence data, to tracking how pathologists view digital slides.
But it doesn't end there. QuPath has been designed to be both flexible and extensible, and can be readily adapted to new applications - either by applying the built-in functionality in new ways, or writing new scripts or extensions to add new features.
If you're undecided whether QuPath is for you, here are some reasons why it might be.
There are many fantastic bioimage analysis software applications out there: ImageJ, Fiji, Icy and CellProfiler are some of the most popular.
However, none of these are ideally suited to the specific demands of analyzing huge 2D images for digital pathology, which are often up to 40 GB in size uncompressed.
This is where QuPath comes in. QuPath's fast, multi-threaded whole slide image viewer and efficient, hierarchical data model is designed to handle such images easily. What's more, QuPath can be used in combination with other applications where necessary, to get the best from all of them.
Many of the popular open source bioimage analysis tools have a focus on microscopy - especially fluorescence microscopy.
QuPath has a focus on the brightfield images more common in routine pathology, including both H&E staining and IHC.
Although the focus is currently on brightfield imaging, QuPath has been designed to handle fluorescence data as well - especially when used in combination with ImageJ.
In time, these fluorescence capabilities will improve. But they are already developed enough to be worth trying out, if you want to take advantage of QuPath's many other features when working with fluorescence data.
Machine learning is becoming increasingly important in bioimage analysis. However, the tools to apply it can be hard to use for non-experts - and it is difficult to visualize what they are doing, or correct any errors.
QuPath tackles this by making it possible for end users to quickly and interactively classify cells or other structures within an image, simply by drawing around some examples and assigning the correct classes to them. This allows a user to teach QuPath how to distinguish, for example, tumor cells from immune cells, and watch as QuPath instantly applies this knowledge across a whole slide - color-coding all the cells to display their classifications at a glance.
For all users, this provides a flexible and extremely powerful way to apply machine learning to a wide range of applications. For those who still want more, QuPath's support for diverse machine learning libraries and varying features and parameters offers a means to get into the details of the classification and refine the methods - while still retaining user-friendliness and visual feedback.
One way to use QuPath is to detect some structures and export summary measurements. But for anyone wanting to look below the surface, QuPath's hierarchical, object-based data model allows much more. It not only enables a detailed investigating of the properties of any individual object (e.g. a cell nucleus) just by double-clicking on it, but objects can also be instantly color-coded according to their classifications or measurements to give 'heat-map' representations that give a visual overview of how properties change across an image. Furthermore, it's possible to select, interrogate and manipulate objects based upon their specific properties to get exactly the result that you need.
It may be that you already have your favorite image analysis tools, and you don't really want to learn a new one. The good news is: QuPath can integrate with some of the most popular tools available - and be adapted to integrate with others.
See the Data exchange section for more details.
Many of the biggest challenges in digital pathology come down to developing powerful techniques based upon machine learning, which can be applied to whole slide images. But handling these images is difficult. QuPath gives the tools to address this - both in terms of visualizing the data, and in creating ground-truth labels.
See the Grand Challenges website for some examples of open challenges.
In general, not every part of an huge image is interesting. Experienced pathologists can glance across a low-powered view of the image, and quickly identify the most relevant parts to look at more closely at high magnification.
Some studies have looked at slide viewing behavior to explore what can be learned from that - either to develop better analysis methods that better mimic an experienced human, or to look for differences between how people of different levels of experience view images.
QuPath incorporates optional view tracking to help with studies like these. See the Viewer tracking section for more details.
It's free & open source - you can make it do something else if you like!
The easiest way to start is by scripting. See the Automation section to get started, and in particular Advanced scripting with IntelliJ to find out how to set up scripting with all the benefits of a full IDE.
One of the primary uses of digital pathology is in cancer research, with the goal of ultimately improving patient care. One way to help researchers in this area is to help improve the software tools they used - including QuPath.
Like many open source bioimage analysis tools, QuPath has been developed by researchers within a university, with limited time, resources and experience in software design. It could benefit hugely from the input of others: be that in the form of filing bug reports or suggesting code improvements, or the development of new extensions to add entirely new functionality.
For the sake of (some) balance, here are some reasons why you might want to use something other than QuPath for your image analysis:
- You want commercial-grade support & quality assurance, or are looking for approved analysis for clinical diagnostics. Then you almost certainly want a commercial software product instead. QuPath is not developed for this purpose, but has rather been designed as part of a university research project, and is made available to help other researchers.
- You don't need whole slide support or QuPath's cell detection classification features. In this case, ImageJ or Icy might be more for you. If you want an alternative focussed on machine learning, be sure to check out ilastik.