QPSC: Controlling Microscopes from QuPath

I've been building QPSC (QuPath Scope Control), an open-source system that lets researchers control their microscope directly from QuPath. The core idea is simple: draw a box around a region of interest in QuPath, and the software handles the rest. It moves the stage, captures tiles, stitches everything into a pyramidal image, and drops the result right back into your QuPath project.

The motivation came from a real gap in the pathology research workflow. Researchers identify regions of interest in scanned whole-slide images, then require restaining or alternate types of imaging in order to clearly visualize collagen presence and structure. QPSC connects QuPath's annotation tools to live microscope hardware through Micro-Manager and Pycro-Manager, enabling high throughput imaging and collection of multiple modalities within a single microscope, with integrated project metadata to make sorting related images/samples and positions easier.

The PPM Focus

We are working in collaboration with Dr. Agnes Loeffler from Cleveland MetroHealth, who has been using PPM for several years now, with an entirely manual microscope. That collaboration has already demonstrated real clinical value, including improved detection and differentiation of crystals in joint fluids and the ability to differentiate chronic pancreatitis from invasive pancreatic adenocarcinoma on human patient samples. Despite these successes and several published articles, the complete lack of automation makes it difficult to scale this to the kind of studies that would be needed to ensure an impact with the larger pathology community.

QPSC automates the entire PPM acquisition sequence: rotating the polarizer, adjusting per-angle exposure times (which can vary by orders of magnitude between crossed and uncrossed positions), and collecting the multi-angle image stacks needed for quantitative analysis. For labs using prism-based cameras like JAI sensors, the system also handles per-channel white balance calibration so that hardware exposures are tuned independently for each color channel.

A standalone Python library, ppm-library, handles the image processing and analysis side. It includes debayering, flat-field correction, and a calibration pipeline that uses a sunburst reference slide to map hue values to actual fiber orientation angles. The end result is quantitative angle maps of collagen organization, not just pretty pictures.

Clinical Applications

The bigger picture is computational PPM with improved color metrics and fiber sensitivity as a tool for clinical pathology. Being able to convert human tissue samples into digital PPM images and reproducibly compute collagen signatures across entire slides opens the door to prognostication in a wide variety of malignancies. We've identified three areas where we think this can make the most impact:

Tumor-associated collagen signatures (TACS) have proven biomedical significance in malignant tumors, but TACS hasn't been adopted for clinical use because the histologic context of collagen alignment in normal tissue and benign processes like scarring and inflammation is poorly understood. Automated whole-slide PPM imaging would let researchers systematically compare collagen alignment patterns across benign and malignant tissues at a scale that isn't feasible with manual microscopy.

With the current PPM tool, we already see a range of signal intensity across different types of collagen. Improving the system's sensitivity to early collagen associated with cancer invasion has direct clinical relevance, particularly for situations like intramucosal colorectal carcinoma where early cancer is difficult to detect on routinely stained (H&E) tissues alone.

Since PPM highlights collagen differences between chronic pancreatitis and invasive pancreatic adenocarcinoma, we're also working toward detecting collagen signatures characteristic of pancreatic adenocarcinoma in fine needle aspiration (FNA) samples. Preliminary data show that even in tiny FNA samples (around 5 mm2), there is enough stroma to detect collagen signatures with PPM. An automated system could improve the diagnostic sensitivity of FNA by accurately analyzing the collagen component of these small tissue samples during the workup of radiologically detected pancreatic masses.

What's Built

The system is modular, split across a QuPath extension (Java), a microscope command server (Python), a hardware abstraction layer (Python/Pycro-Manager), and the PPM library mentioned above. Here's what works today:

Bounding box and existing-image acquisition workflows with automatic tiling, stitching (to OME-TIFF or OME-ZARR), and QuPath project import. Multi-angle PPM acquisition with automatic polarizer rotation and per-angle exposure control. Calibration tools for white balance, polarizer alignment, autofocus tuning, and flat-field background collection. A pluggable modality system so new imaging modes (brightfield, SHG, fluorescence) can be added without rearchitecting. A stage map window with real-time position tracking, click-to-navigate, and macro image overlay. Multi-sample project support that tracks image collections, physical offsets, and coordinate system metadata across a whole project.

What's Next

On the near-term roadmap: improved computational PPM with better color metrics and fiber sensitivity to drive the clinical applications described above, deep learning pixel classification integrated into QuPath with a Python backend for training and inference, live preview scanning to rapidly populate the stage map with a low-res overview of the entire slide (in the case where no slide scanner overview image is available), SHG and multiphoton modality support for labs doing second harmonic generation imaging, and a publication describing the system and demonstrating it on real tissue samples.

The project is pre-release and actively developed at UW-Madison LOCI. All code is on GitHub.