MIKAIA

Fraunhofer’s commercial digital-pathology application, installed locally — and worth knowing precisely because its strongest features are the two things the group’s QuPath stack is weakest at: ISH quantification and multiplex spatial analysis.

Purpose

MIKAIA® is a Windows desktop application from the Fraunhofer Institute for Integrated Circuits IIS, Erlangen — identity and version taken from Doc/changelog.txt and confirmed against the executable’s own file-version metadata, 3.0.0. What is on the drive is an installed application, not source.

The decisive question is not what it does but what it does that the existing stack does not, and for most of its feature list the answer is “nothing” — tissue detection, tile export, annotation metrics and cell counting are all covered by QuPath Tooling. Three areas are genuinely different:

  • ISH quantification. A CISH App, an FL Cell AI / FISH app and a FISH Ratio App. Chromogenic and fluorescent in-situ hybridisation scoring is a real weak spot in the QuPath route, and it is not hypothetical here — HER2 Intratumoral Heterogeneity is built on gastric cases with HER2 IHC and reflex ISH, and the ISH half has no tooling recorded anywhere in this wiki.
  • Multiplex and spatial analysis. A Cell × Gene App (spatial transcriptomics), a Cellular Neighborhood App and a Cell–Cell Connections App. These are neighbourhood and interaction analyses over segmented cells, which is exactly the class of question Spatial Proteomics describes and which the group currently has no tool for at all.
  • Slide alignment across stains, offered for multi-omics work — the commercial counterpart to the problem on Cross-Stain Registration.

It also reads an unusually broad set of vendor formats: DICOM-WSI including sparse multi-file and 3DHistech DICOM without a file extension, Hamamatsu MoxiePlex and NDPIS, Olympus VSI in brightfield and 16-bit fluorescence, Akoya QPTIFF with channel names read from the biomarkers, Zeiss CZI multi-scene, and Roche .bif. For a group whose slides come off more than one vendor’s scanner that is not a trivial capability.

Data used

None held here — it is an application. No cohort, no patient data in the install tree.

Methods

Not source code, so there is nothing to read the way a repository is read. Capability was taken from Doc/changelog.txt, which enumerates the analysis modules MIKAIA calls Apps: Segmentation AI Author, Classification AI Author, Universal IHC Cell AI, CISH, FL Cell AI/FISH, FISH Ratio, Cell × Gene, Cellular Neighborhood, Cell–Cell Connections, Annotation Metrics, Tissue Detection, Mask-by-Color, Tile Export, Slide Alignment, and a generic “Plug in your own AI App”.

One correction worth recording, because the obvious shortcut gives the wrong answer. The install has a plugins/ directory, and it contains nothing about analysis: bearer/, iconengines/, imageformats/ and platforms/ are Qt’s own framework plugins — image codecs and platform integration DLLs bundled with any Qt application. Reading capability off that folder would produce a confidently wrong list. The Apps above are the real capability surface and they are named only in the documentation.

It has a real automation surface, which is what makes it a candidate rather than a curiosity. “Plug in your own AI App” is described as a REST/Python API for registering custom AI scripts as MIKAIA Apps, with later versions adding commands for result-panel captions, CSV export and diagrams. Separately a MIKAIA Worker REST API (renamed from the earlier CLI) can register MIKAIA Apps as plugins inside Proscia Concentriq. So it can be driven from outside rather than only used through its own interface — meaning it could sit alongside the QuPath workflow rather than replacing it, which is the only integration shape that would realistically be adopted here.

Licensing is via a CodeMeter runtime, i.e. a dongle or licence server. [unverified] beyond that, and not investigated further.

Current state / open questions

Not adopted, and the right verdict for now is “evaluate against one real question rather than in general”. A general comparison against QuPath would be a long exercise with a predictable answer — QuPath wins on cost, scriptability and the fact that every workflow here already runs on it. The narrow question is worth answering:

  • Can the FISH Ratio or CISH app score the HER2 reflex-ISH slides in HER2 Intratumoral Heterogeneity? That cohort exists, the ISH is already done, and nothing in the current stack quantifies it. One cohort, one app, a comparison against the reported manual scores — that is a bounded test with a clear answer, and it is the only one of these features with an immediate use.
  • Does the Cell × Gene / Cellular Neighborhood pair do anything the group would otherwise have to build? Spatial Proteomics identifies neighbourhood analysis as a capability gap, and this is the only tool on the premises that claims it. [unverified] whether it handles the marker-channel input format that page describes.
  • The 25 MB MIKAIA Manual.pdf in Doc/ was not read, so the REST API’s actual contract is unknown. That is the document to read before any integration is attempted, and it would settle whether the API can be driven headlessly from an existing pipeline.
  • Whether the installed licence is active, and on what terms, is not recorded. It determines whether any of the above can be tried at all.

Related: HER2 Intratumoral Heterogeneity — the cohort with reflex ISH and no ISH tooling; the one bounded test worth running. Related: Spatial Proteomics — the neighbourhood and cell-interaction analyses MIKAIA claims and the group currently cannot do. Related: Cross-Stain Registration — its slide-alignment feature is the commercial counterpart. Related: QuPath Tooling — the incumbent, which covers most of MIKAIA’s other features already; this is a supplement or not adopted at all, never a replacement. Related: Whole Slide Imaging — its vendor-format coverage is broader than the group’s current readers in several places.

Derived from: install tree read 2026-07-29 — Doc/changelog.txt for identity, version, App list, supported formats and API description; MIKAIA.exe file-version metadata (read as a file property, not executed); plugins/ directory listing. Doc/MIKAIA Manual.pdf (25 MB) and EULA.txt were not opened. Nothing was run.