HistoMetriX was evaluated on a publicly available skin histology dataset with expert annotations. Automated epidermal thickness measurements were compared with the reference measurements across the annotated images.

Segmentation accuracy was evaluated separately for each compartment using positive predictive value (PPV), sensitivity, Dice score and Matthews correlation coefficient (MCC). Results are reported as mean ± standard deviation, with the median in parentheses, across the 8 test slides.

Three test slides were scored within their predefined regions of interest. Oskal et al. evaluated 33 H&E slides from other laboratories using their own post-processing. This comparison therefore concerns the reported protocols, not identical images.

The epidermis reached a Dice score of 0.94 ± 0.02, with a minimum of 0.92 across the test slides. The stratum corneum was the most challenging compartment because it is thin and frequently detached. Its PPV remained high at 0.95, but some thin fragments were missed, resulting in a sensitivity of 0.75.

Two measurements were calculated on each test slide using both the HistoMetriX segmentation and the expert masks: the length of the dermal-epidermal junction, corresponding to the epidermis-dermis boundary, and the mean epidermal thickness, calculated as the epidermal area divided by the junction length.

On a typical slide, the junction length is within 5.4 % of the expert masks and the mean thickness within 3.5 % (medians). The means are pulled up by WC23-00492-A1 (junction +22 %, thickness -19 %): a few sweat ducts of its dermis are segmented as epidermis, and each adds its own small ring of junction, which lowers the mean thickness. On the other 7 slides, the junction length differs by at most 15 % and the thickness by at most 10 %.

Analysis time was measured from the start of the pipeline to the writing of the results. Across the 8 test slides, HistoMetriX processed 103 mm² in 326 seconds, corresponding to 5.3 minutes per cm² overall. Each slide adds a fixed processing time of approximately 20 to 30 seconds, so larger sections are processed more efficiently: 2.9 minutes per cm² for the three sections larger than 15 mm². The timed steps included compartment segmentation and area measurements, without cell detection. Tests were performed on an AMD Ryzen 9 7900 with 64 GB RAM and an NVIDIA GeForce RTX 5060 Ti with 16 GB memory.

The evaluation included 8 test slides from a single dataset and a single stain (PAS). These results therefore describe this dataset and should not be generalized to every staining protocol, species or sample type. The model was trained on 24 slides from the same dataset, and three test slides were evaluated within predefined regions of interest. The stratum corneum may be under-segmented when it is thin or detached, while some skin appendages can occasionally be classified as epidermis.

Salam, A.A., Akram, U., Asaf, Z., Rao, B. and Khan, S. DERMA-PASeg. Mendeley Data, Version 1 (2024). https://doi.org/10.17632/w8vxx8yz55.1

Oskal, K.R.J., Risdal, M., Janssen, E.A.M., Undersrud, E.S. and Gulsrud, T.O. A U-net based approach to epidermal tissue segmentation in whole slide histopathological images. SN Applied Sciences, 1, 672 (2019). https://doi.org/10.1007/s42452-019-0694-y

Beyond the standard compartment segmentation, HistoMetriX can generate depth-based subregions or use custom models trained on annotated images to identify specific skin structures. Measurements can then be reported separately for each sub-compartment.

HistoMetriX detects nuclei within the dermis and measures their distance to the epidermis. Nuclei can then be grouped into distance bands or analyzed along a continuous gradient to compare superficial and deeper dermal populations, quantify marker positivity or assess inflammatory cell infiltration.

The same compartment-based workflow can be applied to fluorescence and in situ hybridization images. The epidermis and dermis can be identified using dedicated markers or, when tissue morphology allows it, directly from the nuclear channel. Signals from ISH, FISH or DISH can then be quantified separately within each compartment.

Automated segmentation of skin compartments and quantitative epidermal morphometry are established fields of computational pathology. The scientific context of the HistoMetriX skin pipeline includes:

• Automated identification of skin layers in histology images: Brindise et al., Annals of Biomedical Engineering, 2023. DOI: 10.1007/s10439-022-03106-9. pubmed.ncbi.nlm.nih.gov
• Deep-learning segmentation of the epidermis in whole-slide images: Oskal et al., SN Applied Sciences, 2019. DOI: 10.1007/s42452-019-0694-y. link.springer.com
• Validation of epidermal segmentation and morphometric measurements using the public DERMA-PASeg dataset. DOI: 10.17632/w8vxx8yz55.1. data.mendeley.com
• Color deconvolution based on established brightfield histology quantification methods: Ruifrok and Johnston, 2001. PMID: 11531144.

Do I need to train a model or annotate images?

No. The Skin Morphology pipeline is ready to use. Load your slides, select a region mode and run the analysis. No parameter adjustment or model training is required for most studies. For specific stains, pathologies or species, or to identify finer sub-compartments, the models can be fine-tuned using your own annotated images with Train AI.

How is epidermal thickness calculated?

HistoMetriX measures the epidermal area, the length of the epidermis along its midline and the length of the dermal-epidermal junction. Mean epidermal thickness is calculated by dividing the epidermal area by either length. Both values are available for each sample, field and batch.

Can I reproduce the standard 10-field protocol?

Yes. In Mode 3, draw one region of interest, duplicate it and place the copies along the epidermis. Results are reported for each field and can then be averaged for each slide.

What if a slide contains several samples?

Mode 2 automatically detects and labels each sample. Before running the analysis, you can remove a region, reshape an outline or rename a sample.

Can I analyze cells according to their distance from the epidermis?

Yes. Each detected cell can be classified according to its distance from the epidermis. For example, dermal cells located within 100 µm of the epidermis can be analyzed separately from cells located deeper in the dermis.

Which samples and stains are supported?

HistoMetriX can analyze human skin biopsies and animal skin, including pig, rabbit, rat and mouse samples. Supported stains include H&E, HES, Masson’s trichrome, PAS, IHC with DAB, fluorescence, ISH, FISH and DISH. Contact us to confirm compatibility with a specific sample type or staining protocol.

Which file formats can I open?

HistoMetriX supports SVS, NDPI, TIFF and MRXS whole-slide images.

Where does the analysis run?

HistoMetriX is a Windows desktop application. Analyses run locally on the user’s computer and do not require image data to be uploaded to the cloud.

Can HistoMetriX be used for diagnosis?

No. HistoMetriX is a research-use-only (RUO) tool. It is not intended for clinical diagnosis or patient management decisions.

How can I try it?

Request a trial licence and run the pipeline on your own skin slides.