Dunn et al.

Diagnostic Pathology (2024) 19:42 Diagnostic Pathology

https://doi.org/10.1186/s13000-024-01461-w

RESEARCH Open Access

Quantitative assessment of H&E staining

for pathology: development and clinical
evaluation of a novel system
Catriona Dunn1,2*, David Brettle1, Martin Cockroft3, Elizabeth Keating3, Craig Revie4 and Darren Treanor1,2,5,6,7

Abstract
Background Staining tissue samples to visualise cellular detail and tissue structure is at the core of pathology
diagnosis, but variations in staining can result in significantly different appearances of the tissue sample. While the
human visual system is adept at compensating for stain variation, with the growth of digital imaging in pathology,
the impact of this variation can be more profound. Despite the ubiquity of haematoxylin and eosin staining in clinical
practice worldwide, objective quantification is not yet available. We propose a method for quantitative haematoxylin
and eosin stain assessment to facilitate quality assurance of histopathology staining, enabling truly quantitative
quality control and improved standardisation.
Methods The stain quantification method comprises conventional microscope slides with a stain-responsive
biopolymer film affixed to one side, called stain assessment slides. The stain assessment slides were characterised with
haematoxylin and eosin, and implemented in one clinical laboratory to quantify variation levels.
Results Stain assessment slide stain uptake increased linearly with duration of haematoxylin and eosin staining
(r = 0.99), and demonstrated linearly comparable staining to samples of human liver tissue (r values 0.98–0.99).
Laboratory implementation of this technique quantified intra- and inter-instrument variation of staining instruments
at one point in time and across a five-day period.
Conclusion The proposed method has been shown to reliably quantify stain uptake, providing an effective
laboratory quality control method for stain variation. This is especially important for whole slide imaging and the
future development of artificial intelligence in digital pathology.
Keywords Digital Pathology, Histopathology, Quality, Stain, Quality Assurance, Histochemical staining

4
*Correspondence: FFEI Limited, The Cube, Hemel Hempstead, UK
5
Catriona Dunn Department of Histopathology, Leeds Teaching Hospitals NHS Trust,
catriona.dunn@nhs.net Leeds, UK
1 6
National Pathology Imaging Co-operative, Leeds Teaching Hospitals NHS Department of Clinical Pathology and Clinical and Experimental
Trust, Leeds, UK Medicine, Linköping University, Linköping, Sweden
2 7
Department of Pathology and Data Analytics, University of Leeds, Leeds, Centre for Medical Image Science and Visualisation, Linköping University,
UK Linköping, Sweden
3
New Technology Group, Futamura Chemical UK Limited, Wigton, UK

© The Author(s) 2024. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use,
sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and
the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this
article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included
in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will
need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The
Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available
in this article, unless otherwise stated in a credit line to the data.
Dunn et al. Diagnostic Pathology (2024) 19:42 Page 2 of 11

Introduction [7]. Importantly, a recent study also found stain normali-

The histopathological examination of tissue is the cor- sation significantly improved pathologist perception of
nerstone of cancer diagnosis globally. It is based on the stain colour quality, diagnostic confidence, and time to
staining of tissue samples with histochemical dyes, such diagnosis [28]. However, they also found that normali-
as haematoxylin and eosin (H&E), to highlight cellular sation reduced inter-pathologist agreement. Although
components for visual interpretation by pathologists. this was only two pathologists it suggests that normali-
This process has not changed for over a century, and it is sation may improve perceived colour and pathologist
well understood that there are variations in the method confidence, but that the normalisation process may be
[1–5]. Staining variation is widely seen in clinical practice a variable in its own right that could negatively impact
in pathology, both within and between laboratories [5–7]. upon inter-observer agreement. Stain normalisation can
Although not often highlighted as a clinical risk, detailed improve image standardisation, AI performance and gen-
evidence in this area is lacking. Professional guidelines eralisability, however image manipulation is relative and
and laboratory practice emphasise the need to maintain can introduce artefacts, lead to loss of information, or
stain quality and reduce variation through internal and bias the training data [23, 29–31].
external quality assessment, but routine quantitative An alternative approach to reduce variation between
assessment of H&E staining has to date been unachiev- images is to reduce stain variation at its source, through
able [8–10]. laboratory quality control (QC). Strict protocols are
The need to quantify and control stain quality is given maintained within histopathology laboratories and
greater impetus with the increasing use of digital pathol- reagents are replenished regularly to minimise varia-
ogy. This is the process of scanning a glass pathology slide tion, with most adopting automated staining instru-
with a whole slide imaging system to produce a digital ments for improved precision. Methods of routine QC
image. The technology has been promoted and adopted have changed little over the years, where both inter-
as it has the potential to improve workflow and quality nal and external quality assessments are based on sub-
in pathology services [11–13]. Its utilisation is growing jective, qualitative observations [5, 32–35]. Although
due to the increasing maturity of whole slide imaging human qualitative assessment is important for assess-
systems, displays, data handling and storage, significant ing quality, it is subject to observer bias and relies on
clinical need for pathology service globally, and the use assessing stained control tissue which, due to intrinsic
of artificial intelligence (AI) to augment human diagnosis biological differences, can be variable between sections.
[10, 14, 15]. Control tissue blocks are finite, being exhausted after a
Image quality, specifically colour, is an important few hundred control sections have been cut, necessitat-
parameter for AI as differences in colour are used to set ing new controls which may have different morphological
thresholds to detect objects and patterns, meaning varia- appearances and staining characteristics. Tissue stain-
tion in the stained colour of tissue can impact upon AI ing may also be confounded by other variables prior to
algorithm performance. An increasing number of papers staining, such as fixation or section thickness variation
in the literature highlight the importance of colour sta- [36–39]. These limitations mean that using tissue-based
bility for AI [5–7, 16–19]. To help mitigate the effect QC approaches alone may not be sufficient as a control
of stain variation, computer assisted methods can be method for stain quality assessment over time, or across
employed such as stain normalisation, the digital nor- institutions.
malisation of an image’s colour, and data augmentation, There has been research into the use of quantitative
where computer simulated images with variable staining controls for immunohistochemistry staining in histopa-
are introduced to training datasets to improve AI robust- thology, and a consortium has recently been launched to
ness [20–23]. With stain normalisation, the accuracy of improve immunohistochemistry reproducibility [40–44].
AI, before and after normalisation, has been shown to But there is limited research focusing on quantitative QC
deliver significant improvements in AI performance [19, methods for H&E staining, which accounts for the major-
23–27]. Examples include improving colorectal cancer ity of stained slides in laboratories worldwide. Gray et al.
classification and prostate cancer detection accuracy by [5] and Chlipala et al. [45] have developed digital meth-
20% and 9% respectively [26, 27]. Other work has found ods of quantifying H&E staining from whole slide images
that prostate cancer classification performance suf- of stained control tissue. Although effective, these meth-
fered when using images from different institutions and ods of quantifying stain can be impacted by confounding
scanners, and that application of stain normalisation to variables as they use tissue as a control and rely on accu-
a variable-quality dataset improved AI performance by rate colour reproduction during digitisation.
5% [24]. Inter-institutional staining characteristics can In this paper we propose a method for absolute quan-
be distinguishable by AI and have the potential to bias tification of H&E staining in the laboratory environment,
accuracy, even with application of stain normalisation using stain assessment slides. Stain assessment slides
Dunn et al. Diagnostic Pathology (2024) 19:42 Page 3 of 11

comprise of a biopolymer film applied as a label to stan- were stained for 13 stain durations, from 15 s to 6 min,
dard pathology glass slides. The biopolymer film is highly with five slides at each stain duration (n = 65 per stain
receptive to stain due to its hydrophilicity and porous technique). Stain durations are shown in Supplementary
structure. We characterise the stain assessment slides, Information Table 2.
compare the stain response with tissue, and validate the
use of this methodology as routine QC testing for H&E Analysis
staining within a clinical laboratory. This technique has The stain assessment slides were scanned in a UV-Vis
the potential to offer truly objective and quantitative QC Cary100 spectrophotometer (Agilent Technologies,
of H&E staining, to augment current QC processes in Santa Clara, USA). Prior to scanning, the spectropho-
laboratories. tometer was calibrated using certified reference materials
traceable to the National Physical Laboratory (Tedding-
Materials and methods ton, UK) primary references, and the baseline and zero
Experiment 1: stain assessment slide H&E characterisation were set following the standard procedure [46–48].
Methodology Absorbance spectra were measured from each slide
A biopolymer film, with a standard thickness of 24.4 μm between 350 and 800 nanometres (nm), at 1 nm incre-
(±2%), was sourced (Futamura Chemical UK Limited, ments. Total absorbance was calculated from each spec-
Wigton, UK). Discs of the biopolymer film (10 mm diam- trum to provide a single number for comparison between
eter) were cut and positioned onto non-coated glass slides. Total absorbance was the sum of all absorbance
slides (25 x 50 mm; Solmedia Ltd, Shrewsbury, UK). values within the visible spectrum (380–740 nm). Aver-
Chemically resistant polyethylene terephthalate (PET) age total absorbance was calculated for each stain dura-
labels with acrylic adhesive (17 x 25 mm; North and tion and technique, and plotted onto a scatter graphs
South labels Ltd, Thornton Heath, UK) had a central with linear trend-lines applied. Error bars of one stan-
8 mm diameter circular aperture removed and were over- dard deviation from the mean were included. Using
laid to adhere the biopolymer film to the slides. Hereafter Minitab Desktop 21.2 statistical software (State College,
these slides will be referred to as stain assessment slides; USA), Pearson’s correlation coefficient (r) was calculated
they are depicted in Fig. 1. to assess the strength of the linear relationship, and coef-
Stain assessment slides were manually stained with ficient of variation (CV) was calculated to show relative
Mayer’s haematoxylin and eosin Y 1% aqueous (see Sup- standard deviation(σ) as a percentage of the mean (µ),
plementary Information Table 1 for information on stains using Eq. (1). The CV was calculated for each stain dura-
and reagents used) according to the protocol in Table 1. tion and averaged for each stain technique with 95% con-
Three stain techniques were used: (1) haematoxylin- fidence intervals provided for each CV average.
only, (2) eosin-only and (3) H&E combined (equal stain
duration for each stain). For each stain technique slides

Fig. 1 Stain assessment slide. An illustration (a) and a photo (b) of an example stain assessment slide, consisting of a disc of biopolymer film positioned
onto a glass slide, with a chemically resistant PET label positioned on top to affix the biopolymer in position. The dotted grey line indicates the area where
tissue sections may be mounted
Dunn et al. Diagnostic Pathology (2024) 19:42 Page 4 of 11

Table 1 Staining protocol used in Experiments 1 and 2

Process Step Solution Duration (m:s) Process Step Solution Duration (m:s)
Dewaxing 1 Xylene (wash 1) 3:00 Staining 14 Running tap water 1:00
2 Xylene (wash 2) 3:00 15 Scott’s Tap Water 2:00
3 Xylene (wash 3) 3:00 16 Running tap water 1:00
4 Xylene (wash 4) 3:00 17 Aqueous Eosin Y (1%) 0:15 − 6:00 (see
SI Table 2)
Rehydration 5 100% ethanol (wash 1) 3:00 18 Running tap water 1:00
6 100% ethanol (wash 2) 3:00 Dehydration 19 100% ethanol (wash 1) 0:30
7 100% ethanol (wash 3) 3:00 20 100% ethanol (wash 2) 1:00
8 100% ethanol (wash 4) 3:00 21 100% ethanol 5:00
(wash 3)
9 75% ethanol 3:00 22 100% ethanol (wash 4) 5:00
10 50% ethanol 3:00 Clearing 23 Xylene (wash 1) 3:00
11 25% ethanol 1:00 24 Xylene (wash 2) 3:00
12 Running tap water 2:00 25 Xylene (wash 3) 3:00
Staining 13 Mayer’s Haematoxylin 0:15 − 6:00 (see SI Table 2) Mounting 26 DPX mountant -
The staining protocol used in Experiment 1 and 2 is described, outlining process, solution and duration used. All steps in this protocol were undertaken at room
temperature. Please note stain technique 1 (haematoxylin only) excluded steps 17–18, and stain technique 2 (eosin only) excluded steps 13–16. Stain technique 3
included all steps. Abbreviations: m:s, minutes : seconds; SI, Supplementary Information; DPX, Dibutylphthalate Polystyrene Xylene

σ
Cv = × 100 (1) folds), to determine the relative relationship. The scanned
µ
images were viewed on Aperio ImageScope 12.1 (Leica
Biosystems) and extracted, using the extract region
Experiment 2: characterisation with tissue tool, as jpeg files using JPEG2000 compression (quality
Methodology score 30). The extracted images were viewed on ImageJ
55 stain assessment slides were constructed using the (Bethesda, Maryland, USA), where colour was measured
technique described in Experiment 1. To allow for in Red (R), Green (G) and Blue (B) – RGB – colour space.
increased space on the slide for tissue mounting, this In this colour space, R is a numerical representation of
experiment used 4 mm discs of biopolymer, overlaid with the stained colour intensity in the red spectrum, G in the
the PET label cut to smaller dimensions, 4.5 × 7.5 mm, green, and B in the blue. Median RGB values of biopoly-
with a 2 mm circular aperture removed. mer and tissue on each slide were calculated and plot-
Surplus human liver tissue was sourced. The tissue was ted against each other on a scatter graph. Using Minitab,
processed using a Leica ASP300S processor (Leica Bio- Pearson’s correlation coefficient (r) was calculated to
systems, Wetzlar, Germany) and embedded into a paraf- assess the strength of any correlation and CV was calcu-
fin wax block. The tissue was sectioned to 5 μm using a lated for each stain duration and averaged, with 95% con-
microtome by a senior research technician and mounted fidence intervals provided.
onto the stain assessment slides above the biopolymer
label (see Fig. 1). The slides were placed onto a hot plate Experiment 3: clinical implementation
at 60 °C for two hours and stained using Mayer’s haema- To validate the stain assessment slides as a QC method,
toxylin and eosin Y aqueous 1% (equal time each stain, two proof of concept studies were conducted in one clini-
see Supplementary Information Table 1 for stain infor- cal laboratory using automated staining instruments. The
mation) according to the protocol in Table 1, for stain two arms to this experiment were (a) assessment of varia-
durations between 15 s and 6 min (see Supplementary tion at one point in time, and (b) assessment of variation
Information Table 2), with five slides stained at each stain over a five-day period. Three clinically active staining
duration (n = 55). instruments of the same manufacturer and model were
tested; assigned as Stainer-1, Stainer-2 and Stainer-3. All
Analysis instruments used identical H&E staining protocols.
The slides were scanned in an Aperio AT 2 whole slide
imaging scanner (Leica Biosystems) at 20x magnification Methodology
(0.5 microns per pixel), with JPEG compression (qual- Experiment 3a) assessment of variation at one point-in-time
ity = 70). Digital images were used in this experiment, as This assessment was undertaken to test the level of varia-
opposed to spectral measurements, to enable an aver- tion of the stain assessment slides within three instru-
age colour measurement across the entire biopolymer ments at one point-in-time. 90 stain assessment slides
and tissue area (excluding areas with artefacts, such as were constructed, as described in Experiment 1. One
Dunn et al. Diagnostic Pathology (2024) 19:42 Page 5 of 11

full rack of slides (n = 30) was positioned in each of three depicting one standard deviation from the mean at each
staining instruments (Stainer-1, Stainer-2 and Stainer-3) time point highlight the variation between samples. The
and stained at one point-in-time, using the laboratory’s average CV, with 95% confidence intervals displayed as
standard H&E staining protocol. (lower limit, upper limit), for all time durations was 11%
(6, 16), 11% (9, 13) and 9% (5, 13) for haematoxylin-only,
Experiment 3b) assessment of stain variation over five days eosin-only and H&E combined respectively. See Supple-
For assessment of variation within staining instruments mentary Information Table 3 for the full range of stan-
over time, the three staining instruments were assessed dard deviation and CV values for each stain duration and
over a period of five days. 75 stain assessment slides were technique.
constructed using the method described in Experiment
1. One stain assessment slide was placed in each of the Experiment 2: comparison of stain assessment slides and
three staining instruments and stained with H&E along- tissue
side tissue samples for routine clinical diagnosis. This Median RGB values of H&E stained biopolymer and
was repeated five times per day, over a period of five human liver tissue are plotted against each other in
days (Monday to Friday). The time of staining was spread Fig. 3a. Figure 3b provides thumbnail images of liver and
across each day between 9:00 and 17:00 h. biopolymer stained with H&E for 1–6 min, for visual
comparison of stained colour. There was a linear correla-
Analysis tion found between the biopolymer and liver tissue for R
Experiments 3a and 3b (r =0.99), G (r =0.98) and B (r =0.98) values. The average
After staining, the stain assessment slides from Experi- CV of all the stain durations for RGB values respectively
ment 3a and 3b were scanned in a spectrophotometer as was 2% (1, 2), 4% (3, 5) and 2% (1, 2) for liver tissue, and
detailed in Experiment 1. From the absorbance spectra, 6% (5, 8), 14% (10, 18) and 7% (5, 10) for the biopolymer.
total absorbance was calculated. Using Minitab, boxplots
were generated showing the spread of results. For Experi- Experiment 3: clinical implementation
ment 3a, CV was calculated to measure intra-instrument Experiment 3a) assessment of variation at one point in time
and inter-instrument variation at one point in time. For Boxplots showing the spread of total absorbance mea-
Experiment 3b CV was calculated to measure intra- sured from the stain assessment slides for each instru-
instrument variation for individual days and across the ment at one point in time can be seen in Fig. 4a.
five days, and inter-instrument variation across the five Intra-instrument variation (CV) was 6% (Stainer-1), 9%
days, with 95% confidence intervals provided. (Stainer-2) and 7% (Stainer-3), showing the level of varia-
Inter-instrument variation in Experiment 3a and 3b tion in stain assessment slides at one point in time. The
was found to be normally distributed using the Ander- inter-instrument variation (CV) at one point in time was
son-Darling normality test and so analysis of variance 8%. This variation was calculated to be statistically sig-
(ANOVA) tests were carried out on the data, where nificant (p = 0.0003).
p < 0.05 is considered significant, to compare results for
inter-instrument variation across five days and at one Experiment 3b) Assessment of daily stain variation
point in time. Boxplots showing the spread of total absorbance for each
instrument across five days can be seen in Fig. 4b. The CV
Results across the five days was 28% (Stainer-1), 23% (Stainer-2)
Experiment 1: stain assessment slide characterisation and 30% (Stainer-3), indicating the intra-instrument
Figure 2a shows an example of six averaged spectra from variation for each stain instrument over the time-period.
H&E-stained stain assessment slides, with sparse data The intra-instrument variation over five days was statis-
shown for clarity (stain durations: 1–6 min). The spectra tically significant for Stainer-2 (p=0.001), but not signifi-
demonstrate that as stain duration increased, the portion cant for Stainer-1 (p=0.699) or Stainer-3 (p=0.062). The
of the spectral curve, where the biopolymer absorbed inter-instrument variation (CV) was 27%, but this was
light, increased incrementally for each stain technique, not statistically significant (p=0.441). See Supplemen-
indicating increasing intensity. tary Information Table 4 for more detailed results from
Average total absorbance within the visible spectrum Experiment 3a and 3b.
(380–740 nm, as highlighted between the reference lines
in Fig. 2a for each stain duration and technique were Discussion
plotted in Fig. 2b. Average total absorbance for each We have proposed that improving stain QC and stan-
stain technique increased linearly over time, with Pear- dardisation is a practical and logical approach to
son’s correlation coefficient (r) values of 0.99 (haematox- ensuring consistency of traditional laboratory stain
ylin-only), 0.99 (eosin-only) and 0.99 (H&E). Error bars quality and the resultant digital data set.
Dunn et al. Diagnostic Pathology (2024) 19:42 Page 6 of 11

Fig. 2 Stain assessment slide H&E stain response. a) Mean absorbance spectra of biopolymer film on stain assessment slides, stained with H&E (equal
time each stain) from 1 to 6 min, with five slides stained at each stain duration. The reference lines provided indicate the portion of the spectrum that
represents visible light wavelengths, between 380 and 740 nm, from which total absorbance was measured. b Average total absorbance of biopolymer
film stained using haematoxylin, eosin and H&E combined, for durations ranging from 15 s to 6 min. Each point plotted is the average of five slides at each
stain duration, with error bars depicting one standard deviation from the mean in each direction and linear trend lines applied
Dunn et al. Diagnostic Pathology (2024) 19:42 Page 7 of 11

Fig. 3 H&E stain response of stain assessment slides and human liver tissue. (a) Stain response scatterplot comparing median Red (R), Green (G) and Blue
(B) colour values of human liver tissue against biopolymer film on stain assessment slides, stained with H&E between 15 s and 6 min (equal duration for
each stain). Five slides were stained at each stain duration, with linear trend-lines applied. (b) Thumbnail images for visual comparison of stain response
measured from whole slide images of human liver tissue and biopolymer film, stained with H&E from 1 – 6 min (equal duration for each stain)

We evaluated a novel method of stain QC in a series were an effective, quantitative measure of staining,
of experiments. Experiment 1 characterised the bio- based on purposefully altering stain duration.
polymer film on stain assessment slides, stained with Experiment 2 compared the H&E staining charac-
H&E (separately and combined) and found a linear teristics of the biopolymer with sections of human
relationship between stain duration and stained colour liver tissue, to contrast the performance of the sys-
of the biopolymer, with r values of 0.99 for all stain tem with the conventional use of tissue-based con-
techniques. This demonstrated that the stain assess- trols. There was a strong correlation between mean
ment slides take up H&E stain linearly over time and biopolymer and liver staining (r values between 0.98
Dunn et al. Diagnostic Pathology (2024) 19:42 Page 8 of 11

Fig. 4 Spread of total absorbance across a five day period. Boxplots showing the spread of results of total absorbance measured from absorbance spectra
of stain assessment slides stained in three staining instruments in a clinical laboratory during Experiment 3b. Five slides were stained in each stain instru-
ment per day, using the same staining protocol, over a period of five days (n = 25 per stain instrument). Stainer-1 box is 79 – 104, whiskers are 50 – 138
with median of 91 and an outlier at 40. Stainer-2 box is 70 – 97, whiskers are 41 – 114 with a median of 83. Stainer-3 box is 70 – 107, whiskers are 43 – 159
with a median of 90

and 0.99) indicating that biopolymer stain uptake was measured. The stain assessment slides offer a simple
linearly comparable to human liver tissue within the method of quantifying variation and characterising
stain durations measured. The linear relationship was staining instruments on a periodic basis. However,
non-proportional (y intercept ≠ 0) due to the biopoly- despite the instruments staining being significantly
mer film having an increased thickness (24.4 μm bio- different, only 8% variation was measured at one point
polymer vs. 5 μm tissue sections), permitting higher in time, which was a low level of variation (similar to
sensitivity of the biopolymer to detect variations in baseline level of variation within the stain assessment
staining. slides), particularly considering the biopolymer has
Experiment 3 implemented stain assessment slides an increased sensitivity to stain compared to human
within a clinical laboratory to establish the clini- tissue.
cal utility of the method. Experiment 3a assessed Experiment 3b assessed variation across five days
variation at one point in time and found the intra- and found an average intra-instrument variation of
instrument variation was 6–9%; a similar level to the between 23 and 28%. This is approximately 2.5–4.5
average variation found across stain durations in times higher than the level of variation found in Exper-
Experiments 1 and 2. This suggests that this varia- iment 3a at one point in time, which highlights the
tion was dominated by intra-batch variation in the increased variation present across five days. The daily
stain assessment slides, rather than variation within variation reached as high as 47% on one day (Stainer-3,
the staining instruments, however this was not pos- day 2). The inter-instrument variation was 27% but
sible to discriminate. The inter-instrument variation was not found to be significant, although this may be
at one point in time was 8%, which was found to be due to paucity of data. The variation was likely caused
statistically significant (p = 0.0003). This indicated that by dilution of reagents and high throughput of slides
despite different instruments using the same protocol, over the course of one week. Daily quantitative QC
inter-instrument variations are present. Varying levels would have a strong potential to limit this variation by
of slide throughput may have contributed to this, e.g. setting thresholds of normal operation; this would also
a higher throughput of slides may equate to a higher provide onward benefits for AI by providing more con-
likelihood of reagents becoming diluted/contaminated. sistent data for both training and utilisation. A limita-
There may also be variations between different H&E tion of this experiment was that information was not
stain batches that could contribute to the variation collected on the frequency of stain reagent changes.
Dunn et al. Diagnostic Pathology (2024) 19:42 Page 9 of 11

As one of the potential benefits of the stain assessment characterised, as well as determination of the level
slides would be to optimise reagent use, that infor- of variation in stain assessment slides that equates to
mation is important and should be included in future visually or diagnostically noticeable differences in dif-
work. If less frequent reagent changes can be identified ferent tissue types.
this could be of financial benefit to laboratories, either Further work will develop an operational process to
way this information potentially informs on future allow stain assessment slides to be readily deployed
guidelines or standards. and utilised in an operational environment. The use of
Additional limitations of this study include the vari- a spectrophotometer is impractical in an operational
ability in stain uptake by the biopolymer at 6–14%. For pathology workflow, however if a laboratory has been
context this variability was subjectively barely perceiv- digitalised already, a whole slide imager could prac-
able compared to the staining instrument variation tically be used to collect stain data. There are two
found across five days, which was readily noticeable at potential limitations of this, one is that not all labo-
23–28%. It is thought that the variation was largely due ratories have gone digital, and the other is that a time
to the high sensitivity of the biopolymer; the hand- lag is introduced between staining and the returned
made nature of constructing stain assessment slides; quantitative data, which may limit the utility of the
and the use of a manual staining process in Experi- stain assessment tool as a near-time quality control.
ments 1 and 2. As such automated manufacture and To address this, we are developing a small, labora-
staining processes should improve this. A further limi- tory-friendly device to measure colour directly from
tation was the use of different techniques to measure the stain assessment slides that can fit easily into the
the colour of the biopolymer film. In Experiments 1 laboratory workflow and provide immediate feedback.
and 3, colour was measured spectrally (total absor- It is important to note that the stain assessment slides
bance), to characterise the absolute stained colour in allow quantification of the stain delivered to tissue. We
the biopolymer. Experiment 2 differed in that colour accept that there are complex relationships between
was measured digitally (RGB values) to characterise haematoxylin, eosin and tissue presentation. The
the relative relationship between the biopolymer and use of stain assessment slides is not for assessing the
tissue stain uptake. Accurate colour measurement impact on clinical presentation, but to provide infor-
from whole slide images relies on accurate colour mation that the staining instrument may or may not be
reproduction of the imaging system. The whole slide performing within pre-defined parameters as that may
images were manually checked for quality, but the AT have a consequence for the clinical presentation.
2 scanner was not specifically colour-calibrated prior In summary, this work presents a novel method using a
to use, other than the out-of-factory calibration, setup biopolymer as a quantitative H&E stain assessment tool
and yearly calibration by the manufacturers follow- that:
ing their standard procedure. Because we scanned
the biopolymer and tissue in the same scanner at the •  demonstrates linear staining with H&E,
same time, we can determine from previous experi- •  shows comparable stain uptake to control tissue
mental work that this scanner would have an expected slides,
variation in colour measurement of 0.47%, which is •  has demonstrable clinical utility in measuring stain
an order of magnitude lower than the stain variation variation.
being measured in the stain assessment slides and tis-
sue [5]. There was no direct comparison between the If adopted into routine practice, the presented QC tool
spectral and digital colour measurements and future could improve stain consistency and optimise reagent
work will compare these methods. use by removing subjectivity in stain assessment. This
The H&E characterisation in this paper was based on technique can be used as a periodic point-in-time test
an intensity measurement of H&E staining with equal for staining instruments, to be used alongside labo-
time for each stain (1:1 ratio), so additional analysis is ratory internal and external qualitative assessment
needed to understand the biopolymer response to dis- protocols. An added benefit of quantifying stain vari-
proportionate H&E stain durations. Early work sug- ability is the potential cost-saving by optimising stain
gests that this will be proportional to the time-stain replenishment and reducing reagent use. There are
uptake curves shown in Fig. 2b. The relative uptake also clinical and operational benefits from reducing
of H&E stains needs to be reported to inform practi- the need to re-section and re-stain tissue if stain qual-
cal instrument optimisation in the laboratory. Digi- ity drops. These benefits will not only help optimise
tal methods do exist to do this already, for example, the speed and quality of diagnosis but also help to pro-
stain deconvolution by Ruifrok et al. [49]. The impact duce consistent digital whole slide images and to help
of varying H&E types/brands also needs to be fully facilitate AI in digital pathology in future.
Dunn et al. Diagnostic Pathology (2024) 19:42 Page 10 of 11

Abbreviations References
ANOVA Analysis of variance 1. Titford M. A Short History of Histopathology Technique, Journal of Histotech-
AI artificial intelligence nology, vol. 29, no. 2, pp. 99–110,2006/06/01 2006, https://doi.org/10.1179/
CV coefficient of variation his.2006.29.2.99.
DPX Dibutylphthalate Polystyrene Xylene 2. Hussein I, Raad M, Safa R, Jurjus RA, Jurjus A. Once upon a microscopic slide:
H&E haematoxylin and eosin the story of histology. J Cytol Histol, 6, 2015.
µ mean 3. Lyon HO et al. Standardization of reagents and methods used in cytological
m s:minutes:seconds and histological practice with emphasis on dyes, stains and chromogenic
nm nanometres reagents. Histochemical J J Article 26, 7, pp. 533–44, July 01 1994, https://doi.
PET polyethylene terephthalate org/10.1007/bf00158587.
QC quality control 4. Bejnordi BE, Timofeeva N, Otte-Höller I, Karssemeijer N. W. M. v. d. Laak,
RGB red, green and blue quantitative analysis of stain variability in histology slides and an algorithm
σ standard deviation for standardization. Med Imaging 2014: Digit Pathol. 2014;9041. https://doi.
SI supplementary information org/10.1117/12.2043683.
5. Gray A, Wright A, Jackson P, Hale M, Treanor D. Quantification of histochemi-
cal stains using whole slide imaging: development of a method and demon-
stration of its usefulness in laboratory quality control, (in eng). J Clin Pathol.
Supplementary Information Mar 2014;68(3):192–9. https://doi.org/10.1136/jclinpath-2014-202526.
The online version contains supplementary material available at https://doi. 6. Bejnordi BE et al. Stain Specific Standardization of Whole-Slide Histopatho-
org/10.1186/s13000-024-01461-w. logical Images, (in eng), IEEE transactions on medical imaging, vol. 35, no. 2, pp.
404 – 15, Feb 2016, https://doi.org/10.1109/tmi.2015.2476509.
Supplementary Material 1 7. Howard FM, et al. The impact of site-specific digital histology signatures on
deep learning model accuracy and bias. Nat Commun. 2021;12(1):1–13.
Supplementary Material 2 8. The Royal College of Pathologists., How to Assess the Quality of
a Pathology Service, www.rcpath.org/uploads/assets/1c3aac02-
3f31-4246-83b9da4aa04899ca/How-to-Assess-the-Quality-
Acknowledgements
of-a-Pathology-Service-https://www.rcpath.org/uploads/
We thank Mike Shires, Doreen Crellin, Allan Gray, Michael Hale and Dr
assets/1c3aac02-3f31-4246-83b9da4aa04899ca/How-to-Assess-the-Quality-
Alexander Wright for their advice, support and contributions.
of-a-Pathology-Service-meeting-report.pdf 2011. [Online]. Available:
9. Robert Lott JT, Sheppard E, Santiago J, Hladik C, Nasim M, Zeitner K, Haas T.
Author contributions
Shane Kohl, Saeid Movahedi-Lankaran, Practical Guide to Specimen Handling
D.T. was the principal investigator. D.T., D.B. and C.D. designed the research.
in Surgical Pathology, <https://cap.objects.frb.io/documents/practical-guide-
C.D. performed laboratory work and statistical analysis. M.C., E.K. and C.R.
specimen-handling.pdf> College of American Pathologists, 2020. [Online].
provided technical knowledge and advice. C.D. wrote the manuscript with
Available: https://cap.objects.frb.io/documents/practical-guide-specimen-
assistance from all authors. All authors reviewed and approved the final
handling.pdf.
manuscript.
10. Williams BJ, Knowles C, Treanor D. Maintaining quality diagnosis with digital
pathology: a practical guide to ISO 15189 accreditation. J Clin Pathol.
Funding
2019;72(10):663–8.
This research is part of the National Pathology Imaging Co-operative, NPIC
11. Griffin J, Treanor D. Digital pathology in clinical use: where are we now and
(Project no. 104687) supported by a £50 m investment from the Data to Early
what is holding us back? Histopathology, vol. 70, no. 1, pp. 134–145, 2017.
Diagnosis and Precision Medicine challenge, managed and delivered by UK
12. Gupta R, Kurc T, Sharma A, Almeida JS, Saltz J. The emergence of pathomics.
Research and Innovation (UKRI).
Curr Pathobiol Rep. 2019;7(3):73–84.
13. Baxi V, Edwards R, Montalto M, Saha S. Digital pathology and artificial intel-
Data availability
ligence in translational medicine and clinical practice. Mod Pathol, 35, 1, pp.
The datasets supporting the conclusions of this article are included within the
23–32, 2022/01/01 2022, https://doi.org/10.1038/s41379-021-00919-2.
articles Supplementary Information.
14. Niazi MKK, Parwani AV, Gurcan MN. Digital pathology and artificial intel-
ligence. Lancet Oncol, 20, 5, pp. e253-e261, 2019/05/01/ 2019.
Declarations 15. Williams BJ, Bottoms D, Treanor D. Future-proofing pathology: the case for
clinical adoption of digital pathology. J Clin Pathol. 2017;70(12):1010–8.
Ethical approval 16. Tellez D, et al. Whole-slide mitosis detection in H&E breast histology using
All methods were carried out in compliance with relevant guidelines and PHH3 as a reference to Train Distilled Stain-Invariant Convolutional Networks.
regulations, and were under appropriate approval including NHS research IEEE Trans Med Imaging. 2018;37(9):2126–36. https://doi.org/10.1109/
ethics committee (Leeds West LREC reference 05/01205/270). The study TMI.2018.2820199.
did not involve any subjects. Samples of human tissue were used as control 17. Schömig-Markiefka B et al. Quality control stress test for deep learning-based
material in part of the work. Informed consent was not obtained for this, diagnostic model in digital pathology. Mod Pathol, 34, 12, pp. 2098–2108,
as these were surplus tissue samples remaining after diagnosis, they were 2021/12/01 2021, https://doi.org/10.1038/s41379-021-00859-x.
fully anonymised, and their use without specific consent was approved by a 18. Wright AI, Dunn CM, Hale M, Hutchins GGA, Treanor DE. The Effect of
research ethics committee. Quality Control on Accuracy of Digital Pathology Image Analysis, (in eng).
IEEE J Biomed Health Inf. Feb 2021;25(2):307–14. https://doi.org/10.1109/
Competing interests JBHI.2020.3046094.
Authors D.T., D.B. and C.D. are inventors on patents filed by the University 19. Bejnordi BE et al. Stain specific standardization of whole-slide histopathologi-
of Leeds and Leeds Teaching Hospitals NHS Trust on the methodology cal images, (in eng). IEEE Trans Med Imaging, 35, 2, pp. 404 – 15, Feb 2015.
in this paper. FFEI Ltd and Futamura Chemical UK Ltd are partners on the 20. Ciompi F et al. The importance of stain normalization in colorectal tissue
National Pathology Imaging Co-operative program, and are providing in-kind classification with convolutional networks, 2017 IEEE 14th International Sym-
contributions to the research activities in NPIC. Authors M.C. and E.K. are posium on Biomedical Imaging (ISBI 2017), pp. 160–163, 18–21 April 2017 2017,
employees of Futamura Chemical UK Ltd, and are inventors on a patent filed https://doi.org/10.1109/ISBI.2017.7950492.
by Futamura Chemical UK Ltd on the composition of the biopolymer film. 21. Khan AM, Rajpoot N, Treanor D, Magee D. A nonlinear Mapping Approach to
Author C.R. is an employee of FFEI Ltd. Stain normalization in Digital Histopathology images using image-specific
Color Deconvolution. IEEE Trans Biomed Eng. 2014;61(6):1729–38. https://doi.
org/10.1109/TBME.2014.2303294.
Received: 8 June 2023 / Accepted: 3 February 2024
Dunn et al. Diagnostic Pathology (2024) 19:42 Page 11 of 11

22. Komura D, Ishikawa S. Machine learning methods for histopathological 38. Bass BP, Engel KB, Greytak SR, Moore HM. A review of preanalytical factors
image analysis. Comput Struct Biotechnol J. 2018;16:34–42. https://doi. affecting molecular, protein, and morphological analysis of formalin-fixed,
org/10.1016/j.csbj.2018.01.001. (in eng). paraffin-embedded (FFPE) tissue: how well do you know your FFPE speci-
23. Tellez D et al. Quantifying the effects of data augmentation and stain color men? Arch Pathol Lab Med. 2014;138(11):1520–30.
normalization in convolutional neural networks for computational pathol- 39. Chlipala EA, et al. Impact of preanalytical factors during histology pro-
ogy. Med Image Anal, 58, p. 101544, 2019/12/01/ 2019, doi: https://doi. cessing on section suitability for digital image analysis. Toxicol Pathol.
org/10.1016/j.media.2019.101544. 2021;49(4):755–72.
24. Anghel A, et al. A high-performance system for robust stain normalization of 40. Hotzel KJ et al. Synthetic Antigen Gels as Practical Controls for Standardized
whole-slide images in histopathology. Front Med. 2019;6:193. and Quantitative Immunohistochemistry, (in eng), The journal of histochemis-
25. Swiderska-Chadaj Z, et al. Impact of rescanning and normalization on con- try and cytochemistry: official journal of the Histochemistry Society, vol. 67, no. 5,
volutional neural network performance in multi-center, whole-slide classifica- pp. 309–334, May 2019, https://doi.org/10.1369/0022155419832002.
tion of prostate cancer. Sci Rep. 2020;10(1):14398. 41. Sompuram SR, Vani K, Tracey B, Kamstock DA, Bogen SA. Standardizing
26. Ciompi F et al. The importance of stain normalization in colorectal tissue Immunohistochemistry: A New Reference Control for Detecting Staining
classification with convolutional networks, in 2017 IEEE 14th International Problems, (in eng), The journal of histochemistry and cytochemistry: official jour-
Symposium on Biomedical Imaging (ISBI 2017), 2017: IEEE, pp. 160–163. nal of the Histochemistry Society, vol. 63, no. 9, pp. 681 – 90, Sep 2015, https://
27. Salvi M, Molinari F, Acharya UR, Molinaro L, Meiburger KM. Impact of stain doi.org/10.1369/0022155415588109.
normalization and patch selection on the performance of convolutional neu- 42. Bogen SA et al. Experimental validation of peptide immunohistochemistry
ral networks in histological breast and prostate cancer classification. Comput controls, (in eng), Applied immunohistochemistry & molecular morphol-
Methods Programs Biomed Update. 2021;1:100004. ogy: AIMM / official publication of the Society for Applied Immunohisto-
28. Salvi M et al. Impact of Stain Normalization on Pathologist Assessment of chemistry, vol. 17, no. 3, pp. 239 – 46, May 2009, https://doi.org/10.1097/
Prostate Cancer: A Comparative Study, Cancers, vol. 15, no. 5, p. 1503, 2023. PAI.0b013e3181904379.
29. Tosta TAA, de Faria PR, Neves LA, do Nascimento MZ. Computational normal- 43. Torlakovic EE, et al. Development and validation of measurement traceability
ization of H&E-stained histological images: Progress, challenges and future for in situ immunoassays. Clin Chem. 2021;67(5):763–71.
potential. Artif Intell Med. 2019;95:118–32. 44. Bogen SA et al. A Consortium for Analytic standardization in immunohisto-
30. Beena M. A Survey on Color Normalization Approach to Histopathology chemistry. Arch Pathol Lab Med, 2022.
images. Int J Adv Eng Res Sci. 2016;3(4):258867. 45. Chlipala E, et al. Optical density-based image analysis method for the
31. Howard FM et al. The Impact of Digital Histopathology Batch Effect on Deep evaluation of hematoxylin and eosin staining precision. J Histotechnology.
Learning Model Accuracy and Bias, bioRxiv, 2020. 2020;43(1):29–37.
32. Cox B, Colgan E. 1 - Pathology laboratory management, in Bancroft’s Theory 46. Allen DW. Holmium Oxide Glass Wavelength standards, (in eng). J Res Natl
and Practice of Histological Techniques (Eighth Edition), S. K. Suvarna, C. Layton, Inst Stand Technol. 2007;112(6):303–6. https://doi.org/10.6028/jres.112.024.
and J. D. Bancroft Eds.: Elsevier, 2019, pp. 1–11. 47. Eckerle KL, Weidner VR, Hsia JJ, Kafadar K. Measurement Assurance Program
33. NSH. Histology Quality Improvement Program (HistoQIP). https://www.nsh. Transmittance standards for Spectrophotometric Linearity Testing:* Prepara-
org/learn/histoqip (accessed 15/09/2020. tion and Calibration. J Res Natl Bureau Stand, 88, 1, 1983.
34. UKNEQAS. UK National External Quality Assessment Service. https://ukneqas. 48. National Physical Laboratory, A National Measurement Good Practise Guide.
org.uk (accessed. https://www.npl.co.uk/special-pages/guides/mgpg97 (accessed 09/10/19,
35. CAP. College of American Pathologists, Laboratory Accreditation Program. 2019).
https://www.cap.org/laboratory-improvement/accreditation/laboratory- 49. Ruifrok AC, Johnston DA. Quantification of histochemical staining by color
accreditation-program (accessed. deconvolution. Anal Quant Cytol Histol. 2001;23(4):291–9.
36. Allison RT, Vincent JFV. Measuring the forces acting during microtomy
by the use of load cells. J Microsc. 1990;159(2):203–10. https://doi.
org/10.1111/j.1365-2818.1990.tb04776.x.
37. McCampbell AS, et al. Tissue thickness effects on immunohistochemical Publisher’s Note
staining intensity of markers of Cancer. Appl Immunohistochem Mol Mor- Springer Nature remains neutral with regard to jurisdictional claims in
phology. 2019;27(5):345–55. https://doi.org/10.1097/pai.0000000000000593. published maps and institutional affiliations.