A transportable hyperspectral imaging setup based on fast, high-

density spectral scanning for in situ quantitative biochemical

mapping of fresh tissue biopsies
Luca Giannoni,a,b,†,* Marta Marradi,a,b,† Kevin Scibilia,c Ivan Ezhov,c Camilla Bonaudo,d
Angelos Artemiou,e Anam Toaha,a,b, Frédéric Lange,e Charly Caredda,f Bruno Montcel,f
Alessandro Della Puppa,d Ilias Tachtsidis,e Daniel Rückert,c,g and Francesco Saverio
Pavonea,b,h
a
University of Florence, Department of Physics and Astronomy, Florence, Italy
b
European Laboratory for Non-Linear Spectroscopy, Sesto Fiorentino, Italy
c
Technical University of Munich, TranslaTUM - Center for Translational Cancer Research, Munich, Germany
d
Azienda Ospedaliero-Universitaria Careggi, University of Florence, Neurosurgery, Department of Neuroscience,
Psychology, Pharmacology and Child Health, Florence, Italy
e
University College London, Department of Medical Physics and Biomedical Engineering, London, UK
f
Univ Lyon, INSA-Lyon, Université Claude Bernard Lyon 1, UJM-Saint Etienne, CNRS, Inserm, CREATIS UMR
5220, Lyon, France
g
Imperial College London, Department of Computing, London, UK
h
National Research Council, National Institute of Optics, Sesto Fiorentino, Italy

Abstract
Significance: Histopathological examination of surgical biopsies, such as in glioma and glioblastoma resection, is
hindered in current clinical practice by the long times required for the laboratory analysis and pathological screening,
typically taking several days or even weeks to be completed.
Aim: We propose here a transportable, high-density, spectral-scanning based hyperspectral imaging (HSI) setup,
named HyperProbe1, that can provide in situ, fast biochemical analysis and mapping of fresh surgical tissue samples,
right after excision, and without the need of fixing, staining nor compromising the integrity of the tissue properties.
Approach: HyperProbe1 is based on spectral scanning via supercontinuum laser illumination filtered with acousto-
optic tuneable filters. Such methodology allows the user to select any number and type of wavelength bands in the
visible and near-infrared range between 510 and 900 nm (up to a maximum of 79), and to reconstruct 3D hypercubes
composed of high-resolution (4-5 µm), widefield images (0.9x0.9 mm2) of the surgical samples, where each pixel is
associated with a complete spectrum.
Results: The HyperProbe1 setup is here presented and characterised. The system is applied on 11 fresh surgical biop-
sies of glioma from routine patients, including different grades of tumour classification. Quantitative analysis of the
composition of the tissue is performed via fast spectral unmixing to reconstruct mapping of major biomarkers, such
as oxy- (HbO2) and deoxyhaemoglobin (HHb), as well as cytochrome-c-oxidase (CCO). We also provided a prelimi-
nary attempt to infer tumour classification based on differences of composition in the samples, suggesting the possi-
bility to use lipid content and differential CCO concentrations to distinguish between lower and higher grade gliomas.
Conclusions: A proof-of-concept of the performances of HyperProbe1 for quantitative, biochemical mapping of sur-
gical biopsies is demonstrated, paving the way for improving current post-surgical, histopathological practice via non-
destructive, in situ streamlined screening of fresh tissue samples in a matter of minutes after excision.

Keywords: hyperspectral imaging, biomedical optics, biophotonics, digital histopathology, neurosurgery.

* Luca Giannoni (corresponding author), E-mail: [email protected]

† The authors contributed equally

1
1 Introduction

Histopathological screening of excised tissue is the current “gold standard” in post-surgical onco-

logical practice1, for clinical and molecular evaluation of critical parameters such as type, grading

and classification of tumours, e.g., in glioma and glioblastoma (GBM) resection 2–4. Normal rou-

tine involves the dispatch of fresh surgical biopsies after resection to the histopathology laboratory:

there, the samples are typically fixed for preservation, sectioned, stained -various staining tech-

niques are used, with haematoxylin and eosin (H&E) staining being the most prominent- and then

imaged with a microscope to determine their structural and molecular composition3. However,

modern histopathological analysis presents several limitations, the most severe one being the

lengthy preparation of the samples that leads to long duration of the procedures to obtain the final

results, which can vary from several days to even weeks after the surgery. Extemporaneous and

intraoperative analyses can be much faster (minutes to hours), but the number of biopsy samples

is limited due to operational logistics and costs of the procedures, and the breadth of information

they can provide is very limited for diagnostic purposes5. Furthermore, diagnosis and classification

can be affected by variability in the subjective interpretation of the results by histopathologists,

with the screening essentially lacking a more quantitative and objective way to systematically pro-

cess the imaging outcomes6. Overall, post-operative prognosis and planning would enormously

benefit from a different approach to histopathology that could provide much faster and more reli-

able information on the tissue biopsies, ideally by having a screening in situ right after the surgery

that could lead to quantitative results in a matter of minutes to hours.

Hyperspectral imaging (HSI) is an optical imaging modality that is becoming increasingly more

notable in recent years in the biomedical and bioimaging fields7, and whose main features can be

particularly suited and advantageous to tackle the abovementioned challenge. HSI acquires and

2
reconstructs images of a target at multiple, narrow, contiguous or adjacent wavelength bands in

the electromagnetic spectrum, typically spanning from the visible to the near infrared (NIR) range8.

This allows the user to obtain 3D spatio-spectral datasets, named “hypercubes”, where each spatial

pixel of the images is associated with a corresponding spectrum of reflected, transmitted and/or

fluorescent light. The information carried by the hypercubes is related to the optical properties of

absorption and scattering of the investigated tissue, from which is then possible to infer, map and

quantify its biochemical and structural composition, without the need for time-consuming staining

procedures or the use of any exogenous contrast agent. Intrinsic biomarkers for physiology and

pathophysiology of the tissue can indeed be identified for diagnostic purposes, such as haemoglo-

bin for haemodynamics, oxygenation and vascularisation, or cytochrome-c-oxidase (CCO) for cel-

lular metabolism8–10, and related to tumour key parameters of classification11. In addition to its

capabilities for non-destructive biochemical analysis of freshly excised tissue, HSI has the addi-

tional advantage of fast image acquisition and data processing, mainly thanks to recent advance-

ments in deep learning and artificial intelligence (AI) algorithms12, achieving near real-time com-

puting and almost immediate visualization of the results13. Finally, HSI technology is typically

compact and relatively inexpensive (compared to other traditional imaging modalities), so that

devices can be developed to be fully transportable and capable to easily fit either within the surgi-

cal room or in its proximity (e.g., in a post-surgical area) without encumbrance.

We present here the first prototype of a compact, fully transportable HSI setup called “Hy-

perProbe1”, which is able to rapidly select at high density, sampling and spectral resolution (3.5

to 7 nm of minimum bandwidth) any desired wavelength band between 510 and 900 nm, and to

image a target at a field of view (FOV) of 0.9x0.9 mm2 with up to 79 spectral bands, in less than

5 minutes. HyperProbe1 has the capability to image broadband reflected light from fresh biopsies

3
and to reconstruct maps of their optical properties, as well as to quantify the content of biomarkers

of interest within the examined tissue (such as the two forms of haemoglobin and CCO) via fast

spectral unmixing algorithms. We provide full technical characterisation of the performances of

HyperProbe1 and a proof-of-concept of its application on samples of freshly excised glioma from

surgical biopsies at different World Health Organization (WHO) gradings14. The success of Hy-

perProbe1 in providing quantitative biochemical analysis and mapping of surgical biopsies can

pave the way to a novel, fast and heightened methodology to perform in situ histopathological

screening right after surgery, without the need for any manipulation or degradation of the samples.

2 Material and methods

2.1 The HyperProbe1 system

HyperProbe1 is a HSI system based on spectral scanning acquisition mode, where the target is

illuminated in rapid sequence at each selected wavelength band, whilst a full-frame image is ac-

quired synchronously at each illumination step.

Table 1 List of components of HyperProbe1 and their specifications.

Component Manufacturer & model Key specifics
NKT Photonics, Broadband illumination (400-2400 nm);
SCL
SuperK FIANIUM FIR20 Maximum total power of 6.5 W;
NKT Photonics, Broadband selection (510-900 nm);
AOTF
SELECT VIS-nIR Spectral resolution of 3.5-7 nm (FWHM);
4.2-MP (2048 x 2048) CMOS sensor;
Hamamatsu, 6.5-µm pixel size;
Camera
ORCA-Flash 3.0 QE up to 82% (Visible and NIR);
Maximum frame rate of 40 fps;
Amplitude Thorlabs;
Amplitude stabilisation within ±0.05%;
stabilisers NEL02A/M + NEL03A/M
Speckle reducer Optotune; LSR-3005-6D-NIR Transmission up to 98%;
Broadband coverage (400-2000 nm);
NKT Photonics,
Optic fibres High power throughput (up to 500 mW);
SuperK CONNECT
1-mm core diameter;
Thorlabs, 15x reflective objective;
Objective
LMM15X-P01 NA = 0.3;
CMOS: Complementary metal-oxide semiconductor; FWHM = Full-width at half maximum; QE = Quantum effi-
ciency; NIR = Near-infrared; FOV = Field of view; NA = Numerical aperture.
4
All the spectral frames are then stacked together to reconstruct the corresponding 3D hypercube

of the target8. A schematics of the configuration of HyperProbe is reported in Fig.1a, whereas a

detailed list of its components is presented in Table 1.

Fig. 1 (a) Schematics of the HyperProbe1 with all its components; (b) Picture of the spectral illumination side of

HyperProbe1, including the SCL source, the AOTF and the controlling devices; (c) Picture of the imaging and de-

tection side of HyperProbe1, depicting the illumination output, the amplitude stabilisers, the speckle reducer and the

camera; (d) Result of the imaging tests on the USAF1951 target with HyperProbe1 at 650 nm; (e) The smallest re-

solved line pair (group 6, element 6) highlighted in blue and its corresponding line profile that shows minimum

FWHM separation (4.38 μm); (f) Example of a sample of excised glioma tissue obtained from the surgical biopsies.

The illumination side of HyperProbe1 (Fig. 1b) is composed of a supercontinuum laser15 (SCL;

5
NKT Photonics, SuperK FIANIUM FIR20), generating a coherent, broadband illumination (400-

2400 nm) at a maximum total power of 6.5 W, and a set of acousto-optic tuneable filters (AOTF;

NKT Photonics, SELECT VIS-nIR) that selectively filters any desired spectral band between 510

and 900 nm, with a minimum bandwidth (full width at half maximum; FWHM) of 3.5 nm in the

visible and a maximum bandwidth of 7 nm in the NIR range. The filtered light from the AOTF is

directed to the sample through an optic fibre delivery system (NKT Photonics, SuperK CON-

NECT) of 1-mm core diameter, coupled with: (1) a pair of laser amplitude stabiliser connected in

series (Thorlabs, NEL02A/M and NEL03A/M), to eliminate intensity noise and maintain illumi-

nation stability over time within 0.05% of a selected output power across the whole spectral range,

(2) a laser speckle reducer (Optotune, LSR-3005-6D-NIR), and (3) an achromatic doublet lens, to

make the beam divergent in order to obtain an illumination spot of 2-3 cm2 on the target. Image

acquisition at each spectral band is obtained on the imaging side of HyperProbe1 (Fig. 1c) by use

of a complementary metal-oxide semiconductor (CMOS) camera (Hamamatsu, ORCA-Flash 3.0).

The CMOS camera has a sensor size of 2048 x 2048 pixels, with pixel size of 6.5 μm, 82% peak

quantum efficiency (at 620 nm), and maximum readout rate of 40 fps. It is coupled with a 15x

reflective objective (Thorlabs, LMM-15X-P01) and an infinity-corrected, conjugated tube lens

(Thorlabs, TTL200-B), to generate a FOV of the target of about 0.9 x 0.9 mm2. The whole setup

is mounted on a wheeled rack (as depicted in Fig.1b) and can be easily transported within or adja-

cent to the surgical room, with the illumination probe and imaging side that can be laid on a small

breadboard for improved stability.

The HyperProbe1 system was characterised in terms of power emission, spectral, temporal and

imaging performances. The complete characteristics and features of HyperProbe1 are reported in

Table 2. HyperProbe1 can scan the entire spectral range of operation (510-900 nm) by sampling

6
sequentially up to 79 wavelength bands at 5-nm steps. Each spectral band is also modulated in

amplitude by the AOTF, to provide an approximately constant output power on the target of about

200-450 μW per band, accounting also for the quantum efficiency (QE) of the camera. This is

aimed at maintaining a fixed integration time of the camera for each spectral frame (between 5 and

30 ms, depending on the target), as well as an adequate signal-to-noise ratio (SNR) throughout

every acquired spectral frame. The typical acquisition time for an entire hypercube (79 bands)

ranges between 1 to 5 minutes (depending on the selected integration time of the camera), for

biological targets. Spatial resolution of HyperProbe1 was assessed for each single spectral band

via imaging of a positive resolution test target, the USAF 1951 (Thorlabs, R1L1S1P). For all these

imaging assessments, the spatial resolution of HyperProbe1 was found equal to 114 lp/mm, corre-

sponding to a smallest resolvable detail of 4.38 μm (as shown in Fig. 1d and Fig. 1e, for 650 nm).

Table 2 Technical characteristics and features of HyperProbe1.

Characteristics Illumination side
Illumination mode: Spectral scanning
Available spectral range: 510-900 nm
Minimum sampling step size: 5 nm
Maximum number of spectral bands: 79 (visible and NIR)
3.5 nm (visible),
Spectral resolution (FWHM):
7 nm (NIR)
~200 μW (visible),
Average output power per spectral band:
~450 μW (NIR)
Power stability over time: ± 0.05%
Characteristics Imaging side
Type of detector: CMOS
Sensor format: 2048 x 2048 pixels
Pixel size: 6.5 μm
Spatial resolution: 4.38 μm
FOV: 0.9 x 0.9 mm2
Frame rate: 40 fps (at full format)
Sensitivity (QE): 82% at 620 nm
Typical acquisition time (per spectral frame): 5 to 30 ms
Typical acquisition time (per hypercube): 1 to 5 min
NIR = Near-infrared; FWHM = Full-width at half maximum; CMOS: Complementary metal-oxide semiconductor;
FOV = Field of view; QE = Quantum efficiency.

7
2.2 Samples preparation and data acquisition

HyperProbe1 was used to image a series of fresh surgical biopsies of glioma tissue, to validate its

performances in retrieving quantitative information of interest on the composition of samples, as

well as to infer pathological characteristics of the tumours akin to what is obtained via traditional

histopathological screening. The brain tissue samples involved in the study (an example is shown

in Fig. 1f) were obtained from fresh surgical excisions of patients taken during routinely performed

neurosurgery for brain tumour resection at the Azienda Ospedaliero-Universitaria Careggi (Uni-

versity Hospital of Florence), in Florence. Authorization for the study (Studio ID: 23672 -

23672_BIO) was granted by the Ethical Committee of the Area Vasta Centro Toscana, under Ital-

ian law and regulations. Informed consent was collected from each patient involved in the study.

A total number of 11 samples (n = 11) were imaged and analysed with HyperProbe1, to guar-

antee a degree of statistical robustness in the reconstructed spectra and to investigate specimen

variability. The samples were composed of portions of the same tissue removed during the resec-

tion that is normally sent to the laboratory for histopathological screening, which classified the

type of the tumour based on WHO gradings14. This class of information for all the samples is

reported in Table 3, providing a broad characterisation of various typologies of glioma.

Table 3 Classification of the tissue samples investigated with HyperProbe1.

Sample identifier WHO grading Additional info
S1 IV
S2 IV
S3 III Discarded due to fragmented size
S4 (FOV1) IV Two separate FOVs were acquired on the same sample
S4 (FOV2) IV Two separate FOVs were acquired on the same sample
S5 IV Labelled with fluorescein
S6 II Discarded due to presence of light interference
S7 II Possibly shifting to higher grade (III)
S8 III Labelled with fluorescein, presence of coagulated tissue
S9 IV Labelled with fluorescein, but negative
S10 II Labelled with fluorescein
S11 IV Labelled with fluorescein, but negative
FOV = Field of view.
8
 Two samples were discarded from the analysis: sample S3 was a biopsy composed of very small

and dispersed fragments of brain tissue (2-3 mm each at most) and, due to their dimensions, the

acquired spectra appeared very flat (we hypothesise a large influence of partial volume effect);

conversely, sample S6 presented ambiguous patterns due to potential light interference at some

wavelengths. In addition, for sample S4, due to its slightly larger dimensions than the rest of the

biopsies (5-6 cm), it was decided to acquire two separate FOVs on the same tissue to further ana-

lyse variability within subjects. Finally, samples S8, S9, S10 and S11 were marked with fluores-

cein during the surgery, albeit S9 and S11 were confirmed to be negative to fluorescent emission.

For the HSI data acquisition, a portion of the glioma samples (average size of 2-3 cm) was pre-

emptively washed in phosphate-buffered saline (PBS) to eliminate blood and other unwanted re-

siduals, then imaged on its surface with HyperProbe1 (79 wavelength bands at 5-nm steps between

510 and 900 nm) within 1 hour after excision. A thin glass coverslip was placed over each sample

to flatten its top surface for uniform focusing of the FOV, whilst a dark absorbing material was

placed at the bottom to avoid any potential reflection of the light back into the tissue. The acquisi-

tion of a single hypercube for every sample was typically less than 5 minutes, a short enough

duration to ensure that the tissue had not deteriorated nor oxidised during the imaging.

2.2 Data processing and spectral unmixing analysis

Reflectance hypercubes 𝑅(𝑥, 𝑦, 𝜆) for each sample were reconstructed by normalising the hyper-

spectral data of the reflected light intensity 𝐼(𝑥, 𝑦, 𝜆) acquired with HyperProbe1 with reference

hypercubes 𝑊(𝑥, 𝑦, 𝜆) obtained using a white calibration standard (Labsphere, Spectralon® 5"),

after dark counts subtraction 𝐷(𝑥, 𝑦, 𝜆), and by weighting the latter two datasets for the ratios of

their corresponding integration times 𝑡 of the camera used during the acquisition. The formula for

the reconstruction of the reflectance hypercubes is

9
 𝑡
𝐼(𝑥, 𝑦, 𝜆) − 𝑡 𝐼 𝐷(𝑥, 𝑦, 𝜆)
𝐷 (1)
𝑅(𝑥, 𝑦, 𝜆) = 𝑡 𝑡𝐼
𝐼
𝑊(𝑥, 𝑦, 𝜆) −
𝑡𝑊 𝑡𝐷 𝐷(𝑥, 𝑦, 𝜆)

where 𝑡𝐼 , 𝑡𝐷 and 𝑡𝑊 are the camera integration times for each frame of the intensity, dark and white

hypercubes, respectively.

A fast, spectral unmixing approach based on modified Beer-Lambert’s law (MBLL) was used

to infer the differences in the molecular composition of the biopsies, as described in Ezhov et al16.

These differences were quantified with respect to sample S1, using the average spectral reflectance

of the central area of the sample as baseline spectrum. Computational time to analyse a full dataset

from each biopsy was about 2-3 minutes with two AMD EPYC 7452 32-Core processors.

We then compared the inferred compositions for two different scenarios: (1) by fitting the whole

measured wavelength range (from 510 nm to 900 nm), and (2) by fitting only the NIR portion of

the available spectrum (in our case, from 740 nm to 900 nm). For the latter, we expected the major

absorbing chromophores to be oxygenated (HbO2) and deoxygenated (HHb) haemoglobin, the ox-

idised (oxCCO) and reduced (redCCO) forms of cytochrome-c-oxidase (CCO), as well as water

and lipids8,17,18. The spectral signatures of the chromophores targeted by the analysis are depicted

in Fig. 2a and Fig. 2b. In the visible range, we also assumed the presence of additional chromo-

phores, specifically the oxidised and reduced forms of cytochrome-b (Cyt-B) and cytochrome-c

(Cyt-C), due to their involvement in the metabolic processes17.

The inferred compositions, in the forms of either concentrations or volumetric contents, are in

units [mM/cm] and [cm-1], respectively, as we used unitary pathlength (1 cm) in our experiments.

We have previously seen that a quasi-constant pathlength only effectively scales the concentra-

tions, at least in the NIR range16, and is therefore sufficient for the preliminary task of attempting

to distinguishing biopsies of different tumour gradings.

10
2.3 Monte Carlo simulations of penetration depth in tissue

A 3D, in silico, optical and geometrical model of brain biopsy was designed to assess and quantify

the depth of penetration of light in the tissue, at the various spectral bands of HyperProbe19. This

was done in MATLAB using a voxel-based Monte Carlo (MC) simulation software. The software

chosen was Monte Carlo eXtreme (MCXLAB)20,21, which simulates photon transport within a 3D,

voxel-based model with arbitrary optical properties. The simulations were carried out on a desktop

computer with an Intel Xeon W5-3425 and two RTX 4090 GPUs.

The simulated geometry consisted of a homogeneous, semi-infinite slab of grey matter with a

thickness of 0.5 cm (taking into account the maximum thickness recorded among the investigated

samples). Isotropic voxels were used, with dimensions of 0.05 mm. This was chosen as it offered

an acceptable trade-off between accuracy and computational performances. A black absorbing

layer of 0.005-cm thickness with a significantly higher absorption coefficient (in the order of 106

cm-1) was placed at the bottom of the slab, to represent the absorbing material used below the

biopsies. Fresnel reflection was implemented at the top and bottom boundaries of the model20.

The reduced scattering coefficient 𝜇𝑠′ of the grey matter of the model was adopted from Jacques

et al18, and was varied with wavelength according to the following equation:

500 nm −𝑏
𝜇𝑠′ = 𝑎( ) , (2)
𝜆

where 𝑎 = 40.8 cm-1, and 𝑏 = 3.089. From 𝜇𝑠′ , the scattering coefficient 𝜇𝑠 was calculated as an

input for MCXLAB with the equation:

𝜇𝑠′
𝜇𝑠 = , (3)
1−𝑔

where 𝑔 is the anisotropy factor of grey matter standing at 0.85 and chosen to be constant, as its

variation with wavelength has been demonstrated to be minimal for cerebral tissue22. Furthermore,

11
the refractive index of the biopsy model was set to 1.3618.

Finally, the 3D model of brain biopsy was assumed to be composed of the most absorbing and

scattering tissue chromophores, i.e., water, lipids, HbO2, HHb, oxCCO and redCCO9. Thus, the

total absorption coefficient 𝜇𝑎 of the model was determined using the equation9,18:

𝜇𝑎 = 𝑊 ∙ 𝜇𝑎,𝐻2 𝑂 + 𝐹 ∙ 𝜇𝑎,𝑓𝑎𝑡 + 𝑙𝑛10 ∙ 𝐶𝐻𝐻𝑏 ∙ 𝜀𝐻𝐻𝑏 + 𝑙𝑛10 ∙ 𝐶𝐻𝑏𝑂2 ∙ 𝜀𝐻𝑏𝑂2 +

(4)
+𝑙𝑛10 ∙ 𝐶𝑜𝑥𝐶𝐶𝑂 ∙ 𝜀𝑜𝑥𝐶𝐶𝑂 + 𝑙𝑛10 ∙ 𝐶𝑟𝑒𝑑𝐶𝐶𝑂 ∙ 𝜀𝑟𝑒𝑑𝐶𝐶𝑂 ,

where 𝑊 and 𝐹 are the water and lipids volumetric contents, respectively; 𝜇𝑎,𝐻2𝑂 and 𝜇𝑎,𝑓𝑎𝑡 are

the absorption coefficients of water and lipid, respectively; 𝐶𝐻𝐻𝑏 , 𝐶𝐻𝑏𝑂2 , 𝐶𝑜𝑥𝐶𝐶𝑂 , and 𝐶𝑟𝑒𝑑𝐶𝐶𝑂 are

the molar concentrations of HHb, HbO2, oxCCO and redCCO, respectively; and 𝜀𝐻𝐻𝑏 , 𝜀𝐻𝑏𝑂2 ,

𝜀𝑜𝑥𝐶𝐶𝑂 and 𝜀𝑟𝑒𝑑𝐶𝐶𝑂 are the molar extinction coefficients of HHb, HbO2, oxCCO and redCCO,

respectively. 𝑊, 𝐹, 𝐶𝐻𝐻𝑏 , 𝐶𝐻𝑏𝑂2 , 𝐶𝑜𝑥𝐶𝐶𝑂 , and 𝐶𝑟𝑒𝑑𝐶𝐶𝑂 are derived from Giannoni et al9, whereas

the absorption coefficients 𝜇𝑎,𝐻2𝑂 and 𝜇𝑎,𝑓𝑎𝑡 (graphed in Fig. 2a), as well as the molar extinction

coefficients 𝜀𝐻𝐻𝑏 , 𝜀𝐻𝑏𝑂2 , 𝜀𝑜𝑥𝐶𝐶𝑂 and 𝜀𝑟𝑒𝑑𝐶𝐶𝑂 (graphed in Fig. 2b) were derived from Giannoni et

al9 and Prahl et al19. Table 4 summarises the main composition of the 3D in silico model of brain

biopsy used for the MC simulations, based on human grey matter.

Table 4 Composition of the 3D in silico model of brain biopsy used for the MC simulations.
Model composition Values
Water content, 𝑊 70%
Lipid content, 𝐹 10%
Molar concentration of HHb, 𝐶𝐻𝐻𝑏 56.7 µM
Molar concentration of HbO2, 𝐶𝐻𝑏𝑂2 56.7 µM
Molar concentration of oxCCO, 𝐶𝑜𝑥𝐶𝐶𝑂 1 µM
Molar concentration of redCCO, 𝐶𝑟𝑒𝑑𝐶𝐶𝑂 4 µM
HHb = Deoxygenated haemoglobin; HbO2 = Oxygenated haemoglobin; oxCCO = oxidised cytochrome-c-oxidase;
redCCO = reduced cytochrome-c-oxidase.

12
Fig. 2 (a) Absorption coefficients for HbO2, HHb, lipids and water, and scattering coefficient of generic brain tissue

in the visible and NIR range (150 g/L concentration of Hb in blood and blood volume content in generic brain tissue

13
equal to 5% are assumed)8,19; b) Molar extinction coefficients of HbO2, HHb, oxCCO and redCCO in the visible and

NIR range8,17; c) Simulated fluence rates within the 3D cerebral biopsy model at different wavelengths (the plots are

in a logarithmic scale and the absorbing layer was excluded from the plots to maintain a visible contrast). The distri-

bution of the fluence showed a noticeably higher penetration of the light at the NIR wavelengths; d) Simulated mean

photon pathlength within the 3D biopsy model, as a function of wavelength; e) Partial pathlength distributions of the

photons simulated within the 3D biopsy model at various wavelengths (as previously, the absorbing layer was not

included in this figure due to a negligible number of photons passing through it).

A planar, divergent light source was positioned above the 3D biopsy model to simulate the

illumination of the grey matter slab at the same wavelengths of operation of the HyperProbe1 (500-

900 nm, at 5-nm sampling). The direction of the illumination was normal the z axis. A total of 107

photons were simulated for each wavelength band, with emission power equal to 1 mW.

Simulated fluence rates for each wavelength were obtained from the MC simulations, allowing

to visualise and assess the spatial distribution of the photons at the different spectral bands within

the 3D model. Examples of these distributions are depicted in Fig. 2c, demonstrating increasing

depth of penetration of the light within the tissue for longer and longer wavelengths. The depth of

penetration of the photons in the model of biopsy ranged from 10-15% of its thickness (about 0.5-

0.75 mm) for the visible light, up to almost the whole size of the sample (5 mm), for the NIR light.

To further investigate the relationship between wavelength and penetration depth of the light,

the mean pathlength across the tissue of the simulated photons was estimated from the MC simu-

lations for each spectral band, as reported in Fig. 2d. The results further demonstrate that the mean

pathlength travelled by the photon within the tissue increases as the wavelength of the light gets

longer, up to about 5 mm (at 900 nm). Nonetheless, this gradual increase is not homogeneously

continuous, as local extrema are identifiable in Fig 2c, corresponding to the peaks of absorption of

haemoglobin at about 530-570 nm and 770 nm (as visible in Fig. 2a).

14
 Finally, distributions of the partial pathlengths of the simulated photons were estimated for each

wavelength of HyperProbe1: Fig. 2e depicts examples of these distributions across the full range

of the system (the absorbing layer at the bottom is excluded). The results, together with the out-

comes previously illustrated, demonstrate empirically that HyperProbe1 does not simply map the

optical properties of the investigated tissue on its visible surface, but actually reconstruct an inte-

grated distribution on 2D of the spectral features of the biopsy samples across their entire 3D

volume, thanks to the use of both visible and NIR light with different penetration capabilities.

3 Results

3.1 Qualitative and comparative evaluation of the HyperProbe1 data

Preliminary qualitative evaluation of the data collected with HyperProbe1 on the cerebral ex vivo

tissue was conducted, to assess both heterogeneity in the spectra within the same sample, as well

as spectral variability between all the biopsies. For intercomparison across the various samples

described in Table 3, each processed reflectance hypercube 𝑅(𝑥, 𝑦, 𝜆)𝑛 , for 𝑛 = S1…S11, was

averaged spatially over its entire FOV, in order to obtain a single averaged reflectance spectrum

for each biopsies. All these reflectance spectra are shown altogether in Fig. 3a. Overall, the average

reflectance spectra between samples share similar trends, reporting low reflectance in the visible

range (where absorption from haemoglobin is at its highest, as per Fig. 2a), which then tends to

increase gradually towards the NIR range beyond 600 nm, where scattering becomes predominant.

However, significant variability in both magnitudes of the spectra in the same regions and in their

local features are also present, differentiating the signatures of the various samples. Such aspect is

further highlighted by averaging the abovementioned spectra according to their WHO grading:

lower grade glioma (LGG) samples (WHO grade II and III) were grouped together and the mean

15
1 Fig. 3 (a) Intercomparison of averaged reflectance spectra over the entire imaged FOVs of each biopsy sample;

2 (b) Comparison between average reflectance spectra grouping LGG (WHO II, III) against HGG samples (WHO IV);

3 (c) Processed spectral image from HyperProbe1, at 560 nm, of HGG (WHO grade IV) biopsy sample S2, with high-

4 lighted, selected ROIs in the FOV in which average reflectance spectra were calculated; (d) Example of average re-

5 flectance spectra in the corresponding ROIs of the biopsy sample S2; (e) Example of average attenuation spectra in

6 different ROIs of biopsy sample S2, with the portion within the NIR range 780-900 nm enlarged.

16
 7 of their average reflectance spectra across the whole FOVs was compared against the same mean

8 for all the higher grade glioma (HGG) samples (WHO grade IV), as reported in Fig. 3b. It can be

9 seen that there are indeed significant differences in the means between LGG and HGG samples,

10 regarding both local spectral trends as well as magnitude, with the LGG mean presenting higher

11 reflectance on average along the entire spectral range. Conversely, HGG shows larger variance

12 among biopsies of their corresponding group, for the same grade.

13 A number of different regions of interest (ROI) were selected on the reflectance hypercubes

14 𝑅(𝑥, 𝑦, 𝜆) to calculate average reflectance spectra, including different sizes and portions of the

15 FOV reporting identifiable visual features, such as blood clusters. This was done to investigate

16 potential spatial changes in the reflectance spectra across the FOV of the samples. Fig. 3c shows

17 an example of a single, processed reflectance spectral image (as described in Sec. 2.2) collected

18 with HyperProbe1 for arbitrary HGG (WHO grade IV) sample S2, at the bandwidth centred at 560

19 nm, whereas Fig. 3d depicts the corresponding average reflectance spectra. The comparison of the

20 reflectance spectra for each sample on different ROIs highlights significant (from 0.0017 to 0.0378

21 average root mean square deviation (RSMD) for the reflectance curves across all samples) differ-

22 ences in their shapes and trends, as visible for the model case of S2 in Fig. 3d, for two specific

23 spectral ranges: (1) between 510 and 660 nm, and (2) between 780 and 880 nm. In contrast, outside

24 of these ranges, the remainder of the spectral signatures displays homogeneous distribution of the

25 optical properties of the ex vivo samples across the entire FOVs. The largest differences (0.0378

26 of average RSMD across samples) are reported for the ROIs where accumulation of blood are

27 clearly visible. For the first mentioned range (510-600 nm), such results could be correlated to the

28 presence and strong influence of the visible peaks of haemoglobin absorption (Fig. 2a). The influ-

29 ence of the absorption of haemoglobin could also be connected to the reported differences in the

17
30 spectra for the second mentioned range (780-880 nm), which is characterised by a broad peak of

31 absorption from HbO2. However, differences are identified in the biopsy samples also for ROIs

32 not including visible blood clusters: this could then be connected to local differences in the con-

33 centrations of CCO, as the identified range overlaps with the NIR absorption peak of the latter

34 (Fig. 2b), as well as for the increasing weight of the absorption of water and lipids towards the end

35 of the NIR range (Fig. 2a).

36 For a more direct comparison between the spectral signatures of the biopsies reconstructed with

37 HyperProbe1 and the pure optical signatures of the chromophores of interest reported in Fig. 2a

38 and Fig. 2b, we calculated the attenuation spectra 𝐴(𝑥, 𝑦, 𝜆) associated with the contribution of

39 optical absorption and scattering in the tissues, from the reflectance spectra 𝑅(𝑥, 𝑦, 𝜆) in the same

40 ROIs of the samples, using the formula: 𝐴(𝑥, 𝑦, 𝜆) = −𝑙𝑜𝑔10 (𝑅(𝑥, 𝑦, 𝜆)). Fig. 3e reports, for in-

41 stance, the average attenuation spectra of the same HGG sample S2 and the previously selected

42 ROIs. In the range 510-600 nm, the attenuation spectra of the brain tissue samples correlate with

43 the combined profiles of the absorption spectra of HbO2 and HHb, with noticeable peaks at around

44 545-555 nm and at 575 nm. Another peak is also identified in all samples at around 755-760 nm,

45 overlapping the equivalent one from the absorption spectra of HHb. In all samples, attenuation in

46 the range 510-600 nm is reported to increase gradually when shifting from ROIs with no visible

47 accumulations of blood towards ROIs that include the latter at different degrees of covering (as

48 visible for the case of S2 in Fig. 3d). Figure 3e highlights and enhances the visualisation of the

49 attenuation spectra from S2 in the range between 780 and 900 nm, where differences are reported

50 for all the ROIs regardless of the presence of any discernible spatial feature or difference in con-

51 trast. This is occurring largely across all the analysed samples of surgical biopsies. In particular,

52 localised peaks of attenuation corresponding to about 840 nm are identified in a number of ROIs

18
53 in the samples, where the optical absorption of oxCCO is also at its highest. The reported discrep-

54 ancies among concentric ROIs of different sizes in relatively homogenous areas of the ex vivo

55 tissue could be linked to either local variation in the abovementioned chromophore, as suggested

56 by the overlapping of the peaks, or to partial volume effects connected to changes in the optical

57 pathlength travelled by the lights within the sampled regions9.

58 3.2 Quantitative biochemical analysis of the HyperProbe1 data with spectral unmixing

59 As mentioned in Sec. 2.2, we compared the inferred compositions of the expected chromophores

60 from the spectral unmixing algorithm in all the biopsy sample, for two different fitting scenarios:

61 in the whole range from 510 nm to 900 nm, and in the NIR portion of the spectrum from 740 nm

62 to 900 nm. For the first scenario, we obtained satisfactory spectral fits matching the measured

63 attenuation: an example is depicted in Fig. 4a and Fig. 4b, for HGG sample S4 in FOV#1 (WHO

64 grade IV) and LGG sample S10 (WHO grade II), showing also the reconstructed quantitative maps

65 for the total concentration of haemoglobin (HbT), given as the sum of HbO2 and HHb, as well as

66 for the concentration of differential CCO (diffCCO), given as the difference between the concen-

67 trations of oxCCO and redCCO8,17. Blood clusters are resolved with high resolution, due to the

68 haemoglobin peaks in the 500-600 nm range and the expected high concentration and absorbance

69 of haemoglobin (compared to the other known chromophores). Similar well-matching spectral fits

70 via the MBLL are found across all patients, as we report the root mean square error (RMSE) means

71 across all pixels and across all patients to be in the range 0.017 to 0.039, which is of similar mag-

72 nitude to the exemplary RMSEs reported in Fig. 4.

73 A preliminary attempt to classification of the biopsy samples from the inferred hyperspectral

74 results was also performed. As shown in Fig. 4c and Fig. 4d, we found that predicted lipids content

75 allows to separate LGG (grade II and III) biopsies from HGG biopsies (grade IV). Even though

76 19
77 Fig. 4 Inferred HbT and diffCCO concentration maps of HGG S4 FOV#1 (a) and LGG grade S10 (b) samples fitting

78 the whole measured wavelength spectrum (510-900 nm), whilst a model fitting of the observed attenuation for the

79 marked pixel in the HbT image is shown in the third column for both biopsies; (c) Histogram showing probability

80 density distribution of inferred lipid volumetric content of each pixel across different LGG (displayed in green) and

81 HGG (displayed in red) grade samples; (d) Distribution means of the lipid content (reported on x-axis) and diffCCO

82 concentrations (reported on y-axis) suggest that lipid mean content could be able to distinguish grading of all sam-

83 ples, whereas no apparent separation is visible for the inferred mean diffCCO concentrations.

84

20
 85 we observe significant overlap between the different distributions (with an overlap coefficient of

86 49.96% assuming normal distributions), we generally noticed a trend of higher differential lipids

87 content in HGG biopsies, with a mean of -0.466 cm-1. Conversely, the LGG samples were found

88 to have a mean lipid content of -2.53 cm-1. The distinction between the two gradings for all samples

89 was possible by computing the means across all the biopsies and using the -2 cm-1 lipid content

90 difference threshold, as shown in Fig. 4d. This difference in means of the tumour grades was also

91 found to be statistically significant via the Mann-Whitney U test (p=0.017), testing for equal

92 means. Conversely, we did not find evidences of CCO or haemoglobin (established biomarkers

93 for cellular metabolism and haemodynamics, respectively8) to be able to separate gradings of gli-

94 oma samples in the whole range 510-900 nm (as it can be seen in Fig. 4d, for CCO). The overlap

95 coefficient between LGG and HGG samples was found to be considerably larger, with 81.2% and

96 84.6% for inferred diffCCO and HbO2 concentrations, and statistical differences in grading were

97 not observed (p=0.84 and p=0.99, respectively). On a final note, as seen on Fig. 4d, by considering

98 both variables (lipid content and diffCCO concentration) on the 2D distribution plot, an oblique

99 line could arguably even better separate the two glioma grades. A larger sample size will be re-

100 quired to test such more complex hypotheses further.

101 In the second scenario, we estimated inferred compositions within the biopsy samples using

102 exclusively the NIR range between 740 nm and 900 nm, which was chosen to target chromophores

103 that are known for their characteristic absorption profiles in such range, particularly oxCCO and

104 redCCO (as seen in Fig. 2b). Indeed, MBLL has been commonly employed in this specific NIR

105 range to infer differences in metabolic activity using CCO as a biomarker8–10,17. As shown in Fig.5a

106 and Fig. 5b, we again fit the observed signal qualitatively well: the inter-biopsy mean RMSE errors

107 are found to be in the range 0.0151 to 0.296, i.e., the spectral fits of all biopsies can be expected

108
21
109 Fig. 5 Inferred HbT and diffCCO concentration maps of HGG S4 FOV#1 (a) and LGG S10 (b) samples fitting ex-

110 clusively the NIR range (740-900 nm), whilst a model fitting of the observed attenuation for the marked pixel in the

111 HbT image is shown in the third column for both biopsies; (c) Histogram showing probability density distribution of

112 inferred diffCCO concentrations of each pixel across different LGG (displayed in green) and HGG (displayed in

113 red) grade samples; (d) Distribution means of diffCCO and HbO 2 concentrations suggest that these parameters could

114 be able together to distinguish lower and higher grade samples of glioma tissue, with diffCCO being the most accu-

115 rate, across all the investigated samples.

116

22
117 to be similar as observed in Fig.5a and Fig. 5b, as for the previous scenario, albeit being slightly

118 worse due to the expected reduction in SNR at the latter end of the measured spectrum, where

119 scattering of light become predominant. Notable loss in the resolution of various blood clusters

120 can be observed, which was also expected due to the exclusion of the 510-600 nm range with larger

121 haemoglobin absorption peaks. Interestingly, we observed significant differences to the inferred

122 molecular concentrations across biopsies of LGG and HHG that do not match the ones highlighted

123 by the spectral analysis on the whole wavelength range: e.g., in this scenario, the inferred lipid

124 contents were not able to distinguish between the different grades of the samples. We reported an

125 overlap coefficient of 77.1% and no statistical differences in the concentration means for lipids in

126 this scenario (p=0.84). However, we observed average concentrations differences of metabolic

127 diffCCO between LGG and HGG samples, with HGG samples showing lower diffCCO mean con-

128 centrations, as seen in Fig. 5c and Fig. 5d. The diffCCO concentration threshold at 0.35 mM/cm

129 was able to distinguish all LGG and HGG samples by computing the intra-biopsy diffCCO con-

130 centration mean (p=0.017), despite visible overlap (with overlap coefficient of 61%) between the

131 distributions as seen in Fig. 5c. Singularly, we also reported that both oxCCO and redCCO indi-

132 vidually are seemingly not able to distinguish biopsy grading, as we find overlap coefficients of

133 84.6% and 73.3%, and no statistically significant differences (p=0.99 and p=0.12, respectively).

134 This is shown in Fig. 6 plotting probability density distributions of each pixel across all biopsies,

135 divided between LGG and HGG. Only the difference diffCCO between the two inferred concen-

136 trations resulted in the suggested distinction of the two classes of tumoral grades.

137 Furthermore, as depicted in Fig. 5d, we also found an approximately linear correlation in the

138 2D domain between the mean differences in concentration of diffCCO and those of HbO2: together

139 with a reduced DiffCCO mean concentration, HGG samples also present a correlated reduction in

23
140 the mean concentration of HbO2, with the combination of the two biomarkers enhancing the pos-

141 sibility of separating samples grading more effectively.

142 Fig. 6 Histogram showing probability density distribution of inferred oxCCO (a) and redCCO (b) concentrations of

143 each pixel across different LGG (displayed in green) and HGG (displayed in red) grade samples. Neither the in-

144 ferred oxCCO and redCCO density distributions using the NIR range (740-900 nm) suggest to be able to differenti-

145 ate tumour grading, contrary to the diffCCO mean concentrations.

146

147 On a final note, we observed that the HGG sample S9 additionally displays diffCCO character-

148 istics close to LGG samples (Fig. 5d): we hypothesise that there could be different (possibly con-

149 current) reasons for this low distinguishability: even though it has not been shown that HGG sam-

150 ples can show characteristics of LGG samples, the opposite (i.e., high grade-typical histological

151 characteristics in low grade lesions) has recently been reported by Motomura et al23. Thus, in gen-

152 eral, we suggest that it might be possible that lower grade lesions could display characteristics of

153 higher grade lesions, although further investigation on this would be needed.

154 4 Conclusions

155 We presented a novel, transportable HSI device, name HyperProbe1, based on fast spectral scan-

156 ning via a combination of SCL illumination and AOTF filtering. At full operational performances,

24
157 HyperProbe1 can scan sequentially up to 79 wavelengths between 510 and 900 nm, with 5-nm

158 sampling and high spectral resolution (3.5-7 nm of bandwidth), as well as acquire corresponding

159 spectral images on a 0.9 x 0.9-cm2 FOV at high spatial resolution (4.38 μm), in less than 5 minutes.

160 Furthermore, we also provided a preliminary assessment of the data acquired and analysed with

161 HyperProbe1 on a number (n = 11) of fresh surgical samples of glioma from patients at different

162 WHO gradings, including both lower grade (WHO grades II and III) and higher grades (WHO

163 grade IV) tumours. The initial findings demonstrated the capability of the device to reconstruct

164 quantitative maps of the distribution of the concentrations of various chromophores of interest, in

165 particular haemoglobin and CCO (both established biomarkers for tissue haemodynamics and me-

166 tabolism, respectively), as well as lipids volumetric content. We also looked at differences in the

167 inferred contributions of these biomolecules between LGG and HGG biopsies, that may suggest a

168 way to distinguish between the two, and thus potentially provide the basis for a HSI-based meth-

169 odology for tumour classification. We found indeed significant differences between mean lipid

170 contents across samples that could potentially be used to distinguish tumour grading, when fitting

171 over the entire work range (510-900 nm). In particular, HGGs presented higher mean lipid content

172 than LGGs: a plausible biological interpretation of this could be connected to a substantial lipid

173 storage for lipid metabolism in higher grade gliomas, a known characteristic feature of GBM24,25.

174 Similarly, we also found that the analysed HGGs presented inferior concentrations of diffCCO

175 compared to LGGs, when fitting the data exclusively in the NIR range (740-900 nm): this may

176 suggest a potential metabolic path with NIR light to distinguish lower and higher grade samples.

177 This finding may also emphasise the potential role of diffCCO in providing metabolic differences

178 that lies beyond its critical use in NIR spectroscopy (NIRS) applications17. The exact mechanisms

179 by which these concentration differences in diffCCO arise still remain uncertain: as changes in the

25
180 concentration of diffCCO are directly proportional to variations in the metabolic activity of the

181 tissue, one would expect an influence of enhanced hypermetabolism in tumours at higher grades.

182 However, HGGs (such as GBMs) are characterised by the unique presence of necrotic tissue as

183 the core part of the lesion, with the hypermetabolic region occurring only at the borders26,27. Such

184 central necrotic areas, composed almost entirely of dead -i.e., ametabolic- cells could explain the

185 overall reduced mean concentration of diffCCO in HGG compared to LGG, which instead do not

186 present necrosis. Furthermore, the results showed a direct correlation between the reduction of

187 diffCCO in HGG with a corresponding decrease in HbO2, which could also be connected to the

188 hypoxic microenvironment that is specific of this type of brain tumours28,29. From a biological

189 perspective, HGGs, such as GBMs, are known to be triggered and driven in their proliferation by

190 hypoxic microenvironments within the cerebral tissue28,29. Such phenomenon could explain the

191 reduced mean concentration of HbO2, which is a biomarker for oxygenation of the brain, and its

192 correlation with the diffCCO, since a direct association between oxygen delivery and its metabolic

193 consumption has been strongly established30.

194 Future investigations on a larger and enriched cohort of biopsy samples of various types will

195 be needed to further confirm all these results on a more robust statistical basis. In particular, com-

196 parisons with control samples composed of healthy cerebral tissue will also be strongly required

197 for increasing the accuracy and reliability of any prediction.

198 The goal of HyperProbe1 is to provide a first proof-of-concept application to rapid and quanti-

199 tative digital histology of ex vivo tissue from excised surgical biopsies, in particular of cerebral

200 glioma, by reconstructing quantitative maps of the distributions of chromophores of interest in the

201 tissue via fast spectral unmixing algorithms. In this perspective, we demonstrated that Hy-

202 perProbe1 can collect and analyse full-range HSI data of light reflectance from surgical biopsies

26
203 in less than 1 hour after their excision (including preparation of the sample and setting up of the

204 measurements), a significantly much faster time than traditional H&E histopathological screening,

205 which normally takes several days up to few weeks. Such result qualifies HyperProbe1 for an

206 envisioned, future utilisation providing in situ, streamlined, and non-destructive screening of fresh

207 tissue samples, ideally within or just next to the operating theatre. This application could consid-

208 erably benefit the outcome of the surgical treatment and provide an all-optical advancement to

209 current post-surgical histopathological practice.

210 From a technological perspective, we also aim and improving the current setup of HyperProbe1

211 to further improve its performances: we are currently working on extending the operational spec-

212 tral range of the device at both ends, by (1) covering the rest of the available visible range below

213 500 nm and including also part of the near ultraviolet (UV), as well as by (2) expanding the NIR

214 coverage beyond 900 nm. This approach could further enrich the collected spectral data and allow

215 us to potentially target even additional chromophores of interest8.

216 The versatility of HyperProbe1 could also pave the way for potential applications of the instru-

217 mentation to virtually any types of surgical and non-surgical biopsy (or other ex vivo tissues), as

218 well as even move beyond digital histopathology. In this perspective, HyperProbe1 can also be

219 envisioned as an investigative, high-performances HSI device for preclinical in vivo applications:

220 it could be used to explore and tailor features of HSI (such as type and number of wavelengths)

221 that could be translated into clinical settings by specifically engineering more compact, cost-effec-

222 tive and user-friendly medical devices. Within the framework of the HyperProbe project and con-

223 sortium31,32, we indeed aim at translating this technology for its use as a new neuronavigation tool

224 during brain surgery, such as in glioma resection. Such a device would aim at providing an inno-

225 vative approach to guided-neurosurgery, by transforming current practice towards an all-optical,

27
226 real-time, quantitative and accurate imaging approach, that could significantly help neurosurgeons,

227 enhance the efficacy of the treatment, and ultimately improve life expectancy of the patients.

228 Disclosures

229 The authors declare no financial conflict of interest.

230 Code, Data, and Materials

231 In support of open science, the data presented in this article are publicly available on Zenodo at

232 https://zenodo.org/records/10908359. Similarly, all our spectral unmixing code is available on

233 GitHub at https://github.com/HyperProbe/SpectraFit.

234 Acknowledgments

235 The HyperProbe consortium and project has received funding from the European Union’s Horizon

236 Europe research and innovation program under grant agreement No 101071040 – Project Hy-

237 perProbe. Views and opinions expressed are however those of the author(s) only and do not nec-

238 essarily reflect those of the European Union. Neither the European Union nor the granting author-

239 ity can be held responsible for them. AA, FL and IL from UCL are supported by UK Research and

240 Innovation (UKRI) grant No. 10048387

241 References

242 1. M. Titford, “A short history of histopathology technique,” J. Histotechnol. 29(2), 99–110 (2006)

243 [doi:10.1179/his.2006.29.2.99].

244 2. N. Sanai and M. S. Berger, “Surgical oncology for gliomas: The state of the art,” Nat. Rev. Clin.

245 Oncol. 15(2), 112–125 (2018) [doi:10.1038/nrclinonc.2017.171].

246 3. P. Kleihues et al., “Histopathology, classification, and grading of gliomas,” in Glia 15(3), pp. 211–

28
247 221, John Wiley & Sons, Ltd (1995) [doi:10.1002/glia.440150303].

248 4. M. Weller et al., “Glioma,” Nat. Rev. Dis. Prim. 1(1), 1–18 (2015) [doi:10.1038/nrdp.2015.17].

249 5. A. Tarsitano et al., “Intraoperative biopsy of the major cranial nerves in the surgical strategy for

250 adenoid cystic carcinoma close to the skull base,” Oral Surg. Oral Med. Oral Pathol. Oral Radiol.

251 113(2), 214–221, Mosby (2012) [doi:10.1016/j.tripleo.2011.02.014].

252 6. M. J. Van Den Bent, “Interobserver variation of the histopathological diagnosis in clinical trials on

253 glioma: A clinician’s perspective,” Acta Neuropathol. 120(3), 297–304 (2010)

254 [doi:10.1007/s00401-010-0725-7].

255 7. G. Lu and B. Fei, “Medical hyperspectral imaging: a review,” J. Biomed. Opt. 19(1), 010901

256 (2014) [doi:10.1117/1.JBO.19.1.010901].

257 8. L. Giannoni, F. Lange, and I. Tachtsidis, “Hyperspectral imaging solutions for brain tissue

258 metabolic and hemodynamic monitoring: past, current and future developments,” J. Opt. 20(4),

259 044009 (2018) [doi:10.1088/2040-8986/aab3a6].

260 9. L. Giannoni, F. Lange, and I. Tachtsidis, “Investigation of the quantification of hemoglobin and

261 cytochrome-c-oxidase in the exposed cortex with near-infrared hyperspectral imaging: a

262 simulation study,” J. Biomed. Opt. 25(04), 1 (2020) [doi:10.1117/1.JBO.25.4.046001].

263 10. L. Giannoni et al., “A hyperspectral imaging system for mapping haemoglobin and cytochrome-c-

264 oxidase concentration changes in the exposed cerebral cortex,” IEEE J. Sel. Top. Quantum

265 Electron. 27(4), 7400411 (2021) [doi:10.1109/JSTQE.2021.3053634].

266 11. L. Giannoni et al., “Optical characterisation and study of ex vivo glioma tissue for hyperspectral

267 imaging during neurosurgery,” Proc. SPIE 12628, Diffus. Opt. Spectrosc. Imaging IX(1262829),

268 153–159 (2023) [doi:10.1117/12.2670854].

269 12. U. Khan et al., “Trends in Deep Learning for Medical Hyperspectral Image Analysis,” IEEE

270 Access 9, 79534–79548 (2021) [doi:10.1109/ACCESS.2021.3068392].

271 13. G. Florimbi et al., “Towards Real-Time Computing of Intraoperative Hyperspectral Imaging for

272 Brain Cancer Detection Using Multi-GPU Platforms,” IEEE Access 8, 8485–8501 (2020)
29
273 [doi:10.1109/ACCESS.2020.2963939].

274 14. D. N. Louis et al., “The 2021 WHO classification of tumors of the central nervous system: A

275 summary,” Neuro. Oncol. 23(8), 1231–1251, Neuro Oncol (2021) [doi:10.1093/neuonc/noab106].

276 15. F. Lange, L. Giannoni, and I. Tachtsidis, “The Use of Supercontinuum Laser Sources in

277 Biomedical Diffuse Optics: Unlocking the Power of Multispectral Imaging,” Appl. Sci. 11(10),

278 4616 (2021) [doi:10.3390/APP11104616].

279 16. I. Ezhov et al., “Shallow learning enables real-time inference of molecular composition changes

280 from broadband-near-infrared spectroscopy of brain tissue,” arXiv: 2309.16735 (2024).

281 17. G. Bale, C. E. Elwell, and I. Tachtsidis, “From Jöbsis to the present day: a review of clinical near-

282 infrared spectroscopy measurements of cerebral cytochrome-c-oxidase,” J. Biomed. Opt. 21(9),

283 091307 (2016) [doi:10.1117/1.JBO.21.9.091307].

284 18. S. L. Jacques, “Optical properties of biological tissues: a review,” Phys. Med. Biol. 58(11), R37–

285 R61, IOP Publishing (2013) [doi:10.1088/0031-9155/58/11/R37].

286 19. S. A. Prahl, “Optical absorption of hemoglobin,” 1999, <http://omlc.org/spectra/hemoglobin>.

287 20. Q. Fang and D. A. Boas, “Monte Carlo Simulation of Photon Migration in 3D Turbid Media

288 Accelerated by Graphics Processing Units,” Opt. Express 17(22), 20178, Optical Society of

289 America (2009) [doi:10.1364/OE.17.020178].

290 21. L. Yu et al., “Scalable and massively parallel Monte Carlo photon transport simulations for

291 heterogeneous computing platforms,” J. Biomed. Opt. 23(01), 1, Society of Photo-Optical

292 Instrumentation Engineers (2018) [doi:10.1117/1.jbo.23.1.010504].

293 22. A. N. Yaroslavsky et al., “Optical properties of selected native and coagulated human brain tissues

294 in vitro in the visible and near infrared spectral range,” Phys. Med. Biol. 47(12), 2059–2073, IOP

295 Publishing (2002) [doi:10.1088/0031-9155/47/12/305].

296 23. K. Motomura et al., “Clinical characteristics and radiological features of glioblastoma, IDH-

297 wildtype, grade 4 with histologically lower-grade gliomas,” in Brain Tumor Pathology 40(2), pp.

298 48–55, Springer (2023) [doi:10.1007/s10014-023-00458-5].

30
299 24. Y. Kou, F. Geng, and D. Guo, “Lipid Metabolism in Glioblastoma: From De Novo Synthesis to

300 Storage,” in Biomedicines 10(8), p. 1943, Multidisciplinary Digital Publishing Institute (2022)

301 [doi:10.3390/biomedicines10081943].

302 25. E. Verdugo, I. Puerto, and M. Á. Medina, “An update on the molecular biology of glioblastoma,

303 with clinical implications and progress in its treatment,” in Cancer Communications 42(11), pp.

304 1083–1111, John Wiley & Sons, Ltd (2022) [doi:10.1002/cac2.12361].

305 26. L. Persano et al., “The three-layer concentric model of glioblastoma: Cancer stem cells,

306 microenvironmental regulation, and therapeutic implications,” ScientificWorldJournal. 11, 1829–

307 1841, ScientificWorldJournal (2011) [doi:10.1100/2011/736480].

308 27. S. M. Markwell et al., “Necrotic reshaping of the glioma microenvironment drives disease

309 progression,” in Acta Neuropathologica 143(3), pp. 291–310, Acta Neuropathol (2022)

310 [doi:10.1007/s00401-021-02401-4].

311 28. A. R. Monteiro et al., “The role of hypoxia in glioblastoma invasion,” in Cells 6(4), p. 45,

312 Multidisciplinary Digital Publishing Institute (2017) [doi:10.3390/cells6040045].

313 29. M. Domènech et al., “Hypoxia: The cornerstone of glioblastoma,” in International Journal of

314 Molecular Sciences 22(22), p. 12608, Multidisciplinary Digital Publishing Institute (2021)

315 [doi:10.3390/ijms222212608].

316 30. M. Caldwell et al., “Modelling blood flow and metabolism in the preclinical neonatal brain during

317 and following hypoxic-ischaemia,” PLoS One 10(10), e0140171, Public Library of Science (2015)

318 [doi:10.1371/journal.pone.0140171].

319 31. L. Giannoni et al., “HyperProbe consortium: innovate tumour neurosurgery with innovative

320 photonic solutions,” in https://doi.org/10.1117/12.2670764 12628, p. 47, SPIE (2023)

321 [doi:10.1117/12.2670764].

322 32. “HyperProbe – HyperProbe Project Website,” <https://hyperprobe.eu/> (accessed 27 March 2024).

31
 323 Luca Giannoni is a postdoctoral researcher at the European Laboratory for

324 Non-Linear Spectroscopy (LENS) of the University of Florence. He received

325 his PhD in Medical Imaging from University College London (UCL) in

326 2020. His current research interests focus on developing cutting-edge optical

327 microscopy systems for neuroimaging, as well as on designing and validat-

328 ing cost-effective, compact HSI devices for clinical translation. He is a member of SPIE.

329 Marta Marradi is a PhD student at the European Laboratory for Non-

330 Linear Spectroscopy (LENS) of the University of Florence. Her work spe-

331 cifically focuses on the development of novel imaging systems based on

332 both multispectral and hyperspectral techniques for biomedical applica-

333 tions, concentrating in the dermatologic and neurological fields.

334 Ivan Ezhov is a research scientist at the Institute of Artificial Intelligence

335 in Medicine at the Technical University of Munich. His research focus is at

336 the intersection of biophysical modelling, inverse problems, and deep learn-

337 ing. His interests include computational oncology, physics-based machine

338 learning and spectral unmixing of large scale datasets, such as medical hyperspectral images.

339 Camilla Bonaudo is a neurosurgeon at the Azienda Ospedaliera of Careggi,

340 and a PhD student in Neuroscience at the University of Florence. Her work

341 focuses on the development of new protocols for the application of technol-

342 ogies, such as Navigated Transcranial Magnetic Stimulation (nTMS), to

343 study cognitive functions, acquiring data about brain plasticity and functional

344 reshaping. Her aim is to study brain connectome and comparing functional mapping with the gold

345 standard of the Direct Cortical Simulation.

32
 346 Angelos Artemiou is an MRes student at University College London (UCL),

347 specializing in the field of Medical Physics and Biomedical Engineering. His

348 research is primarily focused on developing novel phantoms for the metrolog-

349 ical characterisation of NIR and hyperspectral instrumentation.

350 Anam Toaha is a PhD student at the European Laboratory for Non-Linear

351 Spectroscopy (LENS) of the University of Florence. She conducts research

352 in the field of biomedical imaging and spectroscopy. She is also part of the

353 HyperProbe project, working on the development of a novel optical imag-

354 ing device based on hyperspectral imaging, to advance brain surgery by

355 presenting neurosurgeons with enhanced information intraoperatively.

356 Frédéric Lange received his PhD from the University of Lyon and INSA

357 de LYON in 2016. He is now a senior research associate with the Biomed-

358 ical Optics Research Laboratory, Department of Medical Physics and Bio-

359 medical Engineering at University College London. His current main re-

360 search interests are in the development of novel optical technologies to

361 monitor tissue’s oxygenation and metabolism, with a specific interest for non-invasive brain mon-

362 itoring in healthy (i.e., brain development/neuroscience) and pathological conditions.

363 Bruno Montcel is a professor at University Claude Bernard Lyon 1. He

364 leads research focused on optical medical devices at CREATIS laboratory

365 and is the chairman of the Biomedical Engineering Department of Polytech

366 Lyon. Its research explores optical imaging methods and experimental set

367 up for the exploration of tissue physiology and pathologies. It mainly fo-

33
368 cuses on intraoperative and point of care hyperspectral optical imaging methods for medical diag-

369 nosis and gesture assistance.

370 Alessandro Della Puppa is a professor of neurosurgery, a neurosurgeon,

371 and the head of the Neurosurgical Department at the Azienda Ospedaliera

372 of Careggi, in Florence. His major interests concern the surgical improve-

373 ment of neuro-oncology, in particular regarding the development of new

374 intraoperative strategies in awake surgery, in immuno-neuro-oncology, and

375 in navigational TMS in the preoperative and intraoperative mapping of language and cognitive

376 functions, all aiming towards a patient-tailored rehabilitation program.

377 Ilias Tachtsidis is a professor at University College London. He leads the

378 Multimodal Spectroscopy and MetaboLight groups in the Biomedical Op-

379 tics Research Laboratory at the department of Medical Physics and Bio-

380 medical Engineering. Tachtsidis is a multidisciplinary scientist with a re-

381 search portfolio encompassing engineering, physics, computing, neurosci-

382 ence and clinical medicine. His research focus is technology development in optical neuroimaging

383 (devices, algorithms), through applications (clinical, neuroscience), to data analytics towards gen-

384 erating clinical information and knowledge (computational models).

385 Daniel Rückert is Alexander von Humboldt Professor for artificial intelli-

386 gence (AI) in Medicine and Healthcare at the Technical University of Mu-

387 nich, where he directs the Institute for AI and Informatics in Medicine. He is

388 also a professor in the Department of Computing at Imperial College Lon-

389 don. From 2016 to 2020 he served as Head of the Department at Imperial

390 College. His research focuses on medical image computing, data science and AI in medicine.

34
 391 Francesco Saverio Pavone is a professor at the University of Florence. He

392 is the director of the Biophotonics area at the European Laboratory for

393 Non-Linear Spectroscopy, focusing on optical microscopy and imaging.

394 Pavone is also coordinator of several EU projects, including the Hy-

395 perProbe consortium. He participates in the NIH BRAIN initiative and

396 HBP Brain Initiative. He is also the Italian node leader of EBRAIN and the Italian delegate in the

397 Board of Directors of EuroBioImaging.

398 Biographies and photographs for the other authors are not available.

399
400 Caption List
401
402 Fig. 1 (a) Schematics of HyperProbe1; (b) Picture of the illumination side of HyperProbe1; (c)

403 Picture of the imaging side of HyperProbe1; (d) Testing of the spatial resolution of HyperProbe1;

404 (e) Line profile of the smallest resolved element; (f) Example of glioma biopsy sample.

405 Fig. 2 (a) Absorption coefficients for HbO2, HHb, lipids and water, and scattering coefficient of

406 generic brain tissue; b) Molar extinction coefficients of HbO2, HHb, oxCCO and redCCO; c) Sim-

407 ulated fluence rates within the 3D biopsy model at different wavelengths; d) Simulated mean pho-

408 ton pathlength within the 3D biopsy model, as a function of wavelength; e) Partial pathlength

409 distributions of the photons simulated within the 3D biopsy model at various wavelengths.

410 Fig. 3 (a) Intercomparison between average reflectance spectra across all samples; (b) Comparison

411 of average reflectance spectra between LGG and HGG samples; (c) Example of spectral frame at

412 560 nm, acquired with HyperProbe1 on biopsy sample S2; (d) Example of average reflectance

413 spectra in different ROIs of the HyperProbe1 data; (e) Example of average attenuation spectra in

414 different ROIs of the HyperProbe1 data;

415 Fig. 4 Spectral fit and corresponding inferred HbT and diffCCO concentration maps of both a

35
416 HGG (a) and a LGG sample (b), fitting the whole wavelength spectrum (510-900 nm); (c) Histo-

417 gram showing density distribution of inferred lipid content of each pixel across different LGG and

418 HGG samples; (d) Distribution means of the lipid contents and diffCCO concentrations suggest

419 that lipid mean content could be able to distinguish grading of all samples, whereas no apparent

420 separation is visible for the inferred mean diffCCO concentrations.

421 Fig. 5 Spectral fit and corresponding inferred HbT and diffCCO concentration maps of both a

422 HGG (a) and a LGG sample (b), fitting exclusively the NIR range (740-900 nm); (c) Histogram

423 showing density distribution of inferred diffCCO concentrations of each pixel across different

424 LGG and HGG samples; (d) Distribution means of diffCCO and HbO2 concentrations suggest that

425 these parameters could be able to distinguish LGG and HGG samples of glioma tissue, with

426 diffCCO being the most accurate, across all the samples.

427 Fig. 6 Histogram showing probability density distribution of inferred oxCCO (a) and redCCO (b)

428 concentrations of each pixel across different LGG (displayed in green) and HGG (displayed in

429 red) grade samples, showing no significance differences in either cases for the range 740-900 nm.

430 Table 1 List of components of HyperPRobe1 with specifications.

431 Table 2 Technical characteristics and features of HyperProbe1.

432 Table 3 Classification of the tissue samples investigated with HyperProbe1.

433 Table 4 Composition of the 3D in silico model of brain biopsy used for the MC simulations.

36