MINERVA Project, mid- To near Infrared Spectroscopy

for Improved Medical Diagnostics

Valery Naranjo

, Francisco Pe

naranda

, Mariano Alca

niz

, Bruce Napier

Mark Farries

, Gary Stevens

, John Ward

, Cestmir Barta

, Radek Hasal

Angela Seddon

, Slawomir Sujecki

, Samir Lamrini

, Uffe Møller

, Ole Bang

Peter M. Moselund

, Munir Abdalla

, Danny De Gaspari

, Rosa M. Vinella

Hedda Malm

, Gavin R. Lloyd

, Nick Stone

, Jayakrupakar Nallala

Juergen Schnekenburger

, Lena Kastl

and Bj

orn Kemper

Inst. Int. de Investigaci

on en Bioingenier

ıa y Tecnolog

ıa Orientada al Ser Humano,

Universitat Polit

ecnica de Valencia, Valencia, Spain

Vivid Components, Dr.-R

orig-Damm 22, 33102, Paderborn, Germany

Gooch & Housego (Torquay) Ltd., Broomhill Way, Torquay, Devon, TQ2 7QL, U.K.

Gooch & Housego (UK) Ltd., Dowlish Ford, Ilminster, Somerset, TA19 OPF, U.K.

BBT-Materials Processing SRO, Doubicka 11, Praha 8, 184 00, Czech Republic

George Green Institute for Electromagnetics Research, Faculty of Engineering,

University Park, University of Nottingham, Nottingham, NG7 2RD, U.K.

LISA Laser Products OHG, Fuhrberg & Teichmann Max-Planck-Str. 1,

37191, Katlenburg-Lindau, Germany

DTU Fotonik, Department of Photonics Engineering, Technical University of Denmark,

2800 Kgs. Lyngby, Denmark

NKT Photonics A/S, Blokken 84, 3460, Birkerød, Denmark

Xenics NV, Ambachtenlaan 44, BE-3001, Leuven, Belgium

IR Nova AB, Electrum 236, 164 40, Kista, Sweden

Biophotonics Research Unit, Gloucestershire Hospitals NHS Foundation Trust,

Gloucester, U.K.

Department of Physics, Exeter University, Exeter, U.K.

Biomedical Technology Center, University of Muenster, D-48149, Muenster, Germany

vnaranjo@labhuman.com

Abstract. The main idea behind the MINERVA project is the recognition that

for the ﬁrst time, through breakthroughs in photonic technology, it is possible to

open the mid-IR electromagnetic spectrum (3-12 µm) for rapid medical imag-

ing. In particular this could greatly improve the chances of early cancer diagno-

sis. MINERVA will exploit and develop the advances in soft glass optical ﬁbres,

acousto-optic (AO) modulator design, crystal growth, ﬁbre lasers, supercontin-

uum sources and detectors in the mid-IR. Two speciﬁc high impact applications

will be addressed: high volume pathology screening (i.e. automated microscope-

based examination of routine patient samples) and human skin surface examina-

tion (i.e. non-invasive investigation of suspected skin cancer). In an Integrating

Project of this scale it is possible to pursue several targets in parallel, each of

which alone brings signiﬁcant beneﬁts. Together they could begin a new branch

of the bio-medical imaging industry.

Naranjo V.

MINERVA Project, mid- To near Infrared Spectroscopy for Improved Medical Diagnostics.

DOI: 10.5220/0006162400530069

In European Project Space on Intelligent Systems, Pattern Recognition and Biomedical Systems (EPS Lisbon 2015), pages 53-69

ISBN: 978-989-758-095-6

1 Introduction

MINERVA is a project funded by the European Commission through its Seventh Frame-

work Programme (FP7) [1]. It brings together thirteen partners from across Europe with

the common objective of developing mid-infrared (mid-IR) technology to improve the

early diagnosis of cancer (Fig. 1).

Mid-IR radiation is an exciting new area for real-time molecular-sensing with ap-

plications in different areas: medicine and healthcare (e.g. early cancer detection: the

MINERVA application space), environment and energy (e.g. monitoring exhaust gases),

security (e.g. detection of narcotics or explosives, food security), chemical and indus-

trial manufacturing (e.g. process control and quality assurance).

The MINERVA mid-IR range (1.5 to 12 µm) is rich in spectroscopic absorption

peaks of biomolecules such as fats, proteins and carbohydrates. In particular it has been

shown that, through the latest data analysis techniques, this spectral region can be used

to identify the presence of early cancer. Currently there is a lack of practical sources

and components for this spectral region, and so these mid-IR diagnostic techniques are

restricted to laboratory demonstrations.

MINERVA aims to develop ﬁbre, lasers and broadband sources, components, mod-

ulators and detectors to access this important part of the spectrum. In parallel it will

identify analytical techniques using the new photonic hardware to improve early skin

cancer diagnosis and the rapid and automatic assessment of biopsy samples using a

microscope.

Fig. 1. Logos of the thirteen partners of MINERVA’s consortium.

1.1 mid-IR Spectroscopy: A New Tool for Pathologists

The spectral region studied in MINERVA (1.5-12 µm) includes the so-called “ﬁnger-

print region” in which many biomolecules have tell-tale absorption peaks. By studying

the pattern of absorbed radiation it is possible to deduce details of the type and dis-

tribution of these molecules, which in turn provides important information for disease

diagnosis.

It is emphasised that this process is not as straightforward as simply spotting certain

chemicals, or “cancer markers”. The information is buried in the inter-related distri-

bution of species and subtle biochemical changes. It requires a powerful mathematical

EPS Lisbon January 2015 2015 - European Project Space on Intelligent Systems, Pattern Recognition and Biomedical Systems

technique known as multi-variate analysis to extract useful information from the reams

of spectral data in order to spot the warning signs of cancer.

One form of multi-variate analysis is correlation mapping, which enables the visu-

alisation of diseased cells or regions from spectral data (Fig. 2). MINERVA combines

novel mid-IR spectroscopy with correlation mapping and hopes to lead to a break-

through diagnostic technology.

2www.minerva-project.eu

MId- to NEaR infrared spectroscopy

for improVed medical diAgnostics

MINERVA newsletter #1 July 2013

Mid-IR spectroscopy: a new tool for pathologists

The spectral region studied in MINERVA (1.5-12 μm) includes the so-called “fingerprint

region” in which many biomolecules have tell-tale absorption peaks. By studying the pattern

of absorbed radiation it is possible to deduce details of the type and distribution of these

molecules, which in turn provides important information for disease diagnosis.

It is emphasised that this process is not as straightforward as simply spotting certain

chemicals, or “cancer markers.” The information is buried in the inter-related distribution of

species and subtle biochemical changes. It requires a powerful mathematical technique

known as multi-variate analysis to extract useful information from the reams of spectral data

in order to spot the warning signs of cancer.

One form of multi-variate analysis is correlation mapping, which enables the visualisation of

diseased cells or regions from spectral data. MINERVA combines novel mid-IR spectroscopy

with correlation mapping and hopes to lead to a breakthrough diagnostic technology.

MINERVA will develop a suite of mid-IR photonic hardware to improve access to this

information. Working in the mid-IR is extremely challenging, and MINERVA will need to break

new ground in several technical areas:

• The project coordinator (G&H) will develop mid-IR components such as fused combiners

(glass devices used to combine or separate signals into different optical fibres), and

acousto-optic modulators (to switch the signals and separate wavelengths at high speed).

• These AO devices will need new types and sizes of calomel crystals from BBT.

Mid-IR spectroscopy

Correlation mapping

CaF

Collagen I

Collagen III

DNA

Oleic acid

Albumin

20 40 60 80 100

Images courtesy of

Gloucestershire Hospitals

NHS Foundation Trust

• Mid-IR glass fibre to carry the radiation efficiently and conveniently

is being produced at University of Nottingham.

• Novel pump lasers at 2.9 μm and 4.5 μm from LISA Laser will be

used by DTU and NKT to generate a range of supercontinuum

sources in ZBLAN, indium fluoride and chalcogenide glasses,

spanning the MINERVA range from 1.5 μm to 12 μm.

• Xenics and IRnova are advancing the state-of-the-art in Type II

superlattice detectors, which offer a cost effective route to highly

efficient detection in the mid-IR.

• University of Exeter and Gloucestershire Hospitals NHS Trust will

develop the multivariate algorithms and techniques for high volume

screening of human samples.

• WWU Muenster will develop a skin cancer diagnostic process.

• UPVLC (Valencia) is working on the user interface and visualisation.

• The project is managed and administrated by Vivid Components.

Fig. 2. Correlation mapping enables the visualisation of diseased cells or regions from spectra.

MINERVA will develop a suite of mid-IR photonic hardware to improve access to

this information. Working in the mid-IR is extremely challenging, and MINERVA will

need to break new ground in several technical areas:

– Gooch & Housego (G&H), the project coordinator, will develop mid-IR compo-

nents such as fused combiners (glass devices used to combine or separate signals

into different optical ﬁbres), and acousto-optic (AO) modulators (to switch the sig-

nals and separate wavelengths at high speed).

– These AO devices will need new types and sizes of calomel crystals from BBT-

Materials Processing SRO (BBT).

– Mid-IR glass ﬁbre to carry the radiation efﬁciently and conveniently is being pro-

duced at University of Nottingham (NOTT).

– Novel pump lasers at 2.9 µm and 4.5 µm from LISA Laser Products OHG (LISA)

will be used by the Technical University of Denmark (DTU) and NKT Photonics A/S

(NKT) to generate a range of supercontinuum sources in ZBLAN, indium ﬂuoride

and chalcogenide glasses, spanning the MINERVA range from 1.5 µm to 12 µm.

– Xenics and IRnova are advancing the state-of-the-art in Type II superlattice detec-

tors, which offer a cost effective route to highly efﬁcient detection in the mid-IR.

– University of Exeter and Gloucestershire Hospitals NHS Trust (GHNT) will de-

velop the multivariate algorithms and techniques for high volume screening of hu-

man samples.

– Westfaelische Wilhelms-Universitaet Muenster (WWU) will develop a skin cancer

diagnostic process.

– Universitat Polit

ecnica de Val

encia (UPV) is working on novel algorithms for the

analysis of histopathological images and the recognition and classiﬁcation of hiper-

spectral data of cancer samples.

– The project is managed and administrated by Vivid Components.

In the next sections it will be presented an overview of the expectancies of the

project and the main preliminary advances reached by the different groups.

MINERVA Project, mid- To near Infrared Spectroscopy for Improved Medical Diagnostics

2 Mercurous Halides: Unique Acousto-Optic Materials for IR

from BBT

BBT is a world leader in the growth and processing of Mercurous Chloride (Hg

Calomel) single crystals with excellent AO properties and is thus in favourable posi-

tion to address this question. Calomel single crystals (Fig. 3.a) exhibit a wide range of

optical transmission, high indices of refraction, extremely high value of acousto-optic

ﬁgure of merit M2, very low velocity of shear acoustic wave, high value of birefrin-

gence (four times higher than Calcite), etc. The Calomel crystals are well adapted to

fabricate acousto-optic devices operating in the mid and far IR (3 to 20 µm).

At the beginning of the MINERVA project the production technology enabled the

growth of calomel crystal boules with a diameter of 26 to 29 mm (Fig. 3.b) and length

of 45 to 60 mm (typically 55 mm). Within the MINERVA project the new technology

is being developed enabling the growth of cylindrical crystal boules with a diameter up

to 35 mm, which is necessary for the further manufacturing of acousto-optical tuneable

ﬁlters of new design proposed by G&H.

The Calomel crystal growth process is highly demanding, difﬁcult and complex, es-

pecially in case of bigger 35mm boules. The growing process is powered by a dynamic

temperature ﬁeld and corresponding axial and radial temperature gradients. The whole

process has to be carefully maintained within narrow physical condition limits. All the

equipment and accessories have to be newly developed by BBT and adjusted to the spe-

ciﬁc conditions for growing of the 36mm diameter crystals including the temperature

controllers. These controllers are equipped with brand new cultivation programs with

respect to the bigger material mass. A total of six crystallizers will be built within the

project. Currently, four units are operational and tested (Fig. 3.c).

Fig. 3. (a) Polished Calomel AOTF substrate. (b) Growing Calomel crystal, diameter 28mm. (c)

Two cultivation crystallizer units with Calomel crystals.

EPS Lisbon January 2015 2015 - European Project Space on Intelligent Systems, Pattern Recognition and Biomedical Systems

3 Er:ZBLAN Fibre Laser at 2.9 µm from University of

Nottingham and LISA

The partners NOTT and LISA will develop a 2.9 µm laser based on Er-doped ZBLAN

ﬁbres diode-pumped at 976 nm. This ﬁbre laser will be used as pump source for ultra-

long wavelength supercontinuum generation (3-9 µm). The ﬁrst step is the develop-

ment of a ﬁbre laser in an external cavity conﬁguration. For that purpose a simulation

model based on the rate equation and signal propagation equations is implemented by

the NOTT group (Fig. 4). Different parameters will be studied, e.g. absorption cross-

sections, emission cross-sections, and gain cross-sections, to predict the optimum laser

performance.

3www.minerva-project.eu

MId- to NEaR infrared spectroscopy

for improVed medical diAgnostics

MINERVA newsletter #3 May-2014

Er:ZBLAN fibre laser at 2.9 µm

The partners LISA Laser Products OHG (LISA) and Dr.

Slawomir Sujecki‘s team at the University of Nottingham

(NOTT) will develop a 2.9 µm laser based on Er-doped

For more info contact Dr. Samir Lamrini

SLamrini@lisalaser.de

Above: Set-up of the Er:ZBLAN fibre for

absorption studies. The green

fluorescence results from up-conversion

processes.

Above: Set-up of the Er:ZBLAN fibre in an external

cavity configuration pumped with high-power fibre-

coupled diodes. First experiments showed a good

agreement with the simulations carried out at

Nottingham.

Right: Modelling scheme of the

Er:ZBLAN fibre laser with an external

cavity configuration. For the exact

prediction both the 2.9 µm and the

1.6 µm laser signal were analysed in

forward and backward propagation.

ZBLAN fibres diode-pumped at 976 nm. This fibre laser will be used as pump source for ultra-

long wavelength supercontinuum generation(3-9µm).Thefirststepisthedevelopment of a

fibre laser in an external cavity configuration. For that purpose a simulation model based on the

rate equation and signal propagation equations is implemented by the NOTT group (see sketch

above). Different parameters will be studied, e.g. absorption cross-sections, emission cross-

sections, and gain cross-sections, to predict the optimum laser performance.

In parallel, LISA will carry out experiments for the handling (stripping, cleaving, splicing) of the

soft glass fibre and target both high-power and high-energy laser operation with different

resonator configurations. Coated focussing and collimating optics have to be designed and

manufactured for the laser studies. After the evaluation of the first tests in CW operation LISA‘s

scientists and engineers will design a compact and robust cooled housing for the 2.9 µm laser.

Regarding high-energy operation special acousto-optic modulators (AOM) based on TeO

will

be designed and built by G&H and delivered to LISA.

Fig. 4. Modelling scheme of the Er:ZBLAN ﬁbre laser with an external cavity conﬁguration. For

the exact prediction both the 2.9 µm and the 1.6 µm laser signal were analysed in forward and

backward propagation.

In parallel, LISA will carry out experiments for the handling (stripping, cleaving,

splicing) of the soft glass ﬁbre and target both high-power and high-energy laser oper-

ation with different resonator conﬁgurations (Fig. 5).

Fig. 5. (a) Set-up of the Er:ZBLAN ﬁbre for absorption studies. The green ﬂuorescence results

from up-conversion processes. (b) Set-up of the Er:ZBLAN ﬁbre in an external cavity conﬁgura-

tion pumped with high-power ﬁbre-coupled diodes. First experiments showed a good agreement

with the simulations carried out at Nottingham.

Coated focussing and collimating optics have to be designed and manufactured for

the laser studies. After the evaluation of the ﬁrst tests in CW operation LISA’s scientists

and engineers will design a compact and robust cooled housing for the 2.9 µm laser.

MINERVA Project, mid- To near Infrared Spectroscopy for Improved Medical Diagnostics

Regarding high-energy operation special acousto-optic modulators (AOM) based on

TeO2 will be designed and built by G&H and delivered to LISA.

Further information about MINERVA’s ﬁbre laser can be found in [2–10].

4 Extreme IR Supercontinuum Modelling at DTU

DTUs team has the task of ﬁbre modelling in MINERVA in close collaboration with

the ﬁbre manufacturing group at NOTT. The DTU group also models dynamic super-

continuum generation along the ﬁbres using both measured material data and calculated

ﬁbre properties. This advanced modelling requires extensive computational resources

in order to accurately follow the rapid spectral broadening, which covers over four oc-

taves (from 1 µm to 16 µm); made possible by the strong non-linearity of chalcogenide

glasses and the extremely high numerical aperture (NA) of the Nottingham ﬁbres. Fig-

ure 6 shows two graphics with some results of the numerical modeling of mid-IR su-

percontinuum generation. Thorough analysis of the modelling has been presented in

[11–13].

Fig. 6. (a) Zero dispersion wavelengths versus core diameter for step-index ﬁbres (based on ﬁbres

fabricated at the University of Nottingham) with NA as given in the legend. (b) Modelling shows

that a ﬁbre with core diameter 10 µm and NA = 1.0 exhibiting no second zero dispersion is

optimum. Super-continuum generation beyond 12 µm is observed numerically.

5 MINERVA Supercontinuum Sources from NKT

It has been mentioned that the mid-IR region contains a wealth of spectral data which

can yield important information on the chemical composition of samples from gases

and liquids to living cells. However, the investigation of this topic has been limited by

the available photonic sources. Researchers had to choose between a very low intensity

broadband source such as a “globar” thermal source, or a high intensity but narrowband

source, such as a laser diode.

NKT Photonics is dedicated to providing ﬂexible sources of high intensity light in

an easy to use format (Fig. 7.a). It has already established commercial supercontin-

uum systems which can deliver any wavelength from 400 to 2000 nm on demand. It

has recently launched the EXTEND-UV accessory which can extend the wavelength

EPS Lisbon January 2015 2015 - European Project Space on Intelligent Systems, Pattern Recognition and Biomedical Systems

coverage to cover the 270-400 nm UV region. The company would now like to push

the limits of supercontinuum sources at longer wavelengths, reaching into the mid-IR

region.

In MINERVA NKT is developing zirconium ﬂuoride (ZrF

) glass ﬁbre supercon-

tinuum sources to cover the 1.5-4.5 µm spectrum. Subsequently it will investigate even

longer wavelengths by utilising newly developed indium ﬂuoride (InF

) ﬁbres to extend

the spectrum beyond the transmission band of ZrF

glasses.

These sources could detect changes in cells by monitoring absorption in the 2.6-

3.8 µm region which relates to the balance between lipids and proteins (Fig. 8.a). The

increase in wavelength from 4.5 µm up to >5 µm would make it possible to interrogate

additional important gas absorption lines such as carbon monoxide.

Fig. 7. (a) Schematic of a typical NKT supercontinuum source. (b) Graph showing absorption

spectra of some key bio-molecules in the lower end of the “ﬁngerprint region”.

In the ﬁrst 2 years of the MINERVA project NKT has already developed several

supercontinuum sources with output power of up to 2.5 W. These sources are more

than a million times brighter than most thermal light sources and even brighter than a

Synchrotron. We have shown the limits of zirconium-ﬂuoride based systems by setting a

new record for the longest wavelength supercontinuum generated at 4.75 µm. However,

the chemometric specialists in MINERVA found that the main region of interest was the

2.5-3.8 µm region so we have also shown how the main power in the output spectrum

can be shifted down to the main region of interest by altering the design of the nonlinear

ﬁber. These ﬁrst Supercontinuum sources are already at work in the development of the

MINERVA-lite microscopy setup which will soon be applied to bio sample imaging.

Meanwhile NKT is pushing onward in the development of the mid-IR supercon-

tinuum sources. The initial sources were based on rather long pulse nanosecond pump

lasers with relatively low pulse repetition frequencies. This made the sources incompat-

ible with most of the Fourier transform spectrometers (FTIRs) that many researchers

use in the mid-IR region. In addition, the low repetition rate made it time consuming to

counter any noise in the source by averaging over many pulses. NKT is therefore now

developing sources based on much shorter pump pulses and with higher repetition rate

in order to reduce noise and make the sources compatible with standard FTIRs.

As these mid-IR supercontinuum sources become available and known in the ﬁeld,

the MINERVA consortium expects the emergence of new markets. For example, an

MINERVA Project, mid- To near Infrared Spectroscopy for Improved Medical Diagnostics

important spectroscopic application in the petrochemical industry is to monitor single

wavelengths in the 3-3.5 µm band in order to optimise the reﬁning processes. Monitor-

ing the whole spectrum simultaneously would allow a full real-time chemical analysis

of the output chemicals.

Some relevant references concerning the supercontinuum sources within the MIN-

ERVA framework have been already published [14–24]

6 MINERVA type-II Superlattice IR Detectors from IRnova

Type-II superlattice (T2SL) is a material/technology that can be used for high quality

cooled photon detectors, with tailorable bandgap from 2 µm and upwards. The name

comes from the fact that the conduction and valence bands display a so-called “bro-

ken type-II” (sometimes also called “type-IIb” or even “type-III”) alignment between

the constituent materials, which can be InAs/GaSb/AlSb, or alloys thereof (Fig. 8.a).

In contrast to typical quantum well devices, e.g. the active regions of semiconductor

lasers, the superlattice layers in the T2SL material are so thin (typically 3 nm) that

mini-bands are formed in the material. These mini-bands resemble the conduction and

valence bands of a bulk semiconductor material. By carefully selecting the superlat-

tice layer thicknesses and compositions, novel materials can be deﬁned to meet widely

different needs.

irn054479-2

Utskrift: 2013-05-20

(Mallutgåva 2011-02-02)

Date/Datum

2013-05-20

Rev

Page/Sida

Approved/Godkänd

Type of document/Dokumenttyp

REPORT

Issued by/Utfärdare

Carl Asplund / Henk Martijn

Distribution

For information only/För kännedom

IR detection using type-II superlattice photodiodes

T2SL (Type II SuperLattice) or sometimes also called SLS (Strained Layer Superlattice) is a material / technology

that can be used to make high quality cooled infrared photon detectors with a cut-off wavelength ranging from 2 µm

to 30 µm. This covers the SWIR, MWIR, LWIR and VLWIR wavelength bands, loosely defined as 2-3 µm, 3-5 µm,

8-12 µm and >12 µm, respectively.

A superlattice is a system made of a repeating sequence of thin layers of different materials. If the layer thicknesses

are small enough in a quantum mechanical sense, minibands are formed in the material. The result is an artificial

material with properties that can be engineered; in the detector case the bandgap energy corresponding to the desired

cut-off wavelength. When two semiconductors are brought in contact, there are several ways the valance and

conduction bands can align. If both the valence and the conduction band edge of the second material are above the

band edges of the first material, it is called a broken type II band alignment.

GaSb

InAs

GaSb GaSb

InAs

GaSb

InAs

GaSb GaSb

InAs

Figure 1 Band alignment of InAs GaSb and the forming of minibands.

The III/V compound materials InAs and GaSb form such a band alignment (see Figure 1). As can be seen in Figure 2,

InAs has a lattice mismatch of less than 1% on GaSb. Starting with GaSb substrates alternating layers of InAs and

GaSb with atomic layer precision can be deposited using MBE (Molecular Beam Epitaxy). By interface engineering

(create an interface layer of InSb) or using more complicated superlattices like Ga

1-x

Sb/InAs, thick strain

compensated structures with high crystal quality can be grown. If doping in the form of trace amounts of Be, Te or Si

is incorporated, photovoltaic p-i-n structures that can be used to detect IR radiation of the desired wavelength are

formed.

Fig. 8. (a) Schematic of T2SL band-gap structure. (b) Detector/Dewar/cooler assembly for T2SL

from IRnova.

Compared with a traditional bulk material for the 3-5 µm range, such as InSb, T2SL

requires less cooling and thus draws less power, which allows for longer cooler lifetime

and consequently lower life-cycle cost. For the 8-12 µm range, the traditional alloy bulk

material HgCdTe (or “MCT”) is difﬁcult to fabricate with high yield, partly due to the

extreme sensitivity of the bandgap to composition (particularly the HgTe:CdTe alloy

ratio). Here T2SL materials have distinct advantages in fabrication.

Focal plane arrays comprising hundreds of thousands of T2SL detector pixels are

ﬂip-chip bonded to a CMOS read-out-circuit and then mounted on a ceramic carrier,

which in turn is glued to a cold ﬁnger in a vacuum Dewar housing, complete with an IR

window. The cold ﬁnger is cooled to detector operating temperature by a Stirling rotary

cooler. IRnovas detector-Dewar-cooler assembly for T2SL can be seen in Fig. 8.b.

IRnova has recently worked on improving the quantum efﬁciency (QE) of the de-

tection by applying anti-reﬂective coatings to the detector surface. By this method, the

EPS Lisbon January 2015 2015 - European Project Space on Intelligent Systems, Pattern Recognition and Biomedical Systems

QE was increased from approximately 50% to 80% in the wavelength region of interest.

This improves the signal-to-noise ratio and allows a reduced integration time for each

image frame.

Apart from MINERVA applications, IRnova plans to use T2SL technology for gas

detection of key greenhouse gases with absorption lines in the atmospheric transmission

bands, such as methane and perhaps also sulphur hexaﬂuoride (SF6).

More information about the superlatice IR detectors can be found in [25]

7 Infrared Megapixel Camera Development at Xenics and IRnova

The sensing unit for MINERVA is being developed in a joint effort between Xenics and

IRnova. From the start of the project Xenics has been working on the design of a Read-

Out IC (ROIC), to be integrated through ﬂip-chip technology with the T2SL (Type-2

Super Lattice) photodiode material, which is being developed by IRnova.

Fig. 9. (a) Global and zoomed view of the designed ROIC, currently in manufacturing. (b) Mi-

croscopic view of manufactured array of T2SL photo diodes.

To provide sufﬁcient resolution for reliable data analysis, a 1280×1024 pixel array

was chosen, on an aggressive pitch of 12 µm. Project requirements including frame

rate, sensitivity and noise level were taken into account in the design process. After

extensive simulations and test sample manufacturing, the ﬁnal design was taped out to

a manufacturing foundry (Fig. 9.a). The use of advanced 0.18 µm CMOS technology

is required to allow all necessary functionality within the available space of 144 µm

per pixel. The ﬁrst wafers are currently available for post-processing to be followed by

wafer-level veriﬁcation of the electrical functionality.

In parallel, IRnova has been working on the optimisation of the design and pro-

cessing of the T2SL material, towards cut-off wavelength matching and dark current

minimisation (Fig. 9.b). Once the ROIC chips become available later this year, they

will be hybridized to the sensor chip, and IRnova will integrate the resulting hybrid in

a Sterling-cooled Dewar. The so-called engine will in its turn be integrated into a full

camera by Xenics, to be supplied to the other MINERVA project partners to be used for

capturing spectroscopic images of prepared tissue samples and live cell phantoms.

MINERVA Project, mid- To near Infrared Spectroscopy for Improved Medical Diagnostics

8 Pattern Recognition and Data Analysis at GHNT

The ﬁrst task of GHNT was to provide supporting evidence for the MINERVA instru-

ment speciﬁcations. This was achieved by analysing an existing dataset and applying

pattern recognition techniques to discriminate between benign and cancerous samples

from human colon tissue biopsies (Fig. 10). Sensitivity and speciﬁcity of up 86-99%

can be achieved with the existing dataset. Using this study as a baseline GHNT was

able to assess the impact of various factors that will affect the quality and speed of the

MINERVA instrument.

Fig. 10. Partial Least Squares (PLS) scores plot showing the separation between benign and can-

cer samples in baseline study.

Reducing the number of data points per spectrum is one way to potentially speed

up the system; measuring fewer wavenumbers means a faster total acquisition time.

Multivariate pattern recognition algorithms were used to identify potential wavenum-

ber targets for the MINERVA instrument. Figure 11 shows the wavenumber regions

identiﬁed as ‘important’ for the baseline study.

Fig. 11. VIP identiﬁed wavenumber targets for the MINERVA system (red). Reference spectrum

(blue).

EPS Lisbon January 2015 2015 - European Project Space on Intelligent Systems, Pattern Recognition and Biomedical Systems

A minimum acquisition time per spectrum means that the MINERVA instrument

will be able to rapidly assess samples in a clinical timeframe. However, reducing acqui-

sition time also increases the amount of noise. To determine what level of noise can be

tolerated by the pattern recognition algorithms GHNT simulated the addition of noise to

the baseline study until it was no longer able to discriminate between pathology groups.

This allowed a minimum target signal-to-noise ratio (SNR) to be determined for the

MINERVA instrument whilst maintaining an acceptable ability to discriminate between

pathology types.

9 High Resolution mid-IR Imaging at University of Exeter

One of the main objectives of Exeters group within the MINERVA project is large scale

pathology screening using mid-infrared (mid-IR) spectroscopy. Currently FTIR spectral

histopathology, which has the potential to develop as a cancer diagnostic tool, is carried

out using a heated silicon carbide rod (“Globar”) as the mid-IR light source and focal

plane array (FPA) based detectors. This technology is limited by the low ﬂux of the

light source and the limited tissue area that can be measured in a given amount of time.

The novel technologies being developed in the MINERVA project; consisting of

mid-IR super-continuum light source (instead of a “Globar”) and new generation mega-

pixel (FPA) detectors (instead of a 128×128 pixel FPA), will be tested on pathological

samples at the University of Exeter.

Currently the base instrument, a commercially available Agilent FTIR imaging sys-

tem, in addition to the conventional Globar source coupled to FPA based imaging, has

been retro-ﬁtted with a new high-resolution imaging capability. The FTIR images ac-

quired using this set-up provided a ﬁve-fold improvement in image resolution from 5.5

µm

of the current technology to 1.1 µm

using the high magniﬁcation optics (Fig. 12).

Fig. 12. FTIR based K-means cluster images obtained using conventional and high-resolution

imaging compared with the histological image. Histological features based on the bio-molecular

composition are partitioned in the cluster images. In the high-resolution imaging, tissue and cel-

lular features are more apparent.

Future work in MINERVA will combine these novel technologies for large scale

pathology screening, and also high-resolution imaging in tissue regions of interest, with

MINERVA Project, mid- To near Infrared Spectroscopy for Improved Medical Diagnostics

the aim to develop faster and accurate cancer diagnostic tools. Initially this will integrate

with a 4.5 µm NKT MINERVA source, and later in the project it will be extended to

very long mid-IR wavelengths: possibly out beyond 10 µm.

10 Development of Standardised Samples for mid-IR

Spectrometer Instruments Testing at WWU

A key task of WWU is to transfer the MINERVA technologies to skin diagnostics and to

use mid-IR spectroscopy for the fast screening of human body surfaces and identiﬁca-

tion of patho-physiologically altered cells and tissue lesions. This requires standardised

cell and tissue sample standards with marker spectra for technology performance test-

ing of the novel optical components and systems and for training of novel approaches

for advanced data analysis.

The work of WWU in the ﬁrst MINERVA project period thus focused on the estab-

lishment of standard samples with representative spectral information of human skin

and skin cancer cells. WWU has established cell culture models which represent major

cellular skin constituents and skin cancer cell types. Furthermore, sample preparation

procedures on mid-IR compatible substrates have been developed that allow long-term

storage of cell lines without signiﬁcant losses in the quality of the spectral properties.

In order to identify suitable marker spectra of human skin, sample sets with different

preparations and cell types were analysed with mid-IR spectroscopy in collaborative

work with GHNT to retrieve reference data for technology performance testing and

for the evaluation novel algorithms for sample analysis and classiﬁcation developed

by UPV. The principle component analysis (PCA) of mid-IR spectra from different

cell types shows an excellent distinct grouping of skin components as ﬁbroblasts and

keratinocytes and cancer cells.

10 9 8 7 6 5 4 3

wavelength (µm)

1000 1500 2000 2500 3000 3500 4000

0.1

0.2

0.3

0.4

0.5

0.6

wavenumber (cm

−1

)

absorbance

fibroblasts (NIH-3T3)

keratinocytes (HaCaT)

melanoma cells (A-375)

melanoma cells (SK-MEL-28)

mid-IR spectroscopy

numerical analysis

Fig. 13. Principle component analysis (PCA) of mid-IR spectral data from ﬁbroblasts (NIH-3T3),

keratinocytes (HaCaT) and skin cancer cells (A-375, SK-MEL-28) illustrates the differentiation

between different cell types.

EPS Lisbon January 2015 2015 - European Project Space on Intelligent Systems, Pattern Recognition and Biomedical Systems

Figure 13 illustrates the analysis and differentiation of different cell types (cancer/

non-cancer) that have been prepared at WWU for the example of results PCA of mid-IR

spectral data from ﬁbroblasts (NIH-3T3), keratinocytes (HaCaT), and skin cancer cells

(A-375, SK-MEL-28). Based on these results, current and future activities at WWU in

MINERVA focus on the development of novel mid-IR standards models for skin cancer

detection that are based on 3D human skin equivalents in vitro. Further information

about the standardization of the cell samples was presented in [26].

11 First Steps with MINERVA Image Processing at UPV

The ﬁrst objectives of the image and signal processing group at UPV are focused on

segmentation and registration of different kinds of images (Fig. 14): infra-red spectral

images (IR), white light (WL), and those most used by clinicians at present, the haema-

toxylin and eosin (H&E) stained images. The latter is the current “gold standard” used

to distinguish between a healthy or pathological patient sample.

Fig. 14. (a) Infra-red (IR) image. (b) White light (WL) image. (c) Haematoxylin and eosin (H&E)

image.

The objective within MINERVA is to automatically segment regions of interest

(healthy and pathological) in the H&E images and look for their features in the in-

frared spectrum. To achieve this goal the H&E image must be registered with the WL

image (which is already registered with the infrared volume). So, the work is focused

on two interactive steps: registration and segmentation.

Registration allows the matching of elements that clinicians considered important

in the H&E images with the spectral images. A successful registration task would allow

users to learn, and later identify, the areas from which diseased and healthy cells and

patients can be distinguished (Fig. 15.a).

Segmentation concerns the accurate extraction of the cell contours (Fig. 15.b). This

would reduce the huge amount of data to be analysed looking for subtle biochemical

changes (“cancer markers”). Once the contours have been identiﬁed, the regions must be

classiﬁed as healthy or cancerous depending on subtle features including shape, texture

and clustering. This is an extremely difﬁcult task, but the use of the spectral information

in the mid-IR should eventually aid clinicians to improve on the current gold standard.

More details about the work on image processing in MINERVA project was pre-

sented in [27, 28].

MINERVA Project, mid- To near Infrared Spectroscopy for Improved Medical Diagnostics

Fig. 15. (a) Projective registration test. (b) Segmentation sample test.

12 MINERVA Lite

A prototype MINERVA system that operates in the 2-4.5 µm wavelength band has been

assembled so that the individual parts being developed by partners in the project can

be evaluated together. The ﬁnal MINERVA system will operate at even longer wave-

lengths. The key components in the integrated system are: NKT supercontinuum source,

G&H acousto-optic tunable ﬁlter, Xenics IR camera, commercial microscope and IR

optics and control electronics.

Fig. 16. (a) Photo of part of the “MINERVA Lite” laboratory set-up. (b) 0.3 Mpixel images at 49

wavelengths which can be used to form an (x, y, λ) image cube in 0.6 s.

The breadboard system is shown on Fig. 16.a. This system can take 0.3 megapixel

images with 20 µm spatial resolution at a rate of 85 frames per second. Each image is

taken at a different wavelength so that a set of 49 spectral images can be built up in

0.6 s (Fig. 16.b). These images can form an (x, y, λ) “image cube”. Each pixel records

a spectrum and this has enabled MINERVA researchers to identify a polymer ﬁlm in

EPS Lisbon January 2015 2015 - European Project Space on Intelligent Systems, Pattern Recognition and Biomedical Systems

the sample image. This important preliminary result will be extended in MINERVA to

identify spectra from cancerous cells in tissue samples and in real time on live patients.

More related work in MINERVA project has been published in [29–31]. In addition,

the research done in MINERVA has been mention and reviewed in [32, 33].

References

1. Minerva | MId- to NEaR infrared spectroscopy for improVed medical diAgnostics.

(http://minerva-project.eu/) Accessed: 2015-05-22.

2. Seddon, A.B.: Mid-infrared (IR)–A hot topic: The potential for using mid-IR light for non-

invasive early detection of skin cancer in vivo. Physica Status Solidi (B) 250 (2013) 1020–

1027

3. Oladeji, A., Sojka, L., Tang, Z., Furniss, D., Phillips, A., Seddon, A., Benson, T., Sujecki, S.:

Numerical investigation of mid-infrared emission from Pr

doped GeAsGaSe ﬁbre. Optical

and Quantum Electronics 46 (2014) 593–602

4. Sakr, H., Tang, Z., Furniss, D., Sojka, L., Moneim, N., Barney, E., Sujecki, S., Benson, T.,

Seddon, A.: Towards mid-infrared ﬁber-lasers: rare earth ion doped, indium-containing, se-

lenide bulk glasses and ﬁber. In: SPIE BiOS, International Society for Optics and Photonics

(2014) 89380V

5. Sojka, L., Tang, Z., Furniss, D., Sakr, H., Oladeji, A., Bere

s-Pawlik, E., Dantanarayana, H.,

Faber, E., Seddon, A., Benson, T., et al.: Broadband, mid-infrared emission from Pr

doped

GeAsGaSe chalcogenide ﬁber, optically clad. Optical Materials 36 (2014) 1076–1082

6. Sujecki, S.: An efﬁcient algorithm for steady state analysis of ﬁbre lasers operating under

cascade pumping scheme. International Journal of Electronics and Telecommunications 60

(2014) 143–149

7. Dantanarayana, H.G., Abdel-Moneim, N., Tang, Z., Sojka, L., Sujecki, S., Furniss, D., Sed-

don, A.B., Kubat, I., Bang, O., Benson, T.M.: Refractive index dispersion of chalcogenide

glasses for ultra-high numerical-aperture ﬁber for mid-infrared supercontinuum generation.

Optical Materials Express 4 (2014) 1444–1455

8. Seddon, A.: Mid-infrared photonics for early cancer diagnosis. In: Transparent Optical

Networks (ICTON), 2014 16th International Conference on, IEEE (2014) 1–4

9. Sakr, H., Furniss, D., Tang, Z., Sojka, L., Moneim, N., Barney, E., Sujecki, S., Benson, T.,

Seddon, A.: Superior photoluminescence (PL) of Pr

-In, compared to Pr

-Ga, selenide-

chalcogenide bulk glasses and PL of optically-clad ﬁber. Optics express 22 (2014) 21236–

21252

10. Sujecki, S., Oladeji, A., Sojka, L., Phillips, A., Seddon, A., Benson, T., Sakr, H., Tang,

Z., Furniss, D., Scholle, K., et al.: Modelling and design of mir chalcogenide glass ﬁbre

lasers. In: Numerical Simulation of Optoelectronic Devices (NUSOD), 2014 14th Interna-

tional Conference on, IEEE (2014) 111–112

11. Agger, C., Kubat, I., Møller, U., Moselund, P.M., Petersen, C., Napier, B., Seddon, A., Su-

jecki, S., Benson, T., Farries, M., et al.: Numerical demonstration of 3-12µm supercontinuum

generation in large-core step-index chalcogenide ﬁbers pumped at 4.5 µm. In: Nonlinear Op-

tics, Optical Society of America (2013) NW4A–09

12. Møller, U., Yu, Y., Petersen, C.R., Kubat, I., Mechin, D., Brilland, L., Troles, J., Luther-

Davies, B., Bang, O.: High Average Power Mid-infrared Supercontinuum Generation in a

Suspended Core Chalcogenide Fiber. In: Nonlinear Photonics, Optical Society of America

(2014) JM5A–54

13. Møller, U., Yu, Y., Kubat, I., Petersen, C.R., Gai, X., Brilland, L., M

echin, D., Caillaud, C.,

Troles, J., Luther-Davies, B., et al.: Multi-milliwatt mid-infrared supercontinuum generation

in a suspended core chalcogenide ﬁber. Optics express 23 (2015) 3282–3291

MINERVA Project, mid- To near Infrared Spectroscopy for Improved Medical Diagnostics

14. Thomsen, C.L., Nielsen, F.D., Johansen, J., Pedersen, C., Moselund, P.M., Møller, U.,

Sørensen, S.T., Larsen, C., Bang, O.: New horizons for supercontinuum light sources:

from UV to mid-IR. In: SPIE OPTO, International Society for Optics and Photonics (2013)

86370T

15. Moller, U., Bang, O.: Intensity noise of normal-pumped picosecond supercontinuum gener-

ation. In: Lasers and Electro-Optics Europe (CLEO EUROPE/IQEC), 2013 Conference on

and International Quantum Electronics Conference, IEEE (2013) 1

16. Kubat, I., Agger, C., Moselund, P., Bang, O.: Mid-infrared supercontinuum generation in ta-

pered ZBLAN ﬁber with a standard Erbium mode-locked ﬁber laser. In: Lasers and Electro-

Optics Europe (CLEO EUROPE/IQEC), 2013 Conference on and International Quantum

Electronics Conference, IEEE (2013) 1

17. Kubat, I., Agger, C., Moselund, P.M., Bang, O.: Optimized ZBLAN ﬁber for efﬁcient and

broadband mid-infrared supercontinuum generation through direct pumping at 1550nm. In:

1st International Workshop on Spatio-Temporal Complexity in Optical Fibers. (2013)

18. Kubat, I., Agger, C.S., Moselund, P.M., Bang, O.: Mid-infrared supercontinuum generation

to 4.5 µm in uniform and tapered ZBLAN step-index ﬁbers by direct pumping at 1064 or

1550 nm. JOSA B 30 (2013) 2743–2757

19. Moselund, P., Petersen, C., Leick, L., Seidelin Dam, J., Tidemand-Lichtenberg, P., Pedersen,

C.: Highly Stable, All-ﬁber, High Power ZBLAN Supercontinuum Source Reaching 4.75

µm used for Nanosecond mid-IR Spectroscopy. In: Advanced Solid State Lasers, Optical

Society of America (2013) JTh5A–9

20. Møller, U.V., Sørensen, S.T., Petersen, C.R., Kubat, I., Moselund, P.M., Bang, O.: Supercon-

tinuum generation from ultraviolet to mid-infrared. (In: 15th Conference on Optical Fibers

and Their Applications (OFTA 2014))

21. Kubat, I., Rosenberg Petersen, C., Møller, U.V., Seddon, A., Benson, T., Brilland, L., M

echin,

D., Moselund, P.M., Bang, O.: Thulium pumped mid-infrared 0.9–9µm supercontinuum

generation in concatenated ﬂuoride and chalcogenide glass ﬁbers. Optics express 22 (2014)

3959–3967

22. Kubat, I., Petersen, C.R., Møller, U., Seddon, A., Benson, T., Brilland, L., M

echin, D.,

Moselund, P., Bang, O.: Mid-infrared supercontinuum generation in concatenated ﬂuoride

and chalcogenide glass ﬁbers covering more than three octaves. In: CLEO: Science and

Innovations, Optical Society of America (2014) STh3N–1

23. Kubat, I., Agger, C.S., Møller, U., Seddon, A.B., Tang, Z., Sujecki, S., Benson, T.M., Furniss,

D., Lamrini, S., Scholle, K., et al.: Mid-infrared supercontinuum generation to 12.5 µm in

large na chalcogenide step-index ﬁbres pumped at 4.5 µm. Optics express 22 (2014) 19169–

19182

24. Petersen, C.R., Møller, U., Kubat, I., Zhou, B., Dupont, S., Ramsay, J., Benson, T., Sujecki,

S., Abdel-Moneim, N., Tang, Z., et al.: Mid-infrared supercontinuum covering the 1.4–13.3

µm molecular ﬁngerprint region using ultra-high NA chalcogenide step-index ﬁbre. Nature

Photonics 8 (2014) 830–834

25. Martijn, H., Asplund, C., von W

urtemberg, R.M., Malm, H.: High performance MWIR

type-II superlattice detectors. In: Proc. of SPIE Vol. Volume 8704. (2013) 87040Z–1

26. Kastl, L., Rommel, C.E., Kemper, B., Schnekenburger, J.: Standardized cell samples for

midir technology development. In: SPIE BiOS, International Society for Optics and Photon-

ics (2015) 931507

27. Naranjo, V., Villanueva, E., Lloyd, G.R., Stone, N., Lopez-Mir, F., Alcaniz, M.: Stained

and infrared image registration as ﬁrst step for cancer detection. In: Biomedical and Health

Informatics (BHI), 2014 IEEE-EMBS International Conference on, IEEE (2014) 420–423

28. L

opez-Mir, F., Naranjo, V., Morales, S., Angulo, J.: Probability density function of object

contours using regional regularized stochastic watershed. In: Image Processing (ICIP), 2014

IEEE International Conference on, IEEE (2014) 4762–4766

EPS Lisbon January 2015 2015 - European Project Space on Intelligent Systems, Pattern Recognition and Biomedical Systems

29. Stevens, G., Woodbridge, T.: Development of low loss robust soft-glass ﬁber splices. In:

Workshop on Specialty Optical Fibers and their Applications, Optical Society of America

(2013) W3–21

30. Markos, C., Kubat, I., Bang, O.: Hybrid polymer photonic crystal ﬁber with integrated

chalcogenide glass nanoﬁlms. Scientiﬁc reports 4 (2014)

31. Valle, S., Ward, J., Pannell, C., Johnson, N.: Acousto-optic tunable ﬁlter for imaging ap-

plication with high performance in the ir region. In: SPIE OPTO, International Society for

Optics and Photonics (2015) 93590E–93590E

32. Maragkou, M.: Supercontinuum: Reaching the mid-infrared. Nature Photonics 8 (2014)

746–746

33. Steinmeyer, G., Skibina, J.S.: Supercontinuum: Entering the mid-infrared. Nature Photonics

8 (2014) 814–815

MINERVA Project, mid- To near Infrared Spectroscopy for Improved Medical Diagnostics