WO2022208089A1 - Détermination de filtres spectraux pour imagerie spectrale - Google Patents

Détermination de filtres spectraux pour imagerie spectrale Download PDF

Info

Publication number
WO2022208089A1
WO2022208089A1 PCT/GB2022/050805 GB2022050805W WO2022208089A1 WO 2022208089 A1 WO2022208089 A1 WO 2022208089A1 GB 2022050805 W GB2022050805 W GB 2022050805W WO 2022208089 A1 WO2022208089 A1 WO 2022208089A1
Authority
WO
WIPO (PCT)
Prior art keywords
spectral
material type
spectra
filters
subset
Prior art date
Application number
PCT/GB2022/050805
Other languages
English (en)
Inventor
Sarah Elizabeth BOHNDIEK
Dale Jonathan WATERHOUSE
Travis William SAWYER
Original Assignee
Cancer Research Technology Limited
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Cancer Research Technology Limited filed Critical Cancer Research Technology Limited
Publication of WO2022208089A1 publication Critical patent/WO2022208089A1/fr

Links

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/0059—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence
    • A61B5/0075—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence by spectroscopy, i.e. measuring spectra, e.g. Raman spectroscopy, infrared absorption spectroscopy
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B1/00—Instruments for performing medical examinations of the interior of cavities or tubes of the body by visual or photographical inspection, e.g. endoscopes; Illuminating arrangements therefor
    • A61B1/00002—Operational features of endoscopes
    • A61B1/00004—Operational features of endoscopes characterised by electronic signal processing
    • A61B1/00009—Operational features of endoscopes characterised by electronic signal processing of image signals during a use of endoscope
    • A61B1/000094—Operational features of endoscopes characterised by electronic signal processing of image signals during a use of endoscope extracting biological structures
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B1/00—Instruments for performing medical examinations of the interior of cavities or tubes of the body by visual or photographical inspection, e.g. endoscopes; Illuminating arrangements therefor
    • A61B1/273—Instruments for performing medical examinations of the interior of cavities or tubes of the body by visual or photographical inspection, e.g. endoscopes; Illuminating arrangements therefor for the upper alimentary canal, e.g. oesophagoscopes, gastroscopes
    • A61B1/2733—Oesophagoscopes
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/0033—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room
    • A61B5/004—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room adapted for image acquisition of a particular organ or body part
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/0059—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence
    • A61B5/0082—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence adapted for particular medical purposes
    • A61B5/0084—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence adapted for particular medical purposes for introduction into the body, e.g. by catheters
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/42—Detecting, measuring or recording for evaluating the gastrointestinal, the endocrine or the exocrine systems
    • A61B5/4222—Evaluating particular parts, e.g. particular organs
    • A61B5/4233—Evaluating particular parts, e.g. particular organs oesophagus
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H30/00—ICT specially adapted for the handling or processing of medical images
    • G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/50—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B2576/00—Medical imaging apparatus involving image processing or analysis
    • A61B2576/02—Medical imaging apparatus involving image processing or analysis specially adapted for a particular organ or body part
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/0059—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence
    • A61B5/0071—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence by measuring fluorescence emission
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients

Definitions

  • This invention relates to a method for determining spectral filters, a spectral imaging device comprising the spectral filters, a computing device for determining the spectral filters and computer software.
  • BE Barrett’s Esophagus
  • HR-WLE high-resolution white light endoscopy
  • HR-WLE is a conventional type of spectral imaging where a colour camera acquires a measurement of three broad colour bands to replicate the spectral sensitivity of human vision, that is a red band (620 ⁇ 40 nm), a green band (540 ⁇ 40 nm) and a blue band (470 ⁇ 40 nm).
  • a red band 620 ⁇ 40 nm
  • a green band 540 ⁇ 40 nm
  • a blue band 470 ⁇ 40 nm.
  • dysplasia is heterogeneous in shape and size, patchy in distribution and difficult to identify. Therefore, dysplasia can be inconspicuous on HR-WLE due to low visible contrast between dysplasia and non-dysplastic tissue.
  • NBI narrow band imaging
  • a computer-implemented method for determining spectral filters for spectral imaging comprising: obtaining a plurality of spectra including at least a first spectrum associated with a first material type and a second spectrum associated with a second material type; obtaining spectral filter data indicative of a set of spectral filters, each spectral filter defining a spectral response; permuting through a plurality of subsets of the set of spectral filters, and for each subset: calculating image data for each of the first and second materials in dependence on a propagation of the respective spectrum through each of the spectral filters of the subset; and determining a contrast metric indicative of a contrast between the first material and the second material in dependence on the calculated image data; and selecting one of the plurality of subsets for spectral imaging in dependence on the determined contrast.
  • the imaging may be optical imaging, such as spectral endoscopy.
  • the first material may be a first tissue type.
  • the second material may be a second tissue type.
  • the first and second tissue types may be dysplastic and non-dysplastic tissue in Barrett’s Esophagus (BE).
  • tailored spectral filters may be generated to improve contrast between the different materials in spectral imaging, such as to optimise contrast for dysplasia.
  • the spectral filters may be colour filters if imaging in the visible spectrum.
  • the spectral filters may be narrow band spectral filters to accurately target contrasting spectral regions.
  • the first and second spectra may be indicative of a measured reflectance of the material and the second material respectively.
  • the first spectrum may be a reflectance spectrum associated with the first material and the second spectrum may be a reflectance spectrum associated with the second material.
  • the first and second spectra could alternatively be fluorescence, Raman or transmission spectra associated with each material.
  • the first and second spectra may be generated by any optical interaction with each material.
  • obtaining a plurality of spectra comprises: receiving a plurality of measured spectra associated with each material, and augmenting the plurality of measured spectra with augmented spectra.
  • augmenting the measured spectra facilitates the generation of an arbitrarily large data set.
  • the augmented spectra may be determined by: determining a mean spectra indicative of each material; and performing principal component analysis (PCA)-based noising of each mean spectra to obtain a plurality of augmented spectra associated with each material.
  • PCA principal component analysis
  • obtaining the spectral filter data comprises calculating a set of spectral filters.
  • the set of spectral filters may be calculated by simulation or by receiving spectral filter responses of pre-existing spectral filters.
  • the set of spectral filters may be calculated to have a range of spectral responses.
  • the range of spectral responses may result from varying one or more of: a center wavelength of the spectral filters; a full width half maximum (FWHM) of the spectral filters; the overall shape of the spectral filter response; and a relative area under curve of the spectral filters.
  • Each spectral filter may define a Gaussian spectral response band.
  • Calculating the first image data for the first material may comprise: for each of the spectral filters of the subset, propagating the first spectrum through the spectral filter to determine a channel intensity signal.
  • calculating the first image data for the first material may comprise constructing a colour for the first image data in dependence on a combination of the channel intensity signals.
  • Calculating the second image data for the second material may comprise: for each of the spectral filters of the subset, propagating the second spectrum through the spectral filter to determine a channel intensity signal.
  • calculating the second image data may comprise constructing a colour for the second image data in dependence on a combination of the channel intensity signals.
  • the contrast metric may be indicative of a contrast between the channel intensity signals for the first tissue type and second tissue type.
  • the contrast metric may be a sum of square differences or a classification accuracy of an algorithm for classifying the first material type and second material type in dependence on the channel intensity signals, such as for example a kNN classifier or the like.
  • the subset comprises at least three spectral filters.
  • Calculating the image data for each material may comprise: propagating the spectrum associated with the material through each spectral filter to determine at least three channel intensity signals , I2, ; and constructing an rgb colour for the image data in dependence on the at least three channel intensity signals.
  • the rgb colour may be constructed for each material as:
  • the rgb colour is determined to be normalized to a peak value.
  • determining the contrast between the first material and the second material may comprise determining a colour difference between the constructed colour for the first material and second material.
  • the colour difference may be determined as an International Commission on Illumination Delta E 2000 (CIEDE2000) color difference.
  • the method further comprises obtaining at least one mixed spectrum indicative of a combination of at least the first material type and second material type and obtaining ground-truth abundance data indicative of an abundance of each of the first material type and the second material type in the mixed spectrum.
  • the method may comprise determining mixed image data in dependence on a propagation of the mixed spectrum through each of the spectral filters of the subset; determining estimated abundance data in dependence on applying a linear spectral unmixing algorithm to the mixed image data using the first image data and second image data; and determining the contrast metric as an accuracy of the estimated abundance data compared to the ground-truth abundance data.
  • one of the plurality of subsets for imaging is selected in dependence on maximising the determined contrast metric.
  • the method may comprise outputting the image data to a display; receiving a user selection from an input device; and selecting one of the plurality of subsets in dependence on the user selection. For example, image data from several candidate subsets may be displayed, and the user may select that which produces the best perceived contrast.
  • a spectral imaging device comprising the subset of spectral filters determined according to the method above.
  • the spectral imaging device may comprise an illumination source and an imaging device arranged to measure data according to the determined subset of spectral filters.
  • the spectral imaging device may be a multispectral endoscope.
  • a computing device for determining spectral filters for imaging, comprising: an input configured to receive a plurality of spectra including at least a first spectrum associated with a first material type and a second spectrum associated with a second material type; one or more processors; and a memory storing computer executable instructions therein which, when executed by the one or more processors, cause the one or more processors to: obtain spectral filter data indicative of a set of spectral filters, each spectral filter defining a spectral response band; permute through a plurality of subsets of the set of spectral filters, and for each subset: calculate image data for each of the first and second materials in dependence on a propagation of the respective spectrum through each of the spectral filters of the subset; and determine a contrast metric indicative of a contrast between the first material and the second material in dependence on the calculated image data; and select one of the plurality of subsets for imaging in dependence on the determined contrast metric.
  • Figure 1 is a schematic illustration of a spectral imaging system 100
  • Figure 2 is a flowchart illustrating a method 200 according to an embodiment of the invention.
  • Figure 3 is a schematic illustration of a computing device 300 according to an embodiment of the invention.
  • Figure 4A illustrates an example spectrum 410 and an example spectral filter 420
  • Figure 4B illustrates the propagation of the spectrum 410 through each spectral filter of a subset
  • Figure 5 illustrates a detailed view of a step 230 of the method 200 according to an embodiment
  • Figure 6 illustrates simulated image data comparing imaging using filters according to the present invention to prior art imaging techniques for endoscopy
  • Figure 7 illustrates example data generated according to an embodiment of the invention for food inspection
  • Figure 8 illustrates example data generated according to an embodiment of the invention for endoscopy
  • Figure 9 illustrates example data generated according to an embodiment of the invention for remote sensing.
  • Figure 10 illustrates example data generated according to an embodiment of the invention for nailfold capillaroscopy.
  • a conventional spectral imaging system 100 is illustrated in Figure 1, in the form of a spectral endoscope 100.
  • the spectral endoscope may be used to image internal body tissue, for example in surveillance of conditions such as Barrett’s Esophagus (BE).
  • BE Barrett’s Esophagus
  • alternative spectral imaging systems to which the invention may be applied include any spectral camera system, such as for food or drug monitoring, wide-field land surveillance, or surgical field monitoring.
  • an illumination source 110 of broadband illumination 115 is coupled to an illumination fibre 120.
  • the illumination 115 is carried along the illumination fibre 120 to a tip 130 of the endoscope.
  • the tip 130 of the endoscope may be navigable within a patient in order to deliver the illumination 115 to a tissue region 190.
  • the illumination provided by the illumination fibre 120 is reflected by the tissue region 190, and reflected light is relayed back along an imaging fibre 140 to an imaging device 160, for example a standard colour camera, to record image data of the tissue region 190.
  • the resultant image data may then be transmitted to a display device 170 for display to a user of the endoscope 100.
  • the reflected light may be collected by an objective lens 145 and split.
  • a first branch 141 of the imaging fibre 140 may transmit the reflected light to a spectrometer 150, to measure a diffuse tissue reflectance spectrum.
  • a second branch 142 of the imaging fibre 140 may transmit the reflected light to the imaging device 160.
  • the imaging device 160 is a colour imaging system which acquires a measurement of three broad colour bands to replicate the spectral sensitivity of human vision, that is a red band (620 ⁇ 40 nm), a green band (540 ⁇ 40 nm) and a blue band (470 ⁇ 40 nm).
  • a red band 620 ⁇ 40 nm
  • a green band 540 ⁇ 40 nm
  • a blue band 470 ⁇ 40 nm.
  • some materials may not be distinguishable when imaged using these colour bands.
  • the imaging is used for the purpose of identifying a particular material in the image data on the display, improving the contrast in the image data is important.
  • dysplasia may often be inconspicuous in the image data displayed to the user in HR-WLE due to low visible contrast between dysplasia and non-dysplastic tissue.
  • the present invention could be applied to large scale imaging such as land surveillance to distinguish land use (e.g. urban and non-urban) and crop surveillance, such as to distinguish between crop chlorophyll levels.
  • large scale imaging such as land surveillance to distinguish land use (e.g. urban and non-urban) and crop surveillance, such as to distinguish between crop chlorophyll levels.
  • reference throughout some portions of the application will be made to contrasting a first tissue type to a second tissue type, in the context of identifying dysplasia in BE.
  • the first tissue type and second tissue type could equally be replaced with any first material type and second material type in the analogous contexts.
  • Embodiments of the present invention facilitate an improvement in image contrast between materials in imaging systems such as the spectral endoscope 100.
  • a method 200 for determining spectral filters for spectral imaging is illustrated in Figure 2.
  • the spectral filters are determined to provide improved contrast between a first material type and a second material type in image data.
  • the first material type is a first tissue type such as dysplasia
  • the second material type is a second tissue type such as non- dysplastic BE
  • the spectral imaging is spectral endoscopy.
  • this is merely an illustrative embodiment, and the method 200 is equally applicable to determining spectral filters for provide improved contrast between other materials in alternative (non-endoscopic) spectral imaging systems.
  • the method 200 may be performed by a suitable computing device 300.
  • the computing device 300 may be any computer such as a server computer or personal computer, or a portable electronic device such as a laptop, tablet or mobile phone.
  • a computing device 300 according to the invention is illustrated in Figure 3.
  • the computing device 300 comprises one or more processors 310 and a memory device 320.
  • the memory device may store computer-readable instructions which, when executed, cause the one or more processors 310 to perform embodiments of the method 200.
  • the memory device may also be configured to store one or more sets of data 335, 345 for use in the method 200, as will be explained.
  • the computing device 300 may be communicably coupled to one or more external devices (not shown) and be configured to receive the one or more sets of data 335, 345 from the external devices.
  • the computing device 300 is configured to determine selected filter data 355 for use in spectral imaging.
  • the selected filter data 355 may be stored in the memory device 320 or may alternatively be transmitted to one of the external devices, such as an element of the spectral imaging system 100, for use in imaging a sample.
  • the method 200 comprises a step 210 of obtaining a plurality of spectra in the form of spectral data 335.
  • An example spectrum 410 of the plurality of spectra according to an embodiment is illustrated in Figure 4A.
  • the spectrum 410 is indicative of an intensity of radiation across a range of wavelengths.
  • the plurality of spectra comprise at least a first reflectance spectrum 212 associated with a first material, and a second reflectance spectrum 214 associated with a second material.
  • the first and second material may in some embodiments be a first and second tissue type, such as dysplasia and non- dysplastic BE. When reference is made to the first and second tissue type, it will be appreciated that in other embodiments these could also be alternative first and second materials, as discussed.
  • the spectral data 335 comprises a plurality of spectra for each of the first and second material types. That is, the spectral data 335 may comprise a first plurality of spectra associated with the first material type, and a second plurality of spectra associated with the second material type.
  • each spectrum in the first plurality of spectra may be associated with a spectral measurement taken on a respective sample of the first material type.
  • Each spectrum in the second plurality of spectra may be associated with a spectral measurement taken on a respective sample of the second material type.
  • the plurality of spectra comprise reflectance spectra indicative of a measured reflectance of the first material type and second material type.
  • other spectral measurements may be used, such as obtained fluorescence spectra or transmission spectra.
  • the type of spectra used will be dependent on the type of imaging system the spectral filters are being determined for.
  • the obtained spectra correspond to the type of spectra to be detected in the imaging system. For example, in the spectral endoscope 100, it is the reflected illumination from the tissue sample 190 which is detected by the imaging device 160.
  • the plurality of spectra obtained in step 210 should be reflectance spectra.
  • the present invention is equally applicable to a fluorescence imaging device, or a transmission imaging device, and so appropriate spectra should be obtained in step 210 to reflect the application.
  • step 210 comprises determining the spectral data 335 by augmenting the plurality of spectra. At least some of the plurality of spectra are indicative of spectral measurements taken for the tissue types, as explained.
  • the subset of the plurality of spectra indicative of spectral measurements may be denoted measured spectra.
  • the first plurality of spectra may be considered to comprise a first set of measured spectra indicative of spectral measurements taken for the first tissue type
  • the second plurality of spectra may be considered to comprise a second set of measured spectra indicative of spectral measurements taken for the second tissue type.
  • the measured spectra may be received by the computing device 300, for example from a spectrometer 150.
  • the measured spectra may alternatively be stored in the memory 320 or at another location accessible to the computing device 300.
  • Step 210 may then comprise augmenting the measured spectra with augmented spectra to obtain the spectral data 335.
  • Each of the first plurality of spectra and the second plurality of spectra may be augmented independently.
  • Any suitable method of data augmentation may be implemented to augment the first and second plurality of spectra, for example a deep learning technique, or another statistical augmentation method such as by using Gaussian white noise or principal component analysis (PCA)-based noising.
  • PCA principal component analysis
  • a mean first spectrum l) indicative of the first tissue type is determined by averaging over the set of first measured spectra R ⁇ X).
  • PCA-based noising is then performed on 1) to obtain a set of first augmented spectra.
  • Principal components of variation in R t (X) and associated eigenvalues are determined as follows:
  • N is the number of principal components of R ⁇ X). Only n-1 principal components are non-trivial, where n is the number of spectra in the set of first measured spectra R t (X).
  • a noised spectrum R j (X) may then be generated to augment the first plurality of spectra according to:
  • p is a random number drawn from a standard normal distribution
  • s 2 is a sum of variances across all wavelengths in the set of first measured spectra
  • D is a dimensionless constant representing the desired degree of noise.
  • the first plurality of spectra may then be obtained by augmenting the set of first measured spectra R t (X) with the set of first augmented spectra R j 1 (X).
  • the above augmentation may be repeated for the set of second measured spectra R 2 (2) to obtain a set of second augmented spectra R j 2 (X).
  • the second plurality of spectra may then be obtained by augmenting the set of second measured spectra R 2 (2) with the set of second augmented spectra R j 2 (X).
  • the method 200 comprises a step 220 of obtaining spectral filter data 345.
  • the spectral filter data 345 is indicative of a set of spectral filters. Each spectral filter in the set defines a respective spectral response band.
  • the spectral filter data 345 defines a candidate set of spectral filters, of which a subset will be selected for use in spectral imaging, such as to replace the R, G and B filters in the spectral endoscope 100.
  • the spectral filter data 345 may be stored in the memory device 320 or in another location communicably coupled to the computing device 300.
  • step 220 may comprise retrieving or receiving the spectral filter data 345.
  • the spectral filter data 345 may be generated by the computing device 300 during the method 200.
  • Each spectral filter 420 in the spectral filter data 345 may be a narrow band spectral filter. At least some of the spectral filters may be denoted colour filters, if the spectral imaging is in the visible spectrum and thus the spectral response band defined by the spectral filter lies in the visible spectrum.
  • each spectral filter 420 defines a Gaussian spectral response band, although it will be appreciated that this is not a requirement and a variety of spectral response shapes may be utilized.
  • the set of spectral filters in the spectral filter data 345 may be generated by varying a center wavelength of the spectral response band, a full width half maximum (FWHM) of the spectral response band, and/or a relative area under curve of the spectral response band.
  • FWHM full width half maximum
  • each spectral filter 420 may be generated as a Gaussian spectral filter according to: —4 ⁇ log(2) (1 — 2 c fc ) 2
  • the FWHM may be kept constant. For example, a constant value of 10 nm may be utilised. In other embodiments, a different FWHM value may be selected, such as 5 nm or 20 nm. In other embodiments, the FWHM may be varied to provide an additional or alternative degree of freedom for the set of spectral filters.
  • the method 200 comprises a step 230 of permuting through a plurality of subsets of the set of spectral filters obtained in step 220.
  • step 230 comprises simulating use of the subset of spectral filters in spectral imaging to calculate image data for each of the first and second material types.
  • a contrast metric indicative of a contrast between the first material type and second material type can be determined in dependence on the simulated image data.
  • the step 230 may be implemented as an iterative process 230, as illustrated in Figure 5.
  • the process 230 comprises a step 232 of selecting a subset of the spectral filters from the spectral filter data 345.
  • each subset is selected to comprise three spectral filters S 1 ,S 2 ,S 3 .
  • the three spectral filters can be mapped to R, G and B channels in standard display equipment, which is advantageous for implementation in many spectral imaging techniques.
  • fewer or more spectral filters than three may be utilised, as is the case for the embodiments used to provide the data in Figures 7 to 9.
  • the process 230 comprises a step 234 of calculating first image data for the first tissue type.
  • the first image data may comprise a colour or intensity calculated by simulating spectral imaging of the first tissue type using the subset of spectral filters.
  • Figure 4B shows a schematic illustration of the process simulated by step 234.
  • a spectrum 410 indicative of light reflected from a tissue is propagated through each of the respective spectral filters 421, 422, 423 in the subset and a resultant respective intensity h, 1 , 1 is detected at a detector 430 in a spectral imaging device such as spectral endoscope 100.
  • the process illustrated in Figure 4B may be simulated for each of the first and second tissue types as follows.
  • the process is simulated for the first tissue type.
  • a first set of spectra R t (A) was obtained in step 210.
  • the first set of spectra is indicative of the first tissue type and optionally comprises both measured and augmented spectra, as discussed.
  • Each spectrum in the first set of spectra is propagated through each spectral filter of the subset S 1 ,S 2 ,S 3 to simulate a respective measured intensity:
  • a first set of channel intensity signals comprising a respective channel intensity signal for each spectral filter of the subset S 1; S 2 ,S 3 may then be determined by averaging over the measured intensities:
  • M is the number of spectra in the first set of spectra R ⁇ A).
  • M is the number of spectra in the first set of spectra R ⁇ A).
  • three channel intensity signals are determined for the first tissue type, / 1 (S 1 ),/ 1 (S 2 ),/ 1 (S 3 ).
  • the three channel intensity signals are indicative of an intensity of radiation measured at the detector 430 when imaging the first tissue type through each respective spectral filter.
  • step 236 the process may be simulated analogously for the second tissue type.
  • the above process may be performed for the second tissue type using the second set of spectra R 2 (A ) to obtain a second set of channel intensity signals. That is, each spectrum in the first set of spectra is propagated through each spectral filter of the subset Si,S 2 ,S 3 to simulate a respective measured intensity:
  • a second set of channel intensity signals comprising a respective channel intensity signal for each spectral filter of the subset S 1 ,S 2 ,S 3 may then be determined by averaging over the measured intensities:
  • M is the number of spectra in the second set of spectra R 2 (2).
  • M is the number of spectra in the second set of spectra R 2 (2).
  • three channel intensity signals are determined for the first tissue type, / 2 (S 1 ),/ 2 (S 2 ),/ 2 (S 3 ).
  • the three channel intensity signals are indicative of an intensity of radiation measured at the detector 430 when imaging the second tissue type through each respective spectral filter.
  • Steps 234 and 236 may optionally comprise constructing a first colour Ci for the first tissue type in dependence on a combination of the first channel intensity signals / 1 (S 1 ),/ 1 (S 2 ),/ 1 (S 3 ), and a second colour C 2 for the second tissue type in dependence on a combination of the second channel intensity signals / 2 (S 1 ),/ 2 (S 2 ),/ 2 (S 3 ),.
  • the colour may be an RGB colour constructed by assigning each of the channel intensity signals to a respective colour channel:
  • Ci Ii(S 2 )
  • C 2 I 2 (S 2 )
  • simulated first and second image data indicative of the respective colour may be determined for each of the first and second tissue types.
  • the image data may be readily displayed for a user, for example as at least one pixel of the respective colour on standard display equipment.
  • Each colour may optionally be normalized to a peak value p in order to ensure consistent brightness. In this way, variation in intensity across the spectra is mitigated and variation in the constructed colours can be attributed to chromaticity.
  • a suitable value between a predetermined minimum and maximum such as 0 and 1 may be chosen as the peak value p. In an example, a peak value of 0.8 may be selected to produce a bright but not saturated image.
  • a contrast metric indicative of a contrast between the first tissue type and the second tissue type is determined in dependence on the image data calculated in steps 234 and 236.
  • determining the contrast may comprise determining a colour difference between the first colour and the second colour.
  • any parameter indicative of a colour difference may then be utilised as a contrast metric.
  • the contrast metric may be determined in dependence on an International Commission on Illumination Delta E 2000 (CIEDE2000) color difference between the first colour and the second colour.
  • CEDE2000 International Commission on Illumination Delta E 2000
  • Any alternative distance metric between the first colour and the second colour in colour space may be used, such as an sRGB Euclidean distance.
  • contrast metrics may be used in other embodiments. If more than three channel intensity signals are used (i.e. more than three spectral filters are selected in a subset), then colour difference is not applicable and an alternative contrast metric should be used. Even if colour difference is applicable, it could be envisaged that an alternative contrast metric may be calculated in step 238.
  • Such alternative contrast metrics may include for example a classification accuracy or a sum of square differences between the channel intensity signals for the first tissue type and the channel intensity signals for the second tissue type. An example using classification accuracy is illustrated in Figure 8.
  • step 210 may further comprise receiving one or more mixed spectra indicative of a combination of the first material type and second material type, as well as ground-truth abundance data indicative of the abundance of each of the first material type and the second material type in each mixed spectrum.
  • each mixed spectra may also be propagated through the subset of spectral filters.
  • a linear spectral unmixing algorithm may be applied to the propagated mixed spectra using the propagated first spectrum from step 234 and the propagated second spectrum from step 236.
  • the linear spectral unmixing algorithm is used to determine estimated abundance data indicative of an estimated abundance of each of the first material type and the second material type in each mixed spectrum.
  • the contrast metric may thus comprise an accuracy of spectral unmixing, that is an accuracy of the estimated abundance data compared to the ground-truth abundance data received in step 210. Exemplary embodiments using linear spectral unmixing accuracy as a contrast metric will be described with reference to Figures 9 and 10.
  • Step 230 may be repeated for each candidate subset of colour filters by permuting through each possible combination. In this way, a respective value of the contrast metric is determined for each candidate subset.
  • the method 200 then comprises a step 240 of selecting one of the plurality of subsets for spectral imaging in dependence on the determined contrast metric.
  • the computing device 300 is configured to generate the selected filter data 355 in dependence on the selected subset.
  • Step 240 may comprise selecting a subset having a high determined contrast. In some embodiments, the subset having the maximum determined contrast may be selected.
  • a user selection may be considered when selecting the subset.
  • Several candidate subsets may be selected in dependence on the determined contrast, e.g. the candidate subsets having the highest determined contrast.
  • the corresponding first and second image data for each candidate subset may be output to a user via a display device.
  • the user may then determine which of the candidate subsets results in the highest subjective distinction between first and second image data.
  • a user selection for one subset of the candidate subsets may be received from a user input device.
  • the method 200 thus provides a tailored subset of spectral filters which results in highly visible contrast between the first and second materials when used in spectral imaging.
  • Figure 6 illustrates simulated images constructed from example reflectance spectra indicative of a first tissue type (neoplasia) and a second tissue type (non-dysplastic BE).
  • the simulated images comprise a circular region of neoplasia inside a region of non- dysplastic BE.
  • Four images 610, 620, 630, 640 were simulated.
  • white light endoscopy (WLE) was simulated by propagating each spectrum representative of each pixel through standard R, G and B filters.
  • An RGB colour was constructed for each pixel by combining the intensity of the spectrum following propagation through each filter respectively, to mimic detection in a white light endoscope (WLE).
  • the resultant image 610 comprises a central region 611 of first tissue type and a peripheral region 612 of second tissue type. As can be seen, the contrast between the two regions is low.
  • the R, G and B filters have been replaced with narrow band filters to simulate traditional narrow band imaging (NBI).
  • the resultant image 620 again comprises a central region 621 showing the first tissue type and a peripheral region 622 showing the second tissue type. As with the first image, the contrast between the two regions is low.
  • the third image 630 and the fourth image 640 have each been generated using tailored filters determined using the method 200 to simulate the tailored multispectral endoscopy.
  • the third image 630 was generated using the subset of filters exhibiting maximum colour contrast
  • the fourth image 640 was generated using the subset of filters exhibiting the second highest colour contrast.
  • the contrast between the tissue types in the simulations of tailored multispectral endoscopy is thus visibly higher than the contrast produced in the simulations of traditional WLE or NBI, facilitating easier and more accurate classification of tissue.
  • the CIEDE colour difference between the first and second tissue type for each image is also shown in Figure 6, illustrating a significantly higher colour difference between the first and second tissue in the third and fourth images 630, 640 than for the simulations of traditional WLE and NBI 610, 620.
  • a control image indicative of a third tissue type is also provided which is simulated analogously using reflectance spectra of the third tissue type.
  • a first material type is a first food type
  • a second material type is a second food type.
  • the first food type is ripe or firm blueberries
  • the second food type is overripe or soft blueberries.
  • Figure 7(A) illustrates a reference RGB image taken of blueberries, wherein the RGB image has a first region 711 showing the first food type and a second region 712 showing the second food type.
  • the RGB image provides little to no contrast between the two food types and thus the food types cannot be readily differentiated by visual inspection using standard RGB imaging.
  • the method 200 may thus be utilized to select a subset of spectral filters to improve the visual contrast between the first food type and the second food type.
  • the image data determined in steps 234 and 236 is experimentally captured rather than simulated. That is, for each subset of spectral filters, an image is experimentally captured using the subset of spectral filters selected in step 232. However, it is appreciated that the image data could alternatively be simulated by obtaining spectra indicative of each material type and propagating each spectrum through each filter as previously described.
  • Image data is experimentally captured using each filter / to provide a plurality of channel intensity values li j (Si) for each pixel j corresponding to the first food type, and a second plurality of channel intensity values l2 j (Si) for each pixel j corresponding to the second food type.
  • the channel intensity values for each food type can be averaged over the pixels corresponding to that food type to produce an average channel intensity value (Si) for the first food type and an average channel intensity value l2(S,) for the second food type, as previously described.
  • a first colour Ci for the first food type and a second colour C2 for the second food type are constructed.
  • the colour may be an RGB colour constructed by assigning each of the channel intensity signals to a respective colour channel:
  • first and second colours may be determined for each of the first and second food types.
  • the image data may be readily displayed for a user, for example as at least one pixel of the respective colour on standard display equipment.
  • Each colour may optionally be normalized to a peak value p in order to ensure consistent brightness, as previously described.
  • a contrast metric indicative of a contrast between the first food type and the second food type is determined in dependence on the captured image data.
  • the contrast metric is a CIEDE2000 colour difference between the first colour and the second colour.
  • contrast metrics such as an sRGB Euclidean distance.
  • a set of spectral filters is selected in dependence on the determined contrast metric, for example to maximize the contrast metric.
  • Figure 7(B) illustrates image data for the first and second food type captured using the 11 filters which provide the largest value of CIEDE2000 colour difference, alongside the original RGB image.
  • the images are normalized to mean intensity and cropped to show a centre of image where illumination is even.
  • the resultant CIEDE2000 colour difference is shown above each image. It can be seen that the images captured using the selected spectral filters provide a significantly higher contrast between the first and second material types than the original image.
  • contrast metric used is an accuracy of classification of the spectra into the associated material types.
  • the first material type is neoplasia
  • the second material type is non-dysplastic BE
  • the third material type is squamous tissue
  • the spectral band optimization is applied in the context of multispectral endoscopy in the esophagus.
  • Figure 8(A) illustrates mean reflectance spectra of the three material types respectively.
  • the vertical lines and shaded regions represent the optimal center wavelengths and bandwidths of the subset of spectral filters selected in step 240. Four spectral filters are selected in the illustrated embodiment.
  • Steps 210 and 220 may be performed as described previously.
  • step 230 the spectra obtained in step 210 are shuffled and divided into subsets. One subset is reserved for testing and the remaining subsets are used for training of a k nearest neighbour (kNN) classifier.
  • kNN k nearest neighbour
  • the obtained spectra are propagated through the spectral filters to calculate channel intensity signals as previously discussed with reference to steps 234 and 236.
  • the channel intensity signals from the training subset are used to train the kNN classifier to classify the spectra into the material types.
  • the channel intensity signals from the testing subset are used to test the classification accuracy of the trained classifier. This process is repeated multiple times so that all subsets have been used for testing, and the accuracies tested by each subset are averaged.
  • the contrast metric is determined as the 5-fold cross validation classification accuracy of the kNN classifier.
  • step 240 the subset of spectral filters which provide a maximal classification accuracy are selected.
  • Figure 8(B) provides an illustration of a simulated array of tissue shaded by tissue type.
  • the array comprises a central section of neoplasia, a surrounding portion of non- dysplastic BE and an outer section of squamous tissue.
  • Each pixel of the simulated array is assigned a corresponding spectrum associated with its material type. In this embodiment, each of the spectra was derived from a measured spectrum associated with the material type.
  • Figure 8(C) illustrates a simulated RGB image of the simulated array. As can be seen, the simulated RGB image provides a low level of discrimination between the different material types.
  • Figure 8(D) illustrates the resultant classification of each pixel into a respective material type by simulating multispectral imaging using the spectral filters selected in step 240 and classifying the resultant channel intensities using the kNN classifier.
  • the kNN classifier can classify the material of each pixel with an accuracy of 78%.
  • each pixel of an image may contain spectral contributions from more than one material type.
  • the contrast metric to be optimized may be a spectral unmixing accuracy, that is an accuracy of determining an estimated abundance of each material type in each pixel following propagation through the spectral filters.
  • Figure 9 illustrates example data obtained according to an embodiment of the invention.
  • the method 200 was applied to multispectral imaging for remote sensing to distinguish three material types: soil, tree and water.
  • input data is obtained in the form of the publicly available Samson reference dataset.
  • the Samson reference dataset is spectral data associated with an image having N pixels. Each pixel can be composed of a mixture of the three material types.
  • the input data comprises a respective spectrum associated with each of the three material types, as illustrated in Figure 9(A)(iii).
  • the input data further comprises a plurality of mixed spectra in the form of a hypercube array M illustrated in Figure 9(A)(i).
  • the hypercube array M provides a respective spectrum associated with each of the N pixels.
  • the hypercube array M may thus be represented as an Nxp matrix, where N is the number of pixels in the image, and p is the number of wavelengths at which the spectrum is sampled.
  • the entry M kj in the hypercube array is the signal at pixel k, wavelength j.
  • the input data further comprises an abundance matrix A as illustrated in Figure 9(A)(ii).
  • the abundance matrix A is a 3xN matrix, wherein the entry A ik indicates the respective abundance of each material type / at pixel k.
  • the method 200 is thus used to determine a set of spectral filters to be applied to a remote sensing imaging system to accurately retrieve relative abundances of the three material types soil, water and tree in each of the N imaged pixels.
  • a 3x6 channel intensity matrix / may be constructed having elements as shown above.
  • step 230 comprises an additional step of propagating the hypercube array M through each filter of the subset, providing a propagated hypercube array M’.
  • the propagated hypercube array M’ is an Nx6 matrix, wherein element M k i is the intensity signal for pixel k through spectral filter /.
  • step 238 the relative abundance of the different material types in the propagated hypercube array M’ is estimated using linear spectral unmixing.
  • the linear spectral unmixing can be expressed as:
  • A’ denotes an estimated abundance matrix, i.e. an estimation of the abundance matrix A.
  • A’ is a 3xN matrix, wherein the entry A k i indicates the respective estimated abundance of each material type / at pixel k.
  • the above linear spectral unmixing equation can be solved as an inverse problem with non-negative constraints using the non-negative constrained least squares (NNLS) optimisation.
  • the NNLS optimisation tries to find the A’ to minimise errors,
  • A’ is the recovered abundance matrix using the NNLS optimisation.
  • the spectral filters which provide the best contrast between the material types should provide the best recovery of A’. Therefore, the contrast metric used may be the root mean squared error (RMSE):
  • step 240 the subset of spectral filters is selected to minimise the RMSE.
  • Figure 9(B)(i) illustrates the subset of six spectral filters selected in step 240 for the example dataset, wherein one spectral filter is sampled twice to minimise RMSE.
  • Figure 9(C)(i) provides an illustration of the ground truth abundance across N pixels for each of the three material types.
  • Figure 9(C)(ii) provides an illustration of the estimated abundance across the N pixels for each of the three material types, wherein the estimation is achieved by simulating imaging using the six spectral filters selected in step 240.
  • the estimated abundances provide a RMSE of 0.18 compared to the ground truth abundances.
  • FIG 10 illustrates example data obtained according to another embodiment of the invention.
  • the method 200 was applied to select spectral filters to be used in nailfold capillaroscopy, where the capillaries of a finger are spatially resolved at high magnification to assess the severity of connective tissue disorders.
  • the capillaries are illuminated, and the reflected light is collected via a lens system onto an imaging device.
  • Capillary haemoglobin oxygenation which provides additional insight into disease progression, cannot be measured using standard colour bands.
  • the present invention facilitates the determination of optimal spectral filters to be applied to the imaging system that enables a distinction between the relative concentration of oxygenated haemoglobin (HbC ) and deoxygenated haemoglobin (Hb).
  • a first material type is oxygenated haemoglobin (HbC ⁇ )
  • a second material type is deoxygenated haemoglobin (Hb)
  • a third material type is melanin
  • a fourth material type is assigned to the background scattering.
  • the overall absorbance spectrum of nailfold tissue can be modelled as:
  • f bi00d is the blood volume fraction
  • a is the oxygen saturation representing haemoglobin oxygenation
  • C ⁇ b is the total concentration of HbC>2 and Hb set to 150 g/L
  • e H ⁇ o 2 (l) is the molar extinction coefficient of HbC>2
  • £ Hb ( ) is the molar extinction coefficient of Hb
  • l dermis is the round-trip pathlength of light travelling in capillaries for reflectance imaging (set to 40 pm in this example)
  • f meian in ' s the melanin volume fraction contained in epidermis
  • m a, th b ⁇ ah ⁇ h(l ) ' s the melanin absorption coefficient
  • l evidermis is the round-trip pathlength of light travelling in epidermis for reflectance imaging (set to 200 pm example)
  • m 5 ' ( ) is the reduced tissue scattering coefficient
  • S and G are fitting parameters greater than zero.
  • each subset is selected to have four spectral filters.
  • a first absorbance spectrum is shown for HbC>2 and a second absorbance spectrum is shown for Hb.
  • the first absorbance spectrum corresponds to the first spectrum 212 in the context of the method 200
  • the second absorbance spectrum corresponds to the second spectrum 214.
  • Figure 10(a) illustrates four modelled nailfold absorbance spectra for nailfold tissue having different values of f bi00d and a.
  • a set of nailfold spectra are modelled that are representative of those that would be measured from the nailfold, containing various amounts of haemoglobin and melanin.
  • f bi00d ranges from 50% to 100% (with 20 evenly spaced steps) and a ranges from 50% to 100% (with 20 evenly spaced steps). All the combinations of the sampled f bi00d and a are tried, with each added by a randomly chosen f meian in from the range of 1-3% (for fair skin), 11-16% (for medium skin), and 18-34% (for dark skin).
  • R is the corresponding reflectance spectrum.
  • Gaussian noise is added to each reflectance spectrum to define a given signal-to-noise ratio (SNR), in this case 30dE3.
  • SNR signal-to-noise ratio
  • the noisy reflectance spectra are then converted back to nailfold absorbance spectra.
  • Each modelled nailfold absorbance spectra is associated with a ground-truth value of abundance for each of the first material type (HbC>2) and the second material type (Hb).
  • Each modelled nailfold absorbance spectra is functionally equivalent to a pixel of the hypercube in the example shown in Figure 9.
  • the first absorbance spectrum for HbC>2 is propagated through each spectral filter in step 234, and the second absorbance spectrum for Hb is propagated through each spectral filter in step 236.
  • each modelled nailfold absorbance spectra is also propagated through each spectral filter.
  • step 2308 linear spectral unmixing is performed using the propagated nailfold spectra and the propagated first and second spectra analogously to the method discussed with reference to Figure 9, to find an estimated relative abundance of each of the first material type (HbC ⁇ ) and the second material type (Hb).
  • the contrast metric to be determined is the average normalised root mean squared error (NRMSE), which is the RMSE divided by the mean of the ground-truth abundance values,
  • x denotes the ground-truth abundance value for each material type
  • x denotes the estimated abundance using the linear spectral unmixing
  • N is the total number of modelled nailfold spectra.
  • NRMSE is used to normalise the scaling of HbC>2 and Hb.
  • HbC>2 is richer in the capillary than Hb.
  • step 240 the subset of spectral filters having the lowest NRMSE is selected.
  • Figure 10(b) shows a monochrome image of a capillary.
  • a respective hypercube array may be constructed in the spatial pattern of a capillary from the modelled nailfold absorbance spectra.
  • Figures 10(c) and 10(d) show a map of the resultant abundances of the first material type (HbC>2) at each pixel of one of the hypercube arrays, with and without the addition of simulated Gaussian noise.
  • the simulated hypercube arrays are thus functionally equivalent to the hypercube array M from the embodiment shown in Figure 9.
  • Figure 10(e) illustrates the estimated abundances at each pixel using the optimal four spectral filters selected in step 240.
  • the optimal four spectral filters result in an NRMSE of 0.30 across the four hypercubes of different skin types.
  • the subset of spectral filters may be implemented in a multispectral imaging device, such as the spectral endoscope 100, in a range of different hardware configurations.
  • the spectral filters may be implemented either to tune the illumination source, at the detector, or as a separate component.
  • physical filters may be manufactured to replicate the determined subset and implemented on a filter changing apparatus such as a filter wheel in the multispectral imaging device.
  • the measured radiation such as the reflected light in the imaging fibre 140 in endoscopy, is propagated through a filter of the filter changing apparatus before detection at the detector 160.
  • a respective intensity for each filter can be detected at the detector 160.
  • spectrally resolved detector array may be implemented at the detector 160 to incorporate said filters.
  • filters may be applied to the illumination source.
  • a further spatial optimization of the filters may be performed across the array.
  • the spectral filters are arranged in a spatial pattern on the array, such as illustrated in Figure 9(E3)(ii).
  • the arrangement of the spectral filters in the spatial pattern may have an impact on the contrast metric.
  • the method 200 may optionally comprise determining a spatial arrangement of the subset of filters selected in step 240.
  • a spatial optimization is applied to a set of hypercubes, i.e. , a spatial array of input spectra.
  • Each hypercube contains spatial information in the x and y dimensions, and the spectra in the l dimension.
  • the classification of the material type at each pixel is given a suitable spatial arrangement.
  • the spatial optimization process generates all the spatial combinations of the selected filters and applies each to the set of hypercubes.
  • the average contrast metric for example the colour difference, classification accuracy or accuracy of linear spectral unmixing can be calculated across the set of hypercubes for each spatial arrangement of the filters.
  • the spatial arrangement of filters that gives the best result for the relevant contrast metric is determined to be the best spatial arrangement.
  • the present invention provides a multispectral imaging device incorporating filters which are optimized to distinguish the first and second material.
  • the filters may be both spectrally and spatially optimized as discussed. In this way, when imaging the first and second material for which the filters were tailored, improved contrast is perceived in the resultant image data displayed to a user of the multispectral imaging device, facilitating easier classification of materials during imaging.
  • embodiments of the present invention can be realised in the form of hardware, software or a combination of hardware and software. Any such software may be stored in the form of volatile or non-volatile storage such as, for example, a storage device like a ROM, whether erasable or rewritable or not, or in the form of memory such as, for example, RAM, memory chips, device or integrated circuits or on an optically or magnetically readable medium such as, for example, a CD, DVD, magnetic disk or magnetic tape. It will be appreciated that the storage devices and storage media are embodiments of machine-readable storage that are suitable for storing a program or programs that, when executed, implement embodiments of the present invention.
  • embodiments provide a program comprising code for implementing a system or method as claimed in any preceding claim and a machine readable storage storing such a program. Still further, embodiments of the present invention may be conveyed electronically via any medium such as a communication signal carried over a wired or wireless connection and embodiments suitably encompass the same.

Abstract

L'invention concerne un procédé mis en œuvre par ordinateur permettant de déterminer des filtres spectraux pour une imagerie spectrale, un dispositif d'imagerie spectrale, un dispositif informatique et un logiciel informatique. Le procédé consiste : à obtenir une pluralité de spectres comprenant au moins un premier spectre associé à un premier type de matériau et un second spectre associé à un second type de matériau ; à obtenir des données de filtres spectraux indiquant un ensemble de filtres spectraux, chaque filtre spectral définissant une réponse spectrale et à permuter entre une pluralité de sous-ensembles de l'ensemble de filtres spectraux. Pour chaque sous-ensemble, des données d'image sont calculées pour chacun des premier et second types de matériau en fonction d'une propagation du spectre respectif à travers chacun des filtres spectraux du sous-ensemble ; et une mesure de contraste est déterminée indiquant un contraste entre le premier type de matériau et le second type de matériau en fonction des données d'image calculées. Un sous-ensemble de la pluralité de sous-ensembles est sélectionné pour une imagerie spectrale en fonction de la mesure de contraste déterminée.
PCT/GB2022/050805 2021-03-31 2022-03-30 Détermination de filtres spectraux pour imagerie spectrale WO2022208089A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
GBGB2104680.0A GB202104680D0 (en) 2021-03-31 2021-03-31 Determination of spectral filters for spectral imaging
GB2104680.0 2021-03-31

Publications (1)

Publication Number Publication Date
WO2022208089A1 true WO2022208089A1 (fr) 2022-10-06

Family

ID=75783857

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/GB2022/050805 WO2022208089A1 (fr) 2021-03-31 2022-03-30 Détermination de filtres spectraux pour imagerie spectrale

Country Status (2)

Country Link
GB (1) GB202104680D0 (fr)
WO (1) WO2022208089A1 (fr)

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2005031433A1 (fr) * 2003-09-26 2005-04-07 Tidal Photonics, Inc. Appareil et procedes relatifs a des systemes endoscopes a imagerie couleur
JP2013102897A (ja) * 2011-11-11 2013-05-30 Fujifilm Corp 内視鏡診断装置
US20150071514A1 (en) * 2013-09-10 2015-03-12 Ohio State Innovation Foundation Methods and devices for optimization of magnetic resonance imaging protocols
US20180177461A1 (en) * 2016-12-22 2018-06-28 The Johns Hopkins University Machine learning approach to beamforming
US20180247153A1 (en) * 2017-02-28 2018-08-30 Verily Life Sciences Llc System and method for multiclass classification of images using a programmable light source
WO2020212470A1 (fr) * 2019-04-17 2020-10-22 Koninklijke Philips N.V. Systèmes et procédés d'imagerie médicale avec correction automatique de qualité d'image sur la base de l'analyse de journal de dispositifs médicaux
US20200394792A1 (en) * 2019-06-13 2020-12-17 Boston Scientific Scimed, Inc. Methods and apparatus to detect bleeding vessels

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2005031433A1 (fr) * 2003-09-26 2005-04-07 Tidal Photonics, Inc. Appareil et procedes relatifs a des systemes endoscopes a imagerie couleur
JP2013102897A (ja) * 2011-11-11 2013-05-30 Fujifilm Corp 内視鏡診断装置
US20150071514A1 (en) * 2013-09-10 2015-03-12 Ohio State Innovation Foundation Methods and devices for optimization of magnetic resonance imaging protocols
US20180177461A1 (en) * 2016-12-22 2018-06-28 The Johns Hopkins University Machine learning approach to beamforming
US20180247153A1 (en) * 2017-02-28 2018-08-30 Verily Life Sciences Llc System and method for multiclass classification of images using a programmable light source
WO2020212470A1 (fr) * 2019-04-17 2020-10-22 Koninklijke Philips N.V. Systèmes et procédés d'imagerie médicale avec correction automatique de qualité d'image sur la base de l'analyse de journal de dispositifs médicaux
US20200394792A1 (en) * 2019-06-13 2020-12-17 Boston Scientific Scimed, Inc. Methods and apparatus to detect bleeding vessels

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
ESBEN JANNIK BJERRUM; MADS GLAHDER; THOMAS SKOV: "Data Augmentation of Spectral Data for Convolutional Neural Network (CNN) Based Deep Chemometrics", ARXIV.ORG, 5 October 2017 (2017-10-05), pages 1 - 10, XP080826305 *
FANG XIANGLIN; ZENG QIUYAO; YAN XINLIANG; ZHAO ZUYI; CHEN NA; DENG QIANRU; ZHU MENGHAN; ZHANG YANJIAO; LI SHAOXIN: "Fast discrimination of tumor and blood cells by label-free surface-enhanced Raman scattering spectra and deep learning", JOURNAL OF APPLIED PHYSICS, vol. 129, no. 12, 23 March 2021 (2021-03-23), pages 1 - 11, XP012254990, ISSN: 0021-8979, DOI: 10.1063/5.0042662 *
GEORGOULI KONSTANTIA, OSORIO MARIA TERESA, MARTINEZ DEL RINCON JESUS, KOIDIS ANASTASIOS: "Data augmentation in food science: Synthesising spectroscopic data of vegetable oils for performance enhancement", JOURNAL OF CHEMOMETRICS, vol. 32, no. 6, 1 June 2018 (2018-06-01), GB, pages e3004, XP055854020, ISSN: 0886-9383, DOI: 10.1002/cem.3004 *

Also Published As

Publication number Publication date
GB202104680D0 (en) 2021-05-12

Similar Documents

Publication Publication Date Title
US10117582B2 (en) Medical hyperspectral imaging for evaluation of tissue and tumor
US20190133514A1 (en) System and method for optical detection of skin disease
US20200267336A1 (en) Systems and methods for hyperspectral imaging
Johansen et al. Recent advances in hyperspectral imaging for melanoma detection
US9593982B2 (en) Sensor-synchronized spectrally-structured-light imaging
US9135701B2 (en) Medical image processing
US11257213B2 (en) Tumor boundary reconstruction using hyperspectral imaging
US20090136101A1 (en) Method and System for Analyzing Skin Conditions Using Digital Images
US20190279362A1 (en) Method of non-invasive detection of tumour and/or healthy tissue and hyperspectral imaging apparatus
JP4599520B2 (ja) マルチスペクトル画像処理方法
JP5165732B2 (ja) マルチスペクトル画像処理方法、画像処理装置、及び画像処理システム
US11686618B2 (en) Hyperspectral imaging method and device
Spigulis et al. Multi-spectral skin imaging by a consumer photo-camera
EP3716136A1 (fr) Reconstruction de limites de tumeur à l'aide de l'imagerie hyperspectrale
Bochko et al. Lower extremity ulcer image segmentation of visual and near‐infrared imagery
WO2022208089A1 (fr) Détermination de filtres spectraux pour imagerie spectrale
Wirkert Multispectral Image Analysis in Laparoscopy-A Machine Learning Approach to Live Perfusion Monitoring
JP2021001777A (ja) 植物の生育状態評価方法および評価装置
Obukhova et al. Learning from multiple modalities of imaging data for cancer detection/diagnosis
Taylor-Williams et al. Targeted Multispectral Filter Array Design for Endoscopic Cancer Detection in the Gastrointestinal Tract
JP2023129277A (ja) 周波数帯域を用いたハイパースペクトルによる物体画像の検出方法
Mendes Narrow-Band Image Processing for Gastroenterological Examinations
Kelman Techniques for capture and analysis of hyperspectral data

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22715365

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 22715365

Country of ref document: EP

Kind code of ref document: A1