WO2022263515A1 - Co-design von optischem filter und fluoreszenz-anwendungen mit künstlicher intelligenz - Google Patents
Co-design von optischem filter und fluoreszenz-anwendungen mit künstlicher intelligenz Download PDFInfo
- Publication number
- WO2022263515A1 WO2022263515A1 PCT/EP2022/066313 EP2022066313W WO2022263515A1 WO 2022263515 A1 WO2022263515 A1 WO 2022263515A1 EP 2022066313 W EP2022066313 W EP 2022066313W WO 2022263515 A1 WO2022263515 A1 WO 2022263515A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- digital
- machine learning
- training
- learning system
- image
- Prior art date
Links
- 230000003287 optical effect Effects 0.000 title claims abstract description 73
- 238000013473 artificial intelligence Methods 0.000 title description 3
- 238000013461 design Methods 0.000 title description 2
- 238000012549 training Methods 0.000 claims abstract description 121
- 238000010801 machine learning Methods 0.000 claims abstract description 110
- 238000000034 method Methods 0.000 claims abstract description 61
- 238000002073 fluorescence micrograph Methods 0.000 claims abstract description 18
- 230000015654 memory Effects 0.000 claims description 13
- 230000003595 spectral effect Effects 0.000 claims description 12
- 238000003860 storage Methods 0.000 claims description 9
- 238000004590 computer program Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 11
- 238000005457 optimization Methods 0.000 description 11
- 230000035945 sensitivity Effects 0.000 description 11
- 230000006870 function Effects 0.000 description 9
- 230000005540 biological transmission Effects 0.000 description 6
- 239000002872 contrast media Substances 0.000 description 5
- 238000009826 distribution Methods 0.000 description 5
- 238000005286 illumination Methods 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 206010028980 Neoplasm Diseases 0.000 description 4
- 238000001574 biopsy Methods 0.000 description 4
- 210000005013 brain tissue Anatomy 0.000 description 4
- 238000011282 treatment Methods 0.000 description 4
- 230000001419 dependent effect Effects 0.000 description 3
- 238000001454 recorded image Methods 0.000 description 3
- 238000001228 spectrum Methods 0.000 description 3
- ZGXJTSGNIOSYLO-UHFFFAOYSA-N 88755TAZ87 Chemical compound NCC(=O)CCC(O)=O ZGXJTSGNIOSYLO-UHFFFAOYSA-N 0.000 description 2
- 208000003174 Brain Neoplasms Diseases 0.000 description 2
- 238000003491 array Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 201000011510 cancer Diseases 0.000 description 2
- 238000002512 chemotherapy Methods 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000005284 excitation Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 238000002271 resection Methods 0.000 description 2
- 238000001356 surgical procedure Methods 0.000 description 2
- 229960002749 aminolevulinic acid Drugs 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 239000012620 biological material Substances 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000013434 data augmentation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- NJDNXYGOVLYJHP-UHFFFAOYSA-L disodium;2-(3-oxido-6-oxoxanthen-9-yl)benzoate Chemical compound [Na+].[Na+].[O-]C(=O)C1=CC=CC=C1C1=C2C=CC(=O)C=C2OC2=CC([O-])=CC=C21 NJDNXYGOVLYJHP-UHFFFAOYSA-L 0.000 description 1
- 238000000799 fluorescence microscopy Methods 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
- 238000001959 radiotherapy Methods 0.000 description 1
- 230000003014 reinforcing effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000011273 social behavior Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000011477 surgical intervention Methods 0.000 description 1
- 230000004083 survival effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 229910052724 xenon Inorganic materials 0.000 description 1
- FHNFHKCVQCLJFQ-UHFFFAOYSA-N xenon atom Chemical compound [Xe] FHNFHKCVQCLJFQ-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10056—Microscopic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30016—Brain
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30096—Tumor; Lesion
Definitions
- the invention relates to a prediction method using a machine learning model, and more particularly to a computer-implemented method for predicting digital fluorescence images.
- the invention further relates to a prediction system for predicting digital fluorescence images and a computer program product.
- Brain tumors relate to a type of cancer that is not uncommon, which is comparatively aggressive and often has relatively poor treatment success with a survival chance of approximately 1/3. Treatment of such disorders typically requires surgical removal, radiation therapy, and/or subsequent, usually lengthy, chemotherapy. A biopsy often forms the basis for the decision for the respective treatment, whereby molecular tests are also used. Of course, there are medical risks associated with such procedures. Due to the recently very advanced possibilities of analyzing radiometrically recorded images, such tumor examinations can at least complement biopsies. This may be the case when biopsies are not possible or undesirable. Recently, these image-based diagnoses can also be used during an operation. However, the required computing power is currently extremely high, which is why real real-time support has not yet been possible.
- a computer-implemented method for predicting digital fluorescence images can include recording a first digital image of a tissue sample using a microsurgical optical system with a first digital image recording unit with a first number of color channel information items—or number of color channels—using white light and at least one optical filter. Furthermore, the method can predict a second digital image in the form of a digital fluorescence representation of the recorded first digital image using a trained have a machine learning system which has a trained learning model for predicting a corresponding digital fluorescence representation of an input image.
- the first recorded digital image can be used as an input image for the trained machine learning system, and parameter values of the at least one optical filter can have been determined during training of the machine learning system.
- a prediction system for predicting digital fluorescence images can be a microsurgical optical system with a first digital image recording unit with a first number of color channel information - or number of color channels - and an optical filter for recording a first digital image of a tissue sample using white light and at least one trained machine learning system exhibit.
- the machine learning system can have a trained learning model for predicting a corresponding digital fluorescence representation of an input image. It can be adapted to predict a second digital image in the form of a digital fluorescence representation of the captured first digital image.
- the first recorded digital image can serve as an input image for the trained machine learning system, and parameter values of the at least one optical filter can have been determined during training of the machine learning system.
- the practice according to which different powerful underlying computer systems are used for the creation of the machine learning model in the training phase and the use of the machine learning model in the prediction phase, is further developed by the method proposed here.
- the time available for the development of the machine learning model is longer in the training phase than in the prediction phase.
- the prediction phase i.e. during productive use - it should be possible to output the machine learning system with the shortest possible time delay. This is particularly essential when the machine learning system is used in real-time situations - such as to support surgical procedures.
- a hyperspectral camera can be used to create the training data, which has a large number of different makes color channel information available. From this starting point, the number of color channels is then significantly reduced by a digitally simulated optical filter in an integrated approach during the training process in order to anticipate the available resources during the prediction phase already during the training phase. For this purpose, both the machine learning system with its integrated machine learning model and the parameters of the digitally simulated optical filter are adjusted or trained simultaneously in an integrated optimization process. In addition, additional boundary conditions such as the spectrum of the illumination light used for the tissue sample illuminated by the digital recording system can be taken into account.
- the determined parameters of the digitally simulated optical filter can then be transferred to a real, existing physical filter.
- a digital recording unit e.g. an RGB camera
- a digital recording unit e.g. an RGB camera
- the boundary conditions of the digital recording unit that will later be available for productive use - e.g. the RGB camera - could already be specified during the training.
- the surgeon can thus be provided with a representation of the tissue to be operated on in a fluorescence representation, which allows him to clearly and without further distracting his attention distinguish between healthy and diseased tissue. This applies even though in productive operation - i.e. during the prediction phase (inference) - only a comparatively simple camera (in contrast to a hyperspectral camera for the training phase) can be used.
- surgeon has a very advantageous optical representation of the tissue to be operated on in real time without the need for time-consuming, multi-dimensional biopsies.
- the system presented here is ultimately based very advantageously on an optimization of hardware in the form of the optical filter to be physically produced for the prediction phase, the parameters of which are determined during the integrated optimization during the training phase, and the software of the machine learning system used.
- the trained learning model of the trained machine learning system can provide a plurality of first digital training images of tissue samples that were recorded under white light using a microsurgical optical system with a second image recording unit.
- a second number of color channel information items for each first digital training image, a second number of color channel information items—for example 64 color channels of a hyperspectral camera—can be available for different spectral ranges.
- the method can also include providing a plurality of second digital training images, each representing the same tissue samples as the first set of digital training images, wherein the second digital training images can have indications of diseased elements of the tissue samples, and training the machine learning system to form the trained machine learning model to predict a digital image of a type of the plurality of second digital training images.
- the following can be used as input values for the machine learning system: the majority of the first digital training images in the form of the second number of color channel information (ie e.g. 64 color channels), the majority of the second digital training images as "ground truth", parameter values for reducing the second number the color channel information through at least one digitally simulated optical filter to form the first plurality of color channel information.
- the expert knows the term "ground truth" as the learning objective of the training.
- the majority of the first digital training images can be used as training data for predicting a digital image of the type of majority of the second digital training images, after the second number of color channel information items has been reduced to the first number of color channel information items by means of the digitally simulated optical filter .
- at least some of the parameter values of the at least one optical filter can be output as output values of the machine learning system after the training of the machine learning system has been completed.
- the part of the parameter values of the at least one optical filter can also mean that all determined parameter values are output.
- tissue samples can be, for example, diseased tissue in the form of cancerous tissue, in particular diseased brain tissue.
- second training images were generated as images taken under visible light and/or under special lighting conditions - e.g. UV light.
- the parameter values of the at least one optical filter can have the number of first color channel information items and/or a filter shape of the digitally simulated optical filter.
- the filter can also consist of a plurality of digitally simulated optical filters. In this way, different parameter values of the optical filter can be varied. In the production of optical filters, several of these parameter values can be realized at the same time.
- the second number of color channel information (i.e. the number of color channels) can be larger than the first number of color channel information.
- the second number can represent the color channel number of a hyperspectral camera - e.g. 64 color channels (more generally e.g. 30 to 130 color channels) - while the first number represents the color channel number of the first digital image acquisition unit.
- This can have under 10 - or preferably 3 to 4 - color channels.
- the number of color channels can be significantly reduced, which means that the required computing power can be significantly reduced. This has a positive effect on using the method directly during an ongoing operation.
- the parameter values for reducing the second number of color channels can relate to a filter form of the filter or to a respective center frequency of the first number of color channel information items.
- the parameter values for reducing the second number of color channels can also relate to one or more camera sensitivity curves (eg one per color channel) as an envelope or their number and/or shapes of the geometrically described structures; this can refer to a Gaussian distribution, a rectangular shape, etc.
- parameter values for controlling the source of the white light during the recording of the first digital image can be generated as additional output values of the machine learning system after the training of the machine learning system has been completed.
- the environment variables that were used during the training of the machine learning system with its machine learning model can also be used for the actual, productive use of the method.
- the digital fluorescence representation may correspond to a representation as would be produced using a light source in the UV range - e.g., BLUE400.
- spectral ranges can also be displayed preferentially.
- the term “fluorescence display” should also be usable for these other spectral ranges.
- a contrast agent in the tissue sample would make sense for a fluorescence display: for example 5-ALA (aminolevulinic acid) for excitation in the wavelength range 430-440nm or sodium fluorescein for excitation in the wavelength range around 560nm.
- the structure of the learning model can correspond to an encoder-decoder model.
- embodiments may relate to a computer program product, accessible from a computer-usable or computer-readable medium, having program code for use by, or in connection with, a computer or other instruction processing system.
- a computer-usable or computer-readable medium can be any device suitable for storing, communicating, transmitting, or transporting program code.
- FIG. 1 shows a flowchart-like representation of an embodiment of the computer-implemented method according to the invention for the prediction of digital fluorescence images.
- FIG. 2 shows an extension of the exemplary embodiment according to FIG. 1.
- FIG 3 shows a diagram of an embodiment for the training phase and the prediction phase for the machine learning system and a use of the filters.
- Figure 4 shows a diagram of an embodiment of the number of color channels during training.
- Fig. 5 represents two possible wavelength distributions as a result of joint optimization of the machine learning system and the digitally simulated filter.
- FIG. 7 illustrates an embodiment of a computer system that includes the system of FIG.
- the term 'machine learning system' can describe a system or a method here, with the aid of which output values are generated in a non-procedurally programmed manner.
- a machine learning model present in the machine learning system is trained with training data and the associated desired output values (annotated data or ground truth data).
- the productive phase i.e. the prediction phase (inference phase)
- the prediction phase inference phase
- This also includes neural networks, which can be trained and used as classifiers, for example.
- the desired output values for given input values are typically learned using a method called "backpropagation", in which parameter values of nodes in the neural network or connections between the nodes are automatically adjusted.
- backpropagation a method in which parameter values of nodes in the neural network or connections between the nodes are automatically adjusted.
- the inherent machine learning model is adjusted or trained to form the trained machine learning system with the trained machine learning model.
- the term 'prediction' can describe the phase of the productive use of a machine learning system according to what has been said before. During the prediction phase of the machine learning system, output values are generated or predicted on the basis of the trained machine learning model, which is provided with previously unknown input data.
- the term 'digital fluorescence image' here describes a digital image of a tissue sample whose appearance is that of an image which is viewed under light of a specific wavelength (eg UV light) with the aid of a contrast agent previously added to the tissue sample.
- the term 'first digital image' here describes a recording of a digital image with a digital image recording unit (eg RGB camera) with a low number of color channels (eg monochrome or 3 to 4 color channels, generally less than 10 color channels). This first digital image is acquired during the prediction phase in order to predict a digital fluorescence image from it.
- the terms 'number of color channels' and 'number of color channel information' can be understood as synonyms in the context of this application.
- tissue sample' here can describe biological material. In this method, it is advantageously available for investigations. This can be “normal” human tissue or brain tissue.
- microsurgical optical system' can describe a surgical microscope here. This can be used for surgical operations - for example minimally invasive interventions. It can have lighting, a digital recording unit (e.g. a digital camera with one or more color channels), an image processing and operating unit and an output screen.
- a digital recording unit e.g. a digital camera with one or more color channels
- an image processing and operating unit e.g. a digital camera with one or more color channels
- the term 'digital image recording unit' can describe a camera with a digital image converter. This can capture multiple color channel information in parallel. In the case of a hyperspectral camera, for example, 64 color channels can be recorded; however, other numbers of color channels are also conceivable (e.g. 30 to 130). Such a camera could be used for capturing training images, while a camera with a much more common number of color channels - e.g., on the order of 3 to 4 (or monochrome) - would be usable for productive use during the prediction phase.
- the term 'first number of color channel information' can describe the number of color channels of the image acquisition unit that is used in the prediction mode.
- the term 'optical filter' here can be a device that allows incident optical rays to pass selectively according to certain criteria.
- the criteria can be wavelength-selective (or polarization-state dependent), so that the filter is more permeable to certain different wavelength ranges and less permeable to others (up to practically not at all).
- the term 'second digital image' may describe the predicted image generated by the machine learning system in production.
- the term 'digital fluorescence representation' may mean a specific representation of a digital image that appears as if illuminated with light of a specific wavelength (e.g., UV light) causing fluorescence effects related to the light of the specific wavelength to a contrast agent present, for example, in a biological tissue.
- a specific wavelength e.g., UV light
- the term 'white light' describes light from a light source (e.g. white light LED, xenon, etc.) which emits primarily in the visible wave spectrum.
- a light source e.g. white light LED, xenon, etc.
- the term “parameter values of the at least one optical filter” essentially describes wavelengths—or their ranges—in which the filter is transparent.
- the envelope in a transmission-versus-wavelength representation can be another parameter - just like a mean value of the transmission range.
- first digital training image' here describes a digital image for which a large number (e.g. 64 or more generally: 30 to 130) color channels are available.
- the term 'second image acquisition unit' describes an image acquisition unit which is adapted to provide a large number of color channel data for an acquisition, as is the case with a hyperspectral camera, for example.
- the term 'second digital training image' describes a digital image of the ground truth data that is essential for training a machine learning system in supervised learning.
- the ground truth data reflects the kind of digital images that are expected to be predicted (i.e., to be "learned").
- the term 'indications of diseased elements of the tissue samples' can be characterized by different annotations in a digital image. It can be a pixel-by-pixel annotation of a specific image section, an area marked in some other way, or a specific form of segmentation of the digital image.
- the term 'parameter values for reducing the second number of color channel information' can describe characteristics of the digitally simulated filter. It can affect the number or the transmission characteristics of the filter.
- the term 'digitally simulated optical filter' denotes a unit of the machine learning system that modifies the color channel information. These can be digital color transmission filters.
- the term 'filter form of the digitally simulated optical filter' can refer, for example, to the envelope of the spectral lines (or wavelength range) transparent to the filter, their mean values or an associated standard deviation.
- the term 'white light source control parameter values' can describe characteristics of the white light used to create the digital images. Essentially, these can be spectral range and intensity values of the light emitted by the illumination source.
- the term 'encoder-decoder model' here describes an architecture of a machine learning system in which input data is encoded or coded in order to then be decoded again immediately afterwards. In the middle between the encoder and the decoder, the necessary data is available as a type of feature vector. Depending on the training of the machine learning model, certain features in the input data can then be particularly emphasized during decoding.
- 'U-Net architecture' describes here an architecture of a machine learning model, which internally has a contracting and an expanding path. A more detailed definition is given in connection with FIG.
- the method includes recording 102 a first digital image of a tissue sample - for example, from diseased brain tissue - by a microsurgical optical system with a first digital image acquisition unit, eg one RGB camera - with a first number of color channel information - eg with 3 to 4 color channels - using white light and at least one optical filter.
- the filter is typically located between the illuminated tissue sample and the camera.
- the method 100 has a prediction 104 of a second digital image in the form of a digital fluorescence representation of the recorded first digital image by means of a trained machine learning system, which has a trained learning model for predicting a corresponding digital fluorescence representation of an input image.
- the first recorded digital image for the trained machine learning system is used as input data or input image; in addition, parameter values of the at least one optical filter were determined during training of the machine learning system.
- FIG. 2 shows an extension of the exemplary embodiment according to FIG. 1, in particular the training phase 200 of the method described above, which is shown in FIG. 1 during its productive or prediction phase.
- the training of the combined machine learning model of the trained combined machine learning system includes providing 202 a plurality of first digital training images of tissue samples - i.e. again brain tissue or cancer tissue, for example - which are examined under white light by means of a microsurgical optical system - i.e., e.g. a surgical microscope - Were recorded with a second image recording unit, wherein a second number of color channel information for different spectral ranges is available for each first digital training image. This second number is typically larger than the first number of color channels in the productive prediction mode according to Fig. 1.
- a hyperspectral camera with, for example, 64 color channels would be used advantageously.
- the training phase 200 includes providing 204 a plurality of second digital training images, each of which represents the same tissue samples as the first set of digital training images.
- the second digital training images contain indications of diseased elements in the tissue samples. These indications can take various forms, such as pixel-by-pixel annotations, optically emphasized encircling and/or borders of diseased tissue areas or other image segmentations.
- the representation can be both that under normal visible eg white light or under special light and highlighted with a contrast agent (fluorescence imaging).
- the training phase 200 comprises the actual training 206 of the machine learning system - under so-called supervised learning - to form the trained machine learning model for predicting a digital image of a kind of the plurality of the second digital training images, i.e. with visible markers of diseased tissue.
- the following input values are used for the combined machine learning system: (i) the majority of the first digital training images in the form of the second quantity of color channel information - i.e. with the higher quantity of color channel information - (ii) the majority of the second digital training images as ground truth - i.e. the desired result images to be trained - and (iii) parameter values for reducing the second set of color channel information by at least one digitally simulated optical filter to form the first set of color channel information.
- the parameter values can include the number of filters, information about color spectra or filter shapes.
- the plurality of first digital training images are used after reducing the second number of color channel information to the first number of color channel information by the digitally simulated optical filter.
- parameter values of the at least one optical filter are output as output values of the machine learning system after the training of the machine learning system has been completed. This means that the learning system learns the properties of the simulated optical filter, so to speak, during the training of the machine learning system to form the machine learning model.
- the parameter values of the at least one optical filter can, for example, relate to a number of filters or associated frequency bands. The properties of the digitally simulated filter can then be used directly to produce such an optical filter for use in the prediction phase, as also described in more detail in FIG.
- Figure 3 shows a simultaneous representation 300 of the training phase and the prediction phase for the machine learning system and usage of the filters.
- the left side of the figure relates to the training phase, while the right side of the figure relates to the prediction phase .
- Multiple tissue samples 302 are combined by a digital Recording system 304 with a high number of color channels 306 for generating training data from the tissue 302 recorded.
- These digital images (first training data) are passed through a digitally simulated filter 308 so that these digital images 310 are available behind the digitally simulated filter 308 in a representation with a reduced number of color channels 310 .
- the properties of the optical filter(s) 308 are varied, thereby influencing the learning speed and the learning success (fast or slow convergence) of the machine learning system 312.
- the phase of using the trained machine learning system 324 also begins with a recording of a biological tissue 302a.
- a biological tissue 302a This can happen, for example, during an operation or an examination using an image recording unit 320 - i.e. e.g. an RGB camera.
- the potentially diseased tissue 302a is illuminated, for example, with white light (illumination source not shown) and converted into a digital image by the recording unit/camera 320 after the rays have been filtered by the optical filter or filters 318 .
- This digital image consists of the color information of, for example, 3-4 color channels 322.
- the number of color channels 322 corresponds to that of the color channels 312 that are selected from the color channels 306 of the hyperspectral camera 304 by the digitally simulated optical filter 310 during the training of the machine learning system 312 were generated.
- the respective equivalent elements of the training phase and the prediction phase are indicated by dashed arrows from the left side of FIG. 3 to the right side of FIG.
- the digital image captured by the camera 320 (e.g., an RGB camera) with the fewer number of color channels 322 becomes the trained machine Learning system 324 supplied - ie used as input data - to output a digital output image 326 in the form of a fluorescence display.
- the filter 308 is simulated digitally during the training of the machine learning system 312, while the filter 318, which is used during the prediction phase of the trained machine learning system 324, is a real existing filter , which has the optical characteristics or parameters that the trained machine learning system 314 also outputs at the end of the training. Consequently, the real optical filter 318 is manufactured exactly according to the specifications determined during the joint optimization 314 of the digitally simulated filter 308 and the machine learning system 312 parameters in training. Ground truth data 316 are required as additional input values for the training.
- Fig. 4 shows a diagram of an embodiment 400 in terms of the number of color channels during training are neutralized (i.e. made independent), so that these properties can also be used as additional constraints when training the machine learning system 414.
- boundary conditions can be their number as well as the availability of the light at different wavelengths, its density and the respective intensity.
- the color information that is then available through the same number of color channels 406 for a recorded image is then abstracted from the properties of the light source used.
- the boundary conditions of the simulated optical filter 408, which are optimized together with the properties of the machine learning system 414 (cf. correspondingly the machine learning system 312, 324 of FIG. 3), can be taken into account.
- camera sensitivity parameters in particular with regard to specific wavelengths
- additional boundary conditions 410 can also be taken into account as additional boundary conditions 410 .
- the result is a digital image with a reduced number 412 of color channels, which is smaller (or significantly smaller) than the number of color channels 402 of the digital input image.
- the double arrows between the symbolically represented boundary conditions 404, 408, 410 symbolize the influence on the machine learning system 414 during the training.
- the machine learning system 414 is also supplied with reference data (ground truth data) 416 during the training, which data represent the respective expected result for a digital training image.
- This ground truth data 416 ultimately represents the expected output image in fluorescence representation, which is expected (to be predicted) given the presence of a digital input image with the high number of color channels 402 .
- further parameters 418 can be output. These primarily relate to the characteristics of the digitally simulated optical filter 408 and, for example, spectral range properties of the illumination source for the prediction phase or also spectral range sensitivity parameters of the digital recording unit to be used in the prediction phase (i.e. the RGB camera, a monochromatic camera, or similar) .
- a U-Net for example, can be used as the machine learning system 414 .
- this consists of a contracting path and an expanding path, as indicated by the symbol 414 .
- it may consist of a repeated application of two 3x3 unpadded convolutions, each preceded by a reinforcing linear unit (ReLU) and a 2x2 pooling operation with steps of 2 for downsampling. The number of feature channels is doubled with each downsampling step.
- ReLU reinforcing linear unit
- the expanding path consists of upsampling the feature map, followed by a 2x2 convolution (“up-sampling”), which bisects the number of feature channels in each case, and lining them up through the contracting path accordingly "Cut" feature images and two 3x3 folds, each of which is followed by a ReLU.
- Diagram 502 essentially shows three separate frequency /wavelength ranges that can easily be converted into an RGB representation.
- the diagram 504 has four filter areas with different attenuation for different frequency ranges. It should be pointed out that there are actually only two possible examples in which each spectral range can represent a sensor channel of the digital recording unit.
- the height of the vertical lines corresponds to the weighted sensitivities of the respective filter at the respective wavelength.
- An additional boundary condition of the representation 502 relates to the fact that the three channels each have a Gaussian sensitivity distribution with a relatively small standard deviation.
- the representation 504 has four maxima, one of which can be in the blue area, one in the green area and two in the red area, for example.
- the standard deviation is comparatively small. If no optimization were provided during the learning phase of the machine learning system, the intensity lines of potential filters would be distributed mathematically more chaotically along the x-axis (not shown); no separate spectral ranges would be recognizable that could be converted into a real, physical optical filter.
- the real optical filter (cf. 318, Fig. 3) has exactly the same sensitivity curves.
- the representation 502 represents a fully described sensitivity specification (i.e. no variation of the filter parameters is made during training).
- the representation 504 would then correspond to the characteristics (or parameters) of the sensitivity curves of a camera to be used later (during the prediction phase); i.e., Gaussian sensitivity curves with boundary conditions such as small standard deviation and a specific value for the highest sensitivity value of a given wavelength etc.
- Gaussian sensitivity curves with boundary conditions such as small standard deviation and a specific value for the highest sensitivity value of a given wavelength etc.
- the last case described in the penultimate paragraph would probably be difficult to implement in a 1:1 manner in terms of hardware in terms of an actual physical filter to realize.
- properties of the light source used can be used as additional boundary conditions during the training phase in further embodiments.
- Further embodiments can provide a chronological sequence of digitally recorded images.
- the goal here would be to avoid rapidly changing digital speckle formation when using consecutive images (“flickering”).
- post-processing of the individual recorded frames could be provided through smoother transitions in time.
- the generation of a 3D model so that all 2D operations of the 2D model are performed in 3D space ie 3D convolutions, 3D pooling, etc.
- a combination of a 2D model with time-limited models can also be used.
- the prediction system 600 for predicting digital fluorescence images.
- the prediction system has a memory 604, which stores program code, and one or more processors 602 connected to the memory, which, when they execute the program code, cause the prediction system 600 to control the following units: a first digital image acquisition unit 606 with a first Number of color channel information for capturing a first digital image of a tissue sample by a microsurgical optical system using white light and at least one optical filter and a trained machine learning system 608 for predicting a second digital image in the form of a digital fluorescence representation of the captured first digital image.
- the trained machine learning system has a trained learning model for predicting a corresponding digital fluorescence representation of an input image; the first captured digital image is used as an input image for the trained machine learning system; and parameter values of the at least one optical filter were determined during training of the machine learning system.
- the prediction system 600 can either explicitly have a second image acquisition unit 610 for a plurality of first digital training images of tissue samples, which were acquired under white light by means of a microsurgical optical system, or through the interaction of the program code stored in the memory 604 with the processor 602 fulfill their function.
- a second set of color channel information for different spectral ranges is available for each first digital training image.
- a provision unit 612 for providing a plurality of second digital training images can be present, each representing the same tissue samples as the first set of digital training images, the second digital training images having indications of diseased elements of the tissue samples.
- the provisioning unit 612 may be implemented as an image store.
- a training system 614 for training the machine learning system to form the trained machine learning model for predicting a digital image of one type of the plurality of second digital training images may be provided.
- the following is used as input values for the machine learning system: the majority of the first digital training images in the form of the second quantity of color channel information, the majority of the second digital training images as ground truth (or ground truth data) and parameter values for reducing the second Number of color channel information through at least one digitally simulated optical filter to form the first number of color channel information, wherein the plurality of first digital training images is used as training data for predicting a digital image of the type of the plurality of second digital training images after the second number of color channels information was reduced to the first number of color channel information by means of the digitally simulated optical filter.
- Parameter values of the at least one optical filter are also output as output values of the machine learning system after the training of the machine learning system has been completed.
- modules and/or units - in particular the processor 602, the memory 604, the first digital image acquisition unit 606, the trained machine learning system 608, the second image acquisition unit 610, the provision unit 612 and the training system 614 - Can be connected to electrical signal lines or via a system-internal bus system 616 for the purpose of signal and/or data exchange and cooperative behavior.
- Figure 7 illustrates a diagram of a computer system 700 that may include at least portions of the prediction system.
- embodiments of the concept proposed here can be used with practically any type of computer, regardless of the platform used therein for storing and/or executing program codes.
- FIG. 6 shows a computer system 700 by way of example, which is suitable for executing program code in accordance with the method presented here.
- a computer system already present in a surgical microscope can also—possibly with appropriate extensions—serve as a computer system for executing the concept presented here.
- the computer system 700 has a plurality of general purpose functions.
- the computer system can be a tablet computer, a laptop/notebook computer, another portable or mobile electronic device, a microprocessor system, a microprocessor-based system, a computer system with specially set up special functions or also a component of a computer-controlled microscope system.
- the computer system 700 may be configured to execute instructions executable by a computer system - such as program modules - which can be executed to implement functions of the concepts proposed here.
- the program modules can have routines, programs, objects, components, logic, data structures, etc. in order to implement specific tasks or specific abstract data types.
- the components of the computer system may include: one or more processors or processing units 702, a memory system 704, and a bus system 706 that connects various system components, including the memory system 704, to the processor 702.
- computer system 700 includes a plurality of accessible volatile or non-volatile storage media to which computer system 700 has access.
- the data and/or instructions (commands) of the storage media can be stored in volatile form--such as in a RAM (Random Access Memory) 708--to be executed by the processor 702.
- RAM Random Access Memory
- Further components of the memory system 704 can be a permanent memory (ROM) 710 and a long-term memory 712 in which the program modules and data (reference number 716) as well as workflows can be stored.
- the computer system has a number of dedicated devices (keyboard 718, mouse pointing device (not shown), display 720, etc.) for communication. These dedicated devices can also be integrated into a touch-sensitive display.
- a separately provided I/O controller 714 ensures smooth data exchange with external devices.
- a network adapter 722 is available for communication via a local or global network (LAN, WAN, for example via the Internet). The network adapter can be accessed by other components of computer system 700 via bus system 706 . It should be understood that other devices may be connected to computer system 700, although not shown.
- the prediction system 600 for predicting digital fluorescence images can be connected to the bus system 706.
- the prediction system 600 and the computer system 700 may share memory and/or the processor, if desired.
- the principle presented here can be embodied both as a system, as a method, combinations thereof and/or also as a computer program product.
- the computer program product may include one (or more) computer-readable storage medium(s) having computer-readable program instructions for causing a processor or a control system to perform various aspects of the present invention.
- Electronic, magnetic, optical, electromagnetic, infrared media or semiconductor systems are used as the transmission medium; for example SSDs (solid state device/drive), RAM (Random Access Memory) and/or ROM (Read-Only Memory), EEPROM (Electrically Eraseable ROM) or any combination thereof.
- Solid state device/drive solid state device/drive
- RAM Random Access Memory
- ROM Read-Only Memory
- EEPROM Electrically Eraseable ROM
- Propagating electromagnetic waves, electromagnetic waves in waveguides or other transmission media (e.g. light pulses in optical cables) or electrical signals that are transmitted in wires can also be considered as transmission media.
- the computer-readable storage medium may be an embodying device that stores instructions for use by an instruction execution device.
- the computer-readable program instructions described here can also be downloaded onto a corresponding computer system, for example as a (smartphone) app from a service provider via a cable-based connection or a mobile network.
- the computer-readable program instructions for performing operations of the invention described herein may be machine-dependent or machine-independent instructions, microcode, firmware, state-defining data, or any source or object code written, for example, in C++, Java, or the like or written in conventional procedural programming languages, such as the "C" programming language or similar programming languages.
- the computer-readable program instructions can be executed entirely by a computer system. In some According to embodiments, it can also be electronic circuits, such as programmable logic circuits, field-programmable gate arrays (FPGA) or programmable logic arrays (PLA), which execute the computer-readable program instructions by using status information of the computer-readable program instructions in order to to configure or individualize electronic circuits according to aspects of the present invention.
- FPGA field-programmable gate arrays
- PLA programmable logic arrays
- the computer readable program instructions may be provided to a general purpose computer, special purpose computer, or other programmable data processing system to produce a machine such that the instructions are executed by the processor or the computer or other programmable data processing device , generate means to implement the functions or operations illustrated in the flowchart and/or block diagrams.
- these computer-readable program instructions can also be stored on a computer-readable storage medium.
- each block in the illustrated flowchart or block diagrams may represent a module, segment, or portions of instructions, which represent a plurality of executable instructions for implementing the specific logical function.
- the functions that are shown in the individual blocks can be executed in a different order—if appropriate, also in parallel.
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Investigating, Analyzing Materials By Fluorescence Or Luminescence (AREA)
- Image Analysis (AREA)
Abstract
Description
Claims
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202280042864.2A CN117546206A (zh) | 2021-06-16 | 2022-06-15 | 使用人工智能的光学滤波器和荧光应用的协同设计 |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
DE102021115587.8A DE102021115587A1 (de) | 2021-06-16 | 2021-06-16 | Co-design von optischem filter und fluoreszenz-anwendungen mit künstlicher intelligenz |
DE102021115587.8 | 2021-06-16 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022263515A1 true WO2022263515A1 (de) | 2022-12-22 |
Family
ID=82385317
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/EP2022/066313 WO2022263515A1 (de) | 2021-06-16 | 2022-06-15 | Co-design von optischem filter und fluoreszenz-anwendungen mit künstlicher intelligenz |
Country Status (3)
Country | Link |
---|---|
CN (1) | CN117546206A (de) |
DE (1) | DE102021115587A1 (de) |
WO (1) | WO2022263515A1 (de) |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190384048A1 (en) * | 2012-01-20 | 2019-12-19 | The Trustees Of Dartmouth College | Method and apparatus for quantitative hyperspectral fluorescence and reflectance imaging for surgical guidance |
-
2021
- 2021-06-16 DE DE102021115587.8A patent/DE102021115587A1/de active Pending
-
2022
- 2022-06-15 WO PCT/EP2022/066313 patent/WO2022263515A1/de active Application Filing
- 2022-06-15 CN CN202280042864.2A patent/CN117546206A/zh active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190384048A1 (en) * | 2012-01-20 | 2019-12-19 | The Trustees Of Dartmouth College | Method and apparatus for quantitative hyperspectral fluorescence and reflectance imaging for surgical guidance |
Non-Patent Citations (3)
Title |
---|
COLIN L COOKE ET AL: "Physics-enhanced machine learning for virtual fluorescence microscopy", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, 201 OLIN LIBRARY CORNELL UNIVERSITY ITHACA, NY 14853, 9 April 2020 (2020-04-09), XP081641026 * |
ERIC M CHRISTIANSEN ET AL: "In Silico Labeling: Predicting Fluorescent Labels in Unlabeled Images", CELL, vol. 173, no. 3, 19 April 2018 (2018-04-19), USA, pages 792 - 803, e1, XP002788720, ISSN: 1097-4172, Retrieved from the Internet <URL:https://www.sciencedirect.com/science/article/pii/S0092867418303647?via%3Dihub> [retrieved on 20190218], DOI: 10.1016/J.CELL.2018.03.040 * |
NIE SHIJIE ET AL: "Deeply Learned Filter Response Functions for Hyperspectral Reconstruction", 2018 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, IEEE, 18 June 2018 (2018-06-18), pages 4767 - 4776, XP033473388, DOI: 10.1109/CVPR.2018.00501 * |
Also Published As
Publication number | Publication date |
---|---|
DE102021115587A1 (de) | 2022-12-22 |
CN117546206A (zh) | 2024-02-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
EP2074933B1 (de) | Verfahren zur Analyse und Bearbeitung von Fluoreszenzbildern | |
DE102011078278A1 (de) | Verfahren zur Bilderzeugung und Bildauswertung | |
EP0896661A1 (de) | Verfahren zur automatisierten mikroskopunterstützten untersuchung von gewebeproben oder körperflüssigkeitsproben | |
EP3195250B1 (de) | Verfahren zum erzeugen eines ergebnisbilds und optische vorrichtung | |
DE102009035441B4 (de) | Verfahren und Bildverarbeitungssystem zur Erzeugung eines Volumenansichtsbilds vom Inneren eines Körpers | |
EP3332710B1 (de) | Charakterisierung von plaque | |
DE112004001440T5 (de) | Entfalten eines virtuellen Organs zur Visualisierung | |
DE112015006295T5 (de) | Bildverarbeitungsvorrichtung, Vorrichtung zur biologischen Beobachtung und Bildverarbeitungsverfahren | |
EP3540632A1 (de) | Verfahren und analysegeräte zum klassifizieren von gewebeproben | |
WO2019091509A1 (de) | Verfahren zur auswertung des hautbildes eines menschen im rahmen der hautkrebs- vorsorgeuntersuchung sowie vorrichtung zu dessen betrieb | |
DE202018006284U1 (de) | Vorrichtung, insbesondere für Mikroskope und Endoskope, unter Anwendung einer Grundlinienschätzung und halbquadratischen Minimierung für die Unschärfereduzierung von Bildern | |
DE102009042326A1 (de) | Interaktive Veränderung der Darstellung eines mittels Volume Rendering dargestellten Objekts | |
DE102021210435A1 (de) | Verfahren und Vorrichtung zur Visualisierung von dreidimensionalen Objekten | |
DE102009032442B4 (de) | Verfahren und Vorrichtung zur Nachbearbeitung von mittels eines bildgebenden Systems erfassten Messwerten | |
DE112018000393T5 (de) | Endoskopsystem und Bildanzeigevorrichtung | |
DE102006012943B4 (de) | Verfahren zur automatischen Auswertung eines Abbildungsdatensatzes eines Objektes und medizinisches Abbildungssystem | |
DE102017204175A1 (de) | Verfahren zu einem Auffinden von Auffälligkeiten in medizinischen Bilddaten eines Bereichs des Patienten, der außerhalb eines zu untersuchenden Bereichs des Patienten angeordnet ist, sowie eine medizinische Bildgebungsvorrichtung hierzu | |
EP3379487B1 (de) | Kontrastverstärkte wiedergabe von spektralen ct-bilddaten | |
DE102004056589A1 (de) | Verfahren und Vorrichtung zur Durchführung segmentierungsbasierter Bildoperationen | |
WO2022263515A1 (de) | Co-design von optischem filter und fluoreszenz-anwendungen mit künstlicher intelligenz | |
DE102017203048B3 (de) | Verfahren zum Bestimmen eines Projektionsdatensatzes, Projektionsbestimmungssystem, Computerprogrammprodukt und computerlesbares Speichermedium | |
DE102009006765A1 (de) | Verfahren zur Ansteuerung einer Bildausgabeeinrichtung, Verfahren zur Ermittlung von tangentialen Schnittebenen, Ansteuerungsmodul und Schnittebenenermittlungseinheit | |
EP3488281B1 (de) | Ermitteln eines gewebetyps eines gewebes eines tierischen oder menschlichen individuums | |
DE102021115588A1 (de) | Multi-task-lernen von weisslichtaufnahmen für ein operationsmikroskop | |
DE10224756B4 (de) | Verfahren und Vorrichtung zur Erzeugung einer kombinierten Parameterkarte |
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: 22737386 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2023577468 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 202280042864.2 Country of ref document: CN Ref document number: 18571021 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |