Classifications

• • G—PHYSICS
  • G03—PHOTOGRAPHY; CINEMATOGRAPHY; ANALOGOUS TECHNIQUES USING WAVES OTHER THAN OPTICAL WAVES; ELECTROGRAPHY; HOLOGRAPHY
  • G03H—HOLOGRAPHIC PROCESSES OR APPARATUS
  • G03H1/00—Holographic processes or apparatus using light, infrared or ultraviolet waves for obtaining holograms or for obtaining an image from them; Details peculiar thereto
  • G03H1/04—Processes or apparatus for producing holograms
  • G03H1/08—Synthesising holograms, i.e. holograms synthesized from objects or objects from holograms
  • G03H1/0866—Digital holographic imaging, i.e. synthesizing holobjects from holograms
• • G—PHYSICS
  • G02—OPTICS
  • G02B—OPTICAL ELEMENTS, SYSTEMS OR APPARATUS
  • G02B21/00—Microscopes
  • G02B21/24—Base structure
  • G02B21/241—Devices for focusing
  • G02B21/244—Devices for focusing using image analysis techniques
• • G—PHYSICS
  • G02—OPTICS
  • G02B—OPTICAL ELEMENTS, SYSTEMS OR APPARATUS
  • G02B21/00—Microscopes
  • G02B21/36—Microscopes arranged for photographic purposes or projection purposes or digital imaging or video purposes including associated control and data processing arrangements
  • G02B21/365—Control or image processing arrangements for digital or video microscopes
  • G02B21/367—Control or image processing arrangements for digital or video microscopes providing an output produced by processing a plurality of individual source images, e.g. image tiling, montage, composite images, depth sectioning, image comparison
• • G—PHYSICS
  • G03—PHOTOGRAPHY; CINEMATOGRAPHY; ANALOGOUS TECHNIQUES USING WAVES OTHER THAN OPTICAL WAVES; ELECTROGRAPHY; HOLOGRAPHY
  • G03H—HOLOGRAPHIC PROCESSES OR APPARATUS
  • G03H1/00—Holographic processes or apparatus using light, infrared or ultraviolet waves for obtaining holograms or for obtaining an image from them; Details peculiar thereto
  • G03H1/0005—Adaptation of holography to specific applications
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F18/00—Pattern recognition
  • G06F18/20—Analysing
  • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
  • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F18/00—Pattern recognition
  • G06F18/20—Analysing
  • G06F18/24—Classification techniques
  • G06F18/243—Classification techniques relating to the number of classes
  • G06F18/2431—Multiple classes
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T7/00—Image analysis
  • G06T7/0002—Inspection of images, e.g. flaw detection
  • G06T7/0012—Biomedical image inspection
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T7/00—Image analysis
  • G06T7/10—Segmentation; Edge detection
  • G06T7/136—Segmentation; Edge detection involving thresholding
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T7/00—Image analysis
  • G06T7/10—Segmentation; Edge detection
  • G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V10/00—Arrangements for image or video recognition or understanding
  • G06V10/20—Image preprocessing
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V10/00—Arrangements for image or video recognition or understanding
  • G06V10/40—Extraction of image or video features
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V10/00—Arrangements for image or video recognition or understanding
  • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
  • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V10/00—Arrangements for image or video recognition or understanding
  • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
  • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V10/00—Arrangements for image or video recognition or understanding
  • G06V10/98—Detection or correction of errors, e.g. by rescanning the pattern or by human intervention; Evaluation of the quality of the acquired patterns
  • G06V10/993—Evaluation of the quality of the acquired pattern
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V20/00—Scenes; Scene-specific elements
  • G06V20/60—Type of objects
  • G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
  • G06V20/695—Preprocessing, e.g. image segmentation
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V20/00—Scenes; Scene-specific elements
  • G06V20/60—Type of objects
  • G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
  • G06V20/698—Matching; Classification
• • G—PHYSICS
  • G03—PHOTOGRAPHY; CINEMATOGRAPHY; ANALOGOUS TECHNIQUES USING WAVES OTHER THAN OPTICAL WAVES; ELECTROGRAPHY; HOLOGRAPHY
  • G03H—HOLOGRAPHIC PROCESSES OR APPARATUS
  • G03H1/00—Holographic processes or apparatus using light, infrared or ultraviolet waves for obtaining holograms or for obtaining an image from them; Details peculiar thereto
  • G03H1/0005—Adaptation of holography to specific applications
  • G03H2001/005—Adaptation of holography to specific applications in microscopy, e.g. digital holographic microscope [DHM]
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; 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 OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2207/00—Indexing scheme for image analysis or image enhancement
  • G06T2207/20—Special algorithmic details
  • G06T2207/20112—Image segmentation details
  • G06T2207/20132—Image cropping
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2207/00—Indexing scheme for image analysis or image enhancement
  • G06T2207/20—Special algorithmic details
  • G06T2207/20212—Image combination
  • G06T2207/20216—Image averaging
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2207/00—Indexing scheme for image analysis or image enhancement
  • G06T2207/20—Special algorithmic details
  • G06T2207/20212—Image combination
  • G06T2207/20224—Image subtraction
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; 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
  • G06T2207/30024—Cell structures in vitro; Tissue sections in vitro
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; 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
  • G06T2207/30168—Image quality inspection

Definitions

• the present disclosure relates generally to using convolutional neural networks (CNN) to identify the quality of image acquired using digital holographic microscopy (DHM) and other microscopy techniques.
• CNN convolutional neural networks
• DLM digital holographic microscopy
• the various systems, methods, and apparatuses described herein may be applied to, for example, enhance classification workfiows and the subsequent diagnosis decisions by removing out of focus or poor quality images from analysis.
• Digital holographic microscopy also known as interference phase microscopy, is an imaging technology that provides the ability to quantitatively track sub- nanometric optical thickness changes in transparent specimens.
• DHM captures both phase and intensity.
• the phase information captured as a hologram, can be used to reconstruct extended morphological information (such as depth and surface characteristics) about the specimen using a computer algorithm.
• Modern DHM can be used to reconstruct extended morphological information (such as depth and surface characteristics) about the specimen using a computer algorithm.
• implementations offer several additional benefits, such as fast scanning/data acquisition speed, low noise, high resolution and the potential for label-free sample acquisition.
• DHM is particularly well suited for acquiring images of blood cells for classification purposes, or to perform subsequent diagnosis decisions.
• one of the important features of a complete blood count is to classify the white blood cells (WBC) into five different categories as the imbalance of the number of cells in one or more category helps in disease diagnosis.
• WBC white blood cells
• Automatic classification of the WBC can be performed by applying advanced image analysis techniques on the cell images acquired using DHM. The quality of the cell image is crucial and would affect the accuracy of the classification.
• Off-axis holographic microcopy system creates holograms where there is a modulating pattern over the entire field of view due to a small angle between object and reference beam.
• the reference beam is created from the object beam using two lenses and a pinhole to erase the object spatial signature and to provide a uniform plane waves for creating an interference or hologram image.
• the focal length would greatly affect the quality of the acquired cell images.
• the distance between the focal plane and the object impacts the appearance of the phase images and their quality.
• FIG. 2 illustrates example cell images with different quality. In the top row, the cells are in focus and the information content of the image can be used to discriminate among the different cell types.
• the images in the bottom row are of cells that are out of focus and distorted. The image quality is very poor and cannot be used in a diagnosis workflow.
• Embodiments of the present invention address and overcome one or more of the above shortcomings and drawbacks, by providing methods, systems, and apparatuses related to identifying the quality of the cell images acquired with a microscopy device using a convolutional neural network (CNN).
• CNN convolutional neural network
• a CNN is trained to determine whether cells are in focus or out of focus in an acquired image.
• instructions may be provided to the microscopy device to adjust the focal length so as to bring the acquired images into focus.
• a computer-implemented method for detecting out of focus microscopy images includes acquiring microscopy images depicting cells, and extracting one or more sets of pixels from the microscopy images. Each set of pixels corresponds to an independent cell. One of a plurality of image quality labels are assigned to each set of pixels indicating the degree to which the independent cell is in focus. A classifier is trained to classify the set of pixels into the image quality labels. The classifier is configured according to a multi-layer architecture and the training results in determination of weights for connecting layers in the multi-layer architecture. A deployment of the classifier is created based on the multi-layer architecture, the weights, and the image quality labels.
• a computer-implemented method for performing adaptive focusing of a microscopy device includes acquiring a plurality of microscopy images depicting cells using a microscopy device, and extracting one or more sets of pixels from the microscopy images. Each set of pixels corresponds to an independent cell. Then, a trained classifier is used to assign one of a plurality of image quality labels to each set of pixels indicating the degree to which the independent cell is in focus. If the image quality labels corresponding to the sets of pixels indicate that the cells are out of focus, a focal length adjustment for adjusting focus of the microscopy device is determined using a trained machine learning model. Then, executable instructions are sent to the microscopy device to perform the focal length adjustment.
• a system for performing adaptive focusing of a microscopy device comprises a microscopy device configured to acquire microscopy images depicting cells and one or more processors executing instructions for performing a method that includes extracting pixels from the microscopy images.
• Each set of pixels corresponds to an independent cell.
• a trained classifier is used to assign one of a plurality of image quality labels to each set of pixels indicating the degree to which the independent cell is in focus. If the image quality labels corresponding to the sets of pixels indicate that the cells are out of focus, a focal length adjustment for adjusting focus of the microscopy device is determined using a trained machine learning model. Then, executable instructions are sent to the microscopy device to perform the focal length adjustment.
• FIG. 1 shows an example DHM setup
• FIG. 2 illustrates example cell images with different quality
• FIG. 3 shows the system diagram illustrating the actors and operations used to assess image quality, according to some embodiments
• FIG. 4 illustrates a technique for extracting the cells, according to some embodiments
• FIG. 5A illustrates a first example extraction of cells
• FIG. 5B illustrates a second example extraction of cells
• FIG. 6A illustrates an example of the multi-layer architecture that may be employed by the CNN, according to some embodiments
• FIG. 6B provides an alternate view of the multi-layer architecture shown in FIG. 6A;
• FIG. 7 shows an assessment of classification accuracy of the techniques described herein, according to one example implementation
• FIG. 8 provides an example deployment of a trained CNN, according to some embodiments.
• FIG. 9 provides an example of a parallel processing memory architecture that may be utilized by image processing system, according to some embodiments of the present invention.
• the following disclosure describes the present invention according to several embodiments directed at methods, systems, and apparatuses related to identifying the quality of the cell images acquired with digital holographic microscopy (DHM) or another type of microscopy device using convolutional neural networks (CNNs). More specifically, techniques are described herein for differentiation between "good quality" cell images where the cells are captured in focus and the "poor quality” images that are out of focus. In some embodiments, the problem is formulated as a binary image classification problem where the two classes are in-focus/out-of-focus. This problem is then solved using a CNN. As explained in further detail below, this general framework can be expanded upon with various enhancements, refinements, and other modifications in different embodiments of the present invention.
• FIG. 3 shows the system diagram illustrating the actors and operations used to assess image quality, according to some embodiments.
• a Microscopy Device 305 is used to acquire one or more Microscopy Images 310.
• the Microscopy Device 305 may be any system known in the art capable of acquiring microscopy images of cells.
• the Microscopy Images 310 may be acquired using off-axis digital holographic microscope (DHM). The acquisition can alternatively be done using other DHM techniques such as on axis configurations.
• the Microscopy Device 305 uses other cell imaging techniques known in the art which can be used to acquire the
• Microscopy Images 310 include, without limitation, bright field microscopy, dark field microscopy, differential interference contrast, fluorescence microscopy, confocal microscopy, two-photon excitation microscopy, and multiphoton microscopy.
• GANs Generative adversarial networks
• GANs generally represent a class of artificial intelligence algorithms that falls under the category of "unsupervised learning.”
• GANs are a combination of two neural networks: one network is learning how to generate examples (e.g., synthetic DHM images) from a training data set (e.g., images acquired using Microscopy Device 305), and another network attempts to distinguish between the generated examples and the training data set.
• the training process is successful if the generative network produces examples which converge with the actual data such that the discrimination network cannot consistently distinguish between the two.
• the Microscopy Images 310 are received by an Image Processing System 345 that has processing resources for training a CNN 330 based using the Microscopy Images 310.
• a Preprocessing Module 315 extracts the independent cells from the Microscopy Images 310 for training. Each cell is extracted as a Set of Pixels 320.
• each individual Set of Pixels 320 is also sometimes referred to as a "cell image.”
• various types of image classification models can be used as an alternative to CNNs in other embodiments including, without limitation, linear classifiers (e.g., logistic regression, na ' ive bayes classifiers, etc.), kernel estimation k-means clustering, nearest neighbor classification, support vector machines, decision trees, boosted trees, random forests, and different configurations of neural networks.
• FIG. 3 shows the system diagram illustrating the actors and operations used to assess image quality, according to some embodiments.
• FIG. 4 illustrates a technique 400 for extracting the cells, according to some embodiments.
• the acquired images have a dimension of 384x512 and 100 images are acquired per second.
• the average of the first 100 images is computed at step 405 and the average image is subtracted from each acquired image at step 410.
• adaptive thresholding is applied to capture all the bright components in the image.
• Various adaptive thresholding techniques generally known in the art may be applied at step 415. For example, in one embodiment, the intensity values of the local neighborhood of each pixel are used to determine a threshold for the pixel.
• the pixel is below the threshold value, it is set to some minimum value (e.g., pure black).
• one or more exclusion criteria are applied to the images at step 420 to illuminate non-cell components. Examples of exclusion criteria that may be applied at step 420 include, without limitation, width, height, aspect ratio and circularity of the object.
• connected components analysis is applied at step 425 to identify the sets of pixels corresponding to independent cells. Each set of pixels is cropped using a predefined crop box of size 51 x51.
• FIGS. 5A and 5B illustrate the extraction of the different cells.
• the Preprocessing Module 315 can use detection techniques such as probabilistic boosting trees, deep convolutional neural networks to detect the location of the cell.
• Cell segmentation can also be used to extract the cell. This can be performed using energy minimization techniques such as graph cuts, watershed, random walker, or Mumford-Shah. It can also be performed using model based methods that would fit a predefined shape (e.g., a circle) to the desired object.
• the segmentation can be performed with alternative techniques such as edge matching, gradient matching or intensity matching. Additional details on how segmentation may be performed are detailed in U.S. Patent Application Publication No. 2018/0144182A1 entitled "Analyzing digital holographic microscopy data for hematology applications," the entirety of which is incorporated herein by reference.
• the Image Processing System 345 uses the Sets of Pixels 320 to train the CNN 330 to solve the image classification problem.
• the classification problem aims at assigning a label for each image.
• a User 350 manually inspects each Set of Pixels 320 corresponding to a cell to generate Image Quality Labels 325.
• the User 350 may be provided with a graphical user interface (GUI) where each cell is presented and with an input field that allows specification of the Image Quality Labels 325.
• GUI graphical user interface
• the task of generating the Image Quality Labels 325 can be distributed over a plurality of users.
• the Image Quality Labels 325 are 0, for a cell image that is out of focus and 1, for a cell image that is in focus.
• a wider range of labels are given for different focal plane images and this would capture a larger range of variation in the image.
• the label can be a grade for the cell from 1 to 10 where cells with grade 1 are the worst and cells with grade 10 are the best.
• Correlation between these grades and the focal distance can be used to automatically adjust the focal plane or provide feedback to the device operator to perform such adjustment.
• cells belonging to one or more of these grade classes can be included.
• a CNN 330 includes an input layer, one or more hidden layers, and an output layer.
• the objective of training the CNN 330 is to learn a transfer function between the input layer (features that represent the image) and the output layer (the labels for the image).
• the Image Processing System 345 performs iterative forward and backward passes that are made through the CNN 330 as the transfer function is minimized with respect to Weights 335 connecting the different layers of the CNN
• the NVIDIA TensorRT® is used to deploy the CNN 330 into a production environment.
• TensorRT requires 3 files to execute a CNN: a network architecture file, trained weights, and a label file to provide a name for each output class. These 3 files may be generated by the description of the Multi-Layer Architecture 340, Weights 335, and the description of the labelling system, respectively.
• FIG. 6A illustrates an example of the Multi-Layer Architecture 340 that may be employed by the CNN 330, according to some embodiments.
• the first layer is convolutional layer with 20 filters of kernel size 5> ⁇ 5, while the second layer max pooling layer of kernel size 2x2 and stride of 2.
• the third layer is a convolutional layer with 50 filters of kernel size 5 ⁇ 5 and the fourth layer is a convolutional layer of kernel size 2x2 and stride of 2.
• the fifth layer is a fully connected layer of size 500.
• the sixth layer is a drop out layer keeping 0.7 of the weights.
• the seventh layer is an output layer of size 2 that provides a label 0 for out of focus cells, and a label of 1 for in focus cells.
• FIG. 6B provides a visualization of the operations performed by the Multi-Layer Architecture 340 described in FIG. 6A.
• the network architecture shown in FIGS. 6 A and 6B is just an example of a possible architecture.
• the kernel sizes of the filters can also be different and different sizes could be used for different layers.
• the dropout layer can be eliminated, or alternatively, multiple dropout layers may be used and the drop out probabilities can vary from 0 to 1.
• the trained CNN is used to predict the output label based on the image features computed from the input image.
• FIG. 8 provides an example deployment, according to some embodiments.
• the output of the CNN is used to provide adaptive focusing of the microscope of the Microscopy Device 805. More specifically, the Microscopy Device 805 is used to acquire Microscopy Images 810 which are sent to an Image Processing System 850. At the Image Processing System 850, the
• Microscopy Images 810 are preprocessed by Preprocessing Module 815 to generate Sets of Pixels 820 corresponding to cells.
• the Preprocessing Module 815 shown in FIG. 8 operates in a manner similar to the Preprocessing Module 315 described above with respect to FIG. 3.
• the Sets of Pixels 820 are used as input to the Trained CNN 830 that, in turn, output Labelled Cells 831. That is, the output are the Sets of Pixels 820 with labels according to the labelling convention (e.g., 0 for out of focus or 1 for in focus).
• the Labelled Cells 831 are used as input to a Machine Learning Model 833 trained to output a Focal Length Adjustment 835 for the Microscopy Device 805 to adjust any focus issues.
• This Machine Learning Model 833 trained by monitoring, over time, how the Microscopy Device 805 is adjusted in response to the acquired microscopy images and the output of the Trained CNN 830. Such monitoring may be performed, for example, by recording instructions sent to the Microscopy Device 805. Alternatively, an operator can manually enter the focal length changes into the Image Processing System 850.
• a manifold i.e., a basis set
• Example techniques that can be employed to learn the manifold include, without limitation, principal component analysis (PCA), locally-linear embedding, and diffusion maps.
• the Machine Learning Model 833 outputs a Focal Length Adjustment 835 for the Microscopy Device 805.
• This Focal Length Adjustment 835 is then used as input to an Instruction Generator 840 that translates the adjustment into Executable Instructions 845 for the Microscopy Device 805.
• the implementation of the Instruction Generator 840 is dependent on the interface of the Microscopy Device 805. However, in general, the
• Instruction Generator 840 can be understood as software that provides an additional interface layer between the Image Processing System 850 and the Microscopy Device 805.
• the Machine Learning Model 833 can be trained to directly output the
• FIG. 9 provides an example of a parallel processing memory architecture 900 that may be utilized by an image processing system, according to some embodiments of the present invention.
• This architecture 900 may be used in embodiments of the present invention where NVIDIATM CUDA (or a similar parallel computing platform) is used.
• the architecture includes a host computing unit (“host") 905 and a GPU device (“device”) 910 connected via a bus 915 (e.g., a PCIe bus).
• the host 905 includes the central processing unit, or "CPU” (not shown in FIG. 9) and host memory 925 accessible to the CPU.
• the device 910 includes the graphics processing unit (GPU) and its associated memory 920, referred to herein as device memory.
• the device memory 920 may include various types of memory, each optimized for different memory usages. For example, in some embodiments, the device memory includes global memory, constant memory, and texture memory.
• Parallel portions of a CNN may be executed on the architecture 900 as "device kernels" or simply “kernels.”
• a kernel comprises parameterized code configured to perform a particular function.
• the parallel computing platform is configured to execute these kernels in an optimal manner across the architecture 900 based on parameters, settings, and other selections provided by the user. Additionally, in some embodiments, the parallel computing platform may include additional functionality to allow for automatic processing of kernels in an optimal manner with minimal input provided by the user.
• the processing required for each kernel is performed by grid of thread blocks (described in greater detail below).
• the architecture 900 of FIG. 9 may be used to parallelize training of the CNN. For example, in some embodiments, processing of individual cell images may be performed in parallel.
• the device 910 includes one or more thread blocks 930 which represent the computation unit of the device 910.
• the term thread block refers to a group of threads that can cooperate via shared memory and synchronize their execution to coordinate memory accesses.
• threads 940, 945 and 950 operate in thread block 930 and access shared memory 935.
• thread blocks may be organized in a grid structure. A computation or series of computations may then be mapped onto this grid. For example, in embodiments utilizing CUD A, computations may be mapped on one-, two-, or three-dimensional grids. Each grid contains multiple thread blocks, and each thread block contains multiple threads. For example, in FIG.
• the thread blocks 930 are organized in a two dimensional grid structure with m+1 rows and n+1 columns.
• threads in different thread blocks of the same grid cannot communicate or synchronize with each other.
• thread blocks in the same grid can run on the same multiprocessor within the GPU at the same time.
• the number of threads in each thread block may be limited by hardware or software constraints.
• processing of subsets of the training data or operations performed by the algorithms discussed herein may be partitioned over thread blocks automatically by the parallel computing platform software.
• the individual thread blocks can be selected and configured to optimize training of the CNN.
• each thread block is assigned an individual cell image or group of related cell images.
• registers 955, 960, and 965 represent the fast memory available to thread block 930. Each register is only accessible by a single thread. Thus, for example, register 955 may only be accessed by thread 940. Conversely, shared memory is allocated per thread block, so all threads in the block have access to the same shared memory. Thus, shared memory 935 is designed to be accessed, in parallel, by each thread 940, 945, and 950 in thread block 930. Threads can access data in shared memory 935 loaded from device memory 920 by other threads within the same thread block (e.g., thread block 930). The device memory 920 is accessed by all blocks of the grid and may be implemented using, for example, Dynamic Random- Access Memory (DRAM).
• DRAM Dynamic Random- Access Memory
• Each thread can have one or more levels of memory access.
• each thread may have three levels of memory access.
• Second, each thread 940, 945, 950 in thread block 930 may read and write data to the shared memory 935 corresponding to that block 930.
• the time required for a thread to access shared memory exceeds that of register access due to the need to synchronize access among all the threads in the thread block.
• the shared memory is typically located close to the multiprocessor executing the threads.
• the third level of memory access allows all threads on the device 910 to read and/or write to the device memory.
• Device memory requires the longest time to access because access must be synchronized across the thread blocks operating on the device.
• the processing of each cell image is coded such that it primarily utilizes registers and shared memory and only utilizes device memory as necessary to move data in and out of a thread block.
• the embodiments of the present disclosure may be implemented with any combination of hardware and software. For example, aside from parallel processing architecture presented in FIG. 9, standard computing platforms (e.g., servers, desktop computer, etc.) may be specially configured to perform the techniques discussed herein.
• the embodiments of the present disclosure may be included in an article of manufacture (e.g., one or more computer program products) having, for example, computer- readable, non-transitory media.
• the media may have embodied therein computer readable program code for providing and facilitating the mechanisms of the embodiments of the present disclosure.
• the article of manufacture can be included as part of a computer system or sold separately.
• An executable application comprises code or machine readable instructions for conditioning the processor to implement predetermined functions, such as those of an operating system, a context data acquisition system or other information processing system, for example, in response to user command or input.
• An executable procedure is a segment of code or machine readable instruction, sub-routine, or other distinct section of code or portion of an executable application for performing one or more particular processes. These processes may include receiving input data and/or parameters, performing operations on received input data and/or performing functions in response to received input parameters, and providing resulting output data and/or parameters.
• a graphical user interface comprises one or more display images, generated by a display processor and enabling user interaction with a processor or other device and associated data acquisition and processing functions.
• the GUI also includes an executable procedure or executable application.
• the executable procedure or executable application conditions the display processor to generate signals representing the GUI display images. These signals are supplied to a display device which displays the image for viewing by the user.
• the processor under control of an executable procedure or executable application, manipulates the GUI display images in response to signals received from the input devices. In this way, the user may interact with the display image using the input devices, enabling user interaction with the processor or other device.
• module can refer to either or both of: (i) a software component that causes an electronic device to accept various inputs and generate certain outputs; or (ii) an electronic input/output interface, such as a panel, frame, textbox, window or other portion of a GUI.
• the functions and process steps herein may be performed automatically or wholly or partially in response to user command.
• An activity (including a step) performed automatically is performed in response to one or more executable instructions or device operation without user direct initiation of the activity.

Landscapes

• Engineering & Computer Science (AREA)
• Physics & Mathematics (AREA)
• General Physics & Mathematics (AREA)
• Theoretical Computer Science (AREA)
• Multimedia (AREA)
• Computer Vision & Pattern Recognition (AREA)
• General Health & Medical Sciences (AREA)
• Health & Medical Sciences (AREA)
• Evolutionary Computation (AREA)
• Artificial Intelligence (AREA)
• Medical Informatics (AREA)
• Quality & Reliability (AREA)
• Life Sciences & Earth Sciences (AREA)
• Software Systems (AREA)
• Databases & Information Systems (AREA)
• Data Mining & Analysis (AREA)
• Computing Systems (AREA)
• Chemical & Material Sciences (AREA)
• Optics & Photonics (AREA)
• Analytical Chemistry (AREA)
• Biomedical Technology (AREA)
• Molecular Biology (AREA)
• Evolutionary Biology (AREA)
• General Engineering & Computer Science (AREA)
• Bioinformatics & Computational Biology (AREA)
• Bioinformatics & Cheminformatics (AREA)
• Radiology & Medical Imaging (AREA)
• Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
• Image Analysis (AREA)
• Microscoopes, Condenser (AREA)

Priority Applications (5)

Applications Claiming Priority (2)

Related Child Applications (2)

Publications (1)

Family

ID=63077840

Family Applications (1)

Country Status (5)

Cited By (7)

Families Citing this family (22)

Citations (1)

Family Cites Families (35)

• 2018
  • 2018-07-06 US US16/617,162 patent/US11227386B2/en active Active
  • 2018-07-06 EP EP18748860.6A patent/EP3669227B8/en active Active
  • 2018-07-06 JP JP2020508555A patent/JP7072049B2/ja active Active
  • 2018-07-06 WO PCT/EP2018/068345 patent/WO2019034328A1/en not_active Ceased
  • 2018-07-06 CN CN201880052840.9A patent/CN111051955B/zh active Active
• 2021
  • 2021-12-06 US US17/457,716 patent/US11580640B2/en active Active

Patent Citations (1)

Non-Patent Citations (4)

Also Published As

Similar Documents

Legal Events