Image preprocessing for accelerating cytological image classification by full convolutional neural network

A convolutional network and cell technology, applied to biological neural network models, neural architectures, instruments, etc., can solve problems such as re-optimization difficulties, low computational efficiency, and destruction of inherent parallelism

Active Publication Date: 2019-04-26
HONG KONG APPLIED SCI & TECH RES INST
View PDF7 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, a significant disadvantage of using the skip method is that jumping from one ROI to another in the convolution computation breaks the inherent parallelism that exists in successively computing the convolution product over the input image
A dedicated processor with an optimized

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Image preprocessing for accelerating cytological image classification by full convolutional neural network
  • Image preprocessing for accelerating cytological image classification by full convolutional neural network
  • Image preprocessing for accelerating cytological image classification by full convolutional neural network

Examples

Experimental program
Comparison scheme
Effect test

Example Embodiment

[0062] As used herein, a test image means an image that is processed or intended to be processed by FCN for classification. In addition, in the specification and appended claims, it should be understood that "an image containing cells" means that the image contains sub-images of cells, rather than the image containing physical cells.

[0063] In the convolutional layer of FCN, the volume generated for the input image or the usual 2D data array is obtained by sliding the 2D filter on the input image and generating a single product sum each time the position of the filter on the input image stops. A sequence of products. The "stride" as commonly understood in the art is the number of pixels that the filter jumps from one position to the next immediately next position. For example, the stride value may be 2 or less, and the usual value is 3.

[0064] As will be shown soon, the test image is usually a color image with multiple color channels. In the convolution calculation, each col...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

A full convolutional network (FCN) implemented on a dedicated processor optimized for convolution calculations can achieve acceleration of cell classification. Further acceleration is achieved by compressing a test image of cells and processing the compressed test image using the FCN without re-optimizing the dedicated processor. The test image is first segmented into a background and regions of interest (ROI). The ROIs are packed more tightly by rearranging the ROIs under constraints without having to adjust their sizes, such that the pixel distance separated by any two adjacent rearranged ROIs is not less than the minimum distance determined according to a step value of the FCN convolutional layer. The geometric operations of the ROI rearrangement include repositioning the ROIs and optionally rotating the ROIs. The rearranged ROIs are surrounded by a boundary (usually a rectangular boundary) to form the compressed test image having an area smaller than the area of the test image.

Description

[0001] list of acronyms [0002] 2D two-dimensional [0003] ADC adenocarcinoma intracervical [0004] AGC atypical glandular cells [0005] AIS adenocarcinoma in situ [0006] ASC-H Atypical squamous cells - HSIL cannot be ruled out [0007] ASC-US atypical squamous cells of undetermined significance [0008] CNN convolutional neural network [0009] FCN Fully Convolutional Network [0010] GPU graphics processing unit [0011] HSIL high-grade squamous intraepithelial [0012] LSIL low-grade squamous intraepithelial [0013] RGB red, green, blue [0014] ROI region of interest [0015] SCC squamous cell carcinoma [0016] TBS Bethesda System [0017] WSI Whole Slide Image technical field [0018] The present invention relates to preprocessing test images to achieve speedup in cell classification performed by FCNs implemented on dedicated processors optimized for convolution computation of images. Background technique [0019] In automated cancer screening, a s...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
IPC IPC(8): G06K9/00G06K9/32G06K9/34G06N3/04
CPCG06V20/698G06V10/25G06V10/267G06N3/045
Inventor 胡羽王陆梁秉舜
Owner HONG KONG APPLIED SCI & TECH RES INST
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap