Automatic microscopic cell counting by use of unsupervised adversarial domain adaptation and supervised density regression

This paper proposes a novel approach for semi-supervised domain adaptation for holistic regression tasks, where a DNN predicts a continuous value y∈R given an input image x. The current literature generally lacks specific domain adaptation approaches for this task, as most of them mostly focus on classification. In the context of holistic regression, most of the real-world datasets not only exhibit a covariate (or domain) shift, but also a label gap—the target dataset may contain labels not included in the source dataset (and vice versa). We propose an approach tackling both covariate and label gap in a unified training framework. Specifically, a Generative Adversarial Network (GAN) is used to reduce covariate shift, and label gap is mitigated via label normalisation. To avoid overfitting, we propose a stopping criterion that simultaneously takes advantage of the Maximum Mean Discrepancy and the GAN Global Optimality condition. To restore the original label range—that was previously normalised—a handful of annotated images from the target domain are used. Our experimental results, run on 3 different datasets, demonstrate that our approach drastically outperforms the state-of-the-art across the board. Specifically, for the cell counting problem, the mean squared error (MSE) is reduced from 759 to 5.62; in the case of the pedestrian dataset, our approach lowered the MSE from 131 to 1.47. For the last experimental setup, we borrowed a task from plant biology, i.e., counting the number of leaves in a plant, and we ran two series of experiments, showing the MSE is reduced from 2.36 to 0.88 (intra-species), and from 1.48 to 0.6 (inter-species).

Download Full-text

Algorithms for automated montage synthesis of images from laser-scanning confocal microscopes

Proceedings, annual meeting, Electron Microscopy Society of America ◽

10.1017/s0424820100139627 ◽

1995 ◽

Vol 53 ◽

pp. 650-651

Author(s):

D. E. Becker

Keyword(s):

Image Analysis ◽

High Resolution ◽

Laser Scanning ◽

Cell Counting ◽

Wide Area ◽

Data Set ◽

Computational Synthesis ◽

Automated Cell Counting ◽

Partial View ◽

The Individual

An efficient, robust, and widely-applicable technique is presented for computational synthesis of high-resolution, wide-area images of a specimen from a series of overlapping partial views. This technique can also be used to combine the results of various forms of image analysis, such as segmentation, automated cell counting, deblurring, and neuron tracing, to generate representations that are equivalent to processing the large wide-area image, rather than the individual partial views. This can be a first step towards quantitation of the higher-level tissue architecture. The computational approach overcomes mechanical limitations, such as hysterisis and backlash, of microscope stages. It also automates a procedure that is currently done manually. One application is the high-resolution visualization and/or quantitation of large batches of specimens that are much wider than the field of view of the microscope.The automated montage synthesis begins by computing a concise set of landmark points for each partial view. The type of landmarks used can vary greatly depending on the images of interest. In many cases, image analysis performed on each data set can provide useful landmarks. Even when no such “natural” landmarks are available, image processing can often provide useful landmarks.

Download Full-text

Automated cell counting of astrocytes on patterned substrates containing aliphatic and charged properties

Proceedings, annual meeting, Electron Microscopy Society of America ◽

10.1017/s042482010014124x ◽

1995 ◽

Vol 53 ◽

pp. 974-975

Author(s):

W. Shain ◽

H. Ancin ◽

H.C. Craighead ◽

M. Isaacson ◽

L. Kam ◽

...

Keyword(s):

Cell Culture ◽

Cell Attachment ◽

Culture Method ◽

Cell Counting ◽

Data Sets ◽

Nuclear Staining ◽

Double Positive ◽

A Cell ◽

Wafer Test ◽

Cell Densities

Neural protheses have potential to restore nervous system functions lost by trauma or disease. Nanofabrication extends this approach to implants for stimulating and recording from single or small groups of neurons in the spinal cord and brain; however, tissue compatibility is a major limitation to their practical application. We are using a cell culture method for quantitatively measuring cell attachment to surfaces designed for nanofabricated neural prostheses.Silicon wafer test surfaces composed of 50-μm bars separated by aliphatic regions were fabricated using methods similar to a procedure described by Kleinfeld et al. Test surfaces contained either a single or double positive charge/residue. Cyanine dyes (diIC18(3)) stained the background and cell membranes (Fig 1); however, identification of individual cells at higher densities was difficult (Fig 2). Nuclear staining with acriflavine allowed discrimination of individual cells and permitted automated counting of nuclei using 3-D data sets from the confocal microscope (Fig 3). For cell attachment assays, LRM5 5 astroglial cells and astrocytes in primary cell culture were plated at increasing cell densities on test substrates, incubated for 24 hr, fixed, stained, mounted on coverslips, and imaged with a 10x objective.

Download Full-text

Going Beyond 3-D Imaging: Automated 3-D Montaged image Analysis of Cytological Specimens

Proceedings, annual meeting, Electron Microscopy Society of America ◽

10.1017/s0424820100163873 ◽

1996 ◽

Vol 54 ◽

pp. 282-283

Author(s):

Badrinath Roysam ◽

Hakan Ancin ◽

Douglas E. Becker ◽

Robert W. Mackin ◽

Matthew M. Chestnut ◽

...

Keyword(s):

Image Analysis ◽

Structural Changes ◽

Sampling Methods ◽

Spatial Clustering ◽

Three Dimensional ◽

Field Of View ◽

Cell Counting ◽

Depth Of Field ◽

Tissue Cell ◽

Tissue Organization

This paper summarizes recent advances made by this group in the automated three-dimensional (3-D) image analysis of cytological specimens that are much thicker than the depth of field, and much wider than the field of view of the microscope. The imaging of thick samples is motivated by the need to sample large volumes of tissue rapidly, make more accurate measurements than possible with 2-D sampling, and also to perform analysis in a manner that preserves the relative locations and 3-D structures of the cells. The motivation to study specimens much wider than the field of view arises when measurements and insights at the tissue, rather than the cell level are needed.The term “analysis” indicates a activities ranging from cell counting, neuron tracing, cell morphometry, measurement of tracers, through characterization of large populations of cells with regard to higher-level tissue organization by detecting patterns such as 3-D spatial clustering, the presence of subpopulations, and their relationships to each other. Of even more interest are changes in these parameters as a function of development, and as a reaction to external stimuli. There is a widespread need to measure structural changes in tissue caused by toxins, physiologic states, biochemicals, aging, development, and electrochemical or physical stimuli. These agents could affect the number of cells per unit volume of tissue, cell volume and shape, and cause structural changes in individual cells, inter-connections, or subtle changes in higher-level tissue architecture. It is important to process large intact volumes of tissue to achieve adequate sampling and sensitivity to subtle changes. It is desirable to perform such studies rapidly, with utmost automation, and at minimal cost. Automated 3-D image analysis methods offer unique advantages and opportunities, without making simplifying assumptions of tissue uniformity, unlike random sampling methods such as stereology.12 Although stereological methods are known to be statistically unbiased, they may not be statistically efficient. Another disadvantage of sampling methods is the lack of full visual confirmation - an attractive feature of image analysis based methods.

Download Full-text

Domain Adaptation for Visual Recognition

10.1561/9781680830316 ◽

2015 ◽

Author(s):

Raghuraman Gopalan ◽

Ruonan Li ◽

Vishal M. Patel ◽

Rama Chellappa

Keyword(s):

Visual Recognition ◽

Domain Adaptation

Download Full-text

Cost-Effects of Increased Automation in Blood Cell Counting Used for Centralized Regional Service

Scandinavian Journal of Clinical and Laboratory Investigation ◽

10.3109/00365518809168537 ◽

1988 ◽

Vol 48 ◽

pp. 129-129

Author(s):

J. Kekoni ◽

S. Bjorkqvist ◽

I. Penttila ◽

E. Kentta

Keyword(s):

Blood Cell ◽

Cell Counting ◽

Regional Service ◽

Blood Cell Counting

Download Full-text

Learning Hidden Markov Models with Distributed State Representations for Domain Adaptation

10.3115/v1/p15-2086 ◽

2015 ◽

Cited By ~ 1

Author(s):

Min Xiao ◽

Yuhong Guo

Keyword(s):

Hidden Markov Models ◽

Markov Models ◽

Domain Adaptation ◽

Hidden Markov

Download Full-text

Manufacturing In-Line Whole Wafer Focused Ion Beam (FIB) Memory Address Descramble Verification

ISTFA 2008: Conference Proceedings from the 34th International Symposium for Testing and Failure Analysis ◽

10.31399/asm.cp.istfa2008p0128 ◽

2008 ◽

Author(s):

Steven B. Herschbein ◽

Hyoung H. Kang ◽

Scott L. Jansen ◽

Andrew S. Dalton

Keyword(s):

Flip Chip ◽

Focused Ion Beam ◽

Ion Beam ◽

Random Access ◽

Cell Counting ◽

Test Results ◽

Laboratory Procedure ◽

Memory Array ◽

Dual Beam ◽

Process Level

Abstract Test engineers and failure analyst familiar with random access memory arrays have probably encountered the frustration of dealing with address descrambling. The resulting nonsequential internal bit cell counting scheme often means that the location of the failing cell under investigation is nowhere near where it is expected to be. A logical to physical algorithm for decoding the standard library block might have been provided with the design, but is it still correct now that the array has been halved and inverted to fit the available space in a new processor chip? Off-line labs have traditionally been tasked with array layout verification. In the past, hard and soft failures could be induced on the frontside of finished product, then bitmapped to see if the sites were in agreement. As density tightened, flip-chip FIB techniques to induce a pattern of hard fails on packaged devices came into practice. While the backside FIB edit method is effective, it is complex and expensive. The installation of an in-line Dual Beam FIB created new opportunities to move FA tasks out of the lab and into the FAB. Using a new edit procedure, selected wafers have an extensive pattern of defects 'written' directly into the memory array at an early process level. Bitmapping of the RAM blocks upon wafer completion is then used to verify correlation between the physical damaged cells and the logical sites called out in the test results. This early feedback in-line methodology has worked so well that it has almost entirely displaced the complex laboratory procedure of backside FIB memory array descramble verification.

Download Full-text