Computer image analysis of comparative genomic hybridization.

We describe and evaluate the image-processing and analysis techniques we have developed for the quantitative analysis of comparative genomic hybridization (CGH; Science 258:818, 1992). In a typical CGH application, two genomic DNA samples are simultaneously hybridized to metaphase chromosomes and detected with different fluorochromes. The primary data in CGH are contained in the intensity ratios of the fluorochromes as a function of position on the chromosomes, which reflect variation in DNA copy number ratio between the two DNA samples. Analysis involves chromosome segmentation, intensity normalization, background corrections, and calculation of the fluorescence intensity profiles and the ratio profile along the chromosome's length. Profiles from several copies of the same chromosome in different metaphases are averaged to reduce the noise. Confidence intervals are calculated and displayed for the mean profiles. The techniques were evaluated by examining the variability found in comparisons of two normal genomic DNAs, where the ratio was expected to be constant, and by measuring the ratios obtained for cell lines with cytogenetically documented copy number changes involving several chromosomal segments. The limits of sensitivity of CGH analysis were investigated by simulation. Guidelines for the interpretation of CGH data and indications of areas for future development of the analytical techniques are also presented.

A collaborative trial of a semi-automatic system for slide preparation and screening in cervical cytopathology.

We report a test of an experimental system for machine-aided screening in cervical cytology, comprising the 'CYTOPRESS' semi-automatic slide preparation system (Nijmegen) and the 'CERVIFIP' interactive scanner (Edinburgh). Material from women attending clinics in Edinburgh and Nijmegen was stratified according to the severity of the conventional laboratory diagnosis and selected randomly within strata for inclusion in the test. Monolayered slides were prepared by CYTOPRESS from cervical scrape material remaining after preparation of conventional smears and scanned by CERVIFIP to determine the positions of the most 'suspicious' objects. The test was based on a set of 701 monolayers, divided equally between 'negatives' and 'abnormals' of various grades, of which 585 (83.4%) were passed automatically as adequate for machine-aided analysis. Approximately 15% of adequate slides were passed as 'negative' without operator interaction. In the remaining 85%, the suspicious objects were inspected by a human operator and a decision was then made either to refer each monolayer for conventional microscopic analysis, or to pass it as 'negative'. Where discrepancies occurred between the conventional laboratory and the system results, a consensus diagnosis was reached by taking into account all relevant information including clinical data. Of those with a consensus diagnosis of CIN 3 or worse an estimated 9.3 +/- 4.1% were passed by the system as 'negative'. Closer investigation of these false-negatives revealed that most, and perhaps all, were preventable by system improvements either planned or in progress. Corresponding false-negative rates for those graded 'CIN 1 or 2' and 'negative-early recall' were estimated, respectively as 18.9 +/- 5.3% and 22.9 +/- 3.1%. Of those with a 'negative-routine recall' consensus, 19.4 +/- 2.5% were referred for conventional microscopic analysis, a level well within acceptable limits for cost-effectiveness. Women whose initial laboratory smears were negative, but whose consensus diagnosis was 'negative-early recall' or CIN, are being investigated further to determine whether cervical abnormalities were in fact present. Over two-thirds of this group were referred by the machine-aided system for conventional microscopic analysis.

Automatic scoring of sister chromatid exchanges by image analysis in a dose response experiment.

A system which automatically selects second division metaphases and then, automatically scores the number of SCEs of each cell is described. In an initial set of experiments, the performance of the components of the system was measured using a data set in which metaphases had been visually classified as either 2nd division or other; and in 2nd division metaphases, every SCE had been marked on a hard copy. SCE scoring had a true positive rate of about 75% and a false positive rate of about 1.5 false SCEs per metaphase analyzed. Second division detection had a true positive rate of 80% and a false positive rate of about 10% of the non-2nd division cells. Next, the overall system was compared to human visual scoring in a dose-response experiment by analyzing the effect of mitomycin C on human chromosomes scored visually by two observers and by the fully automatic scoring. Human visual scoring and machine analysis showed similar dose responses, but the variability between them was considerable.

Comparative genomic hybridization: a rapid new method for detecting and mapping DNA amplification in tumors.

Recent evidence indicates that many more genes than the currently known oncogenes may undergo amplification in tumors. We have developed a new technique, Comparative Genomic Hybridization (CGH), which allows rapid detection of DNA amplification anywhere in the tumor genome and maps the amplified sequences on normal chromosomes. CGH is based on a competitive in situ hybridization of differentially labeled tumor DNA and normal DNA to a normal human metaphase spread. Regions of gain of DNA sequences are seen as an increased color ratio of two fluorochromes used to detect the labeled DNAs. Over 20 different regions of amplification have been identified using CGH.

Semi-automated cervical smear pre-screening systems: an evaluation of the Cytoscan-110.

This paper reports a test of a system for provision of machine assistance in cervical cytology screening. The hypothesis tested was that if the results of examination by a screener of a small number of high-ploidy cells on specially prepared monolayers, automatically selected and presented by the system, were combined with machine measurement of cell and cell population characteristics, it would be possible to distinguish conditions requiring further action on the part of a cytology service from those in which the patient could safely be signed out. The system appeared broadly capable of this discrimination, with a false-negative error not significantly different (for the numbers tested) on CIN1 and more severe cases to that obtaining for routine screening of the parallel PAP smears, and also to results obtained by a panel of three observers. The machine system appeared to do better than other systems in selecting borderline cases for review, but this may have been an artefact of the method of evaluation used: all results were compared with a 'reference diagnosis', which was computed using statistical techniques to integrate diagnostic information from all available sources. The false-negative error-rate of the system amounted to 5% of high-grade cases, 17% of CIN1's and 29% of borderlines, but were not substantially different from the FN rates for other reporting systems on the same material. The proportion of negative cases referred back for full cytological diagnosis was 34%. Despite this high false-positive rate, the system is potentially cost-effective in use.

Comparative genomic hybridization for molecular cytogenetic analysis of solid tumors.

Comparative genomic hybridization produces a map of DNA sequence copy number as a function of chromosomal location throughout the entire genome. Differentially labeled test DNA and normal reference DNA are hybridized simultaneously to normal chromosome spreads. The hybridization is detected with two different fluorochromes. Regions of gain or loss of DNA sequences, such as deletions, duplications, or amplifications, are seen as changes in the ratio of the intensities of the two fluorochromes along the target chromosomes. Analysis of tumor cell lines and primary bladder tumors identified 16 different regions of amplification, many in loci not previously known to be amplified.

Radiation dosimetry by automatic image analysis of dicentric chromosomes.

A system for scoring dicentric chromosomes by image analysis comprised fully automatic location of mitotic cells, automatic retrieval, focus and digitization at high resolution, automatic rejection of nuclei and debris and detection and segmentation of chromosome clusters, automatic centromere location, and subsequent rapid interactive visual review of potential dicentric chromosomes to confirm positives and reject false positives. A calibration set of about 15,000 cells was used to establish the quadratic dose response for 60Co gamma-irradiation. The dose-response function parameters were established by a maximum likelihood technique, and confidence limits in the dose response and in the corresponding inverse curve, of estimated dose for observed dicentric frequency, were established by Monte Carlo techniques. The system was validated in a blind trial by analysing a test set comprising a total of about 8000 cells irradiated to 1 of 10 dose levels, and estimating the doses from the observed dicentric frequency. There was a close correspondence between the estimated and true doses. The overall sensitivity of the system in terms of the proportion of the total population of dicentrics present in the cells analysed that were detected by the system was measured to be about 40%. This implies that about 2.5 times more cells must be analysed by machine than by visual analysis. Taking this factor into account, the measured review time and false positive rates imply that analysis by the system of sufficient cells to provide the equivalent of a visual analysis of 500 cells would require about 1 h for operator review.

Multiprocessing interval processor for automated cytogenetics.

In this paper we describe the genesis, use, and possible future of a device specialized for high-speed searching of microscope slides, but which also has considerable power of analysis of individual fields.

The use of cationic polyelectrolytes in the preparation of cell monolayers for automated cell scanning and diagnostic cytopathology.

The introduction of cationic polyelectrolytes as cellular adherents has significantly advanced the preparation of cervical scrape specimens for automated cell scanning and has also provided an efficient technique for the preparation of cell monolayers of other cytologic specimens, e.g., breast cyst fluids, urines and serous effusions, for diagnostic cytopathology. Variable thickness of cell preparation and cell overlap have both been resolved by laying cells onto glass slides coated with the cationic polyelectrolyte poly-L-lysine. We have determined the optimal conditions for pH, molecular weight, concentration and temperature for the application of poly-L-lysine as a cell-to-slide adhesive.

An efficient multiple-cell approach to automatic aneuploidy screening.

We describe a statistical method of discriminating efficiently, on the basis of multiple-cell measurements without operator interaction, between chromosomally normal human cell lines and those either containing a single additional chromosome or missing one chromosome. We begin by defining hypothetical but realistic "confusion matrices," which give the probabilities of (1) assigning each chromosome to each of various possible groups and (2) rejecting it as unclassifiable. From these, false-positive and false-negative rates of 0.01 and 0.001, respectively, are found to be attainable by processing 16 to 32 cells if the average probability of misclassifying or rejecting individual chromosomes is 5% to 9% for "Denver" groups or 10% to 17% for homologous pairs. Since these values are probably within the reach of current technology, the method is a basis for a realistic, fully automatic screening system. We also show how the method can be extended to the detection of quite general types of chromosomally abnormal cell lines.

A fast interval processor (FIP) for cervical prescreening.

A cervical prescreening device (FIP: fast interval processor) designed to scan and classify a slide-mounted specimen within two minutes is described. The image analysis techniques are based directly on the MRC Cerviscan equipment with the minimal conversion needed to adapt these techniques for interval processing. A high scanning rate is achieved by scanning with a charge-coupled diode linear image sensor along one axis and by stepping the microscope stage continuously along the other axis. High processing rates are achieved using an asynchronous pipeline approach. Operations on pixels are carried out by parallel dedicated hardware units, while operations on intervals (segments) are carrie out using a dual microprocessing configuration. A determined attempt has been made to minimize the cost of the components required. Preliminary results showing scanning and processing performance are given.

Efficient interaction for automated chromosome analysis using asynchronous parallel processes.

It is likely that any practical automated chromosome analysis system will be interactive. To prevent long pauses in the stream of operator interactions, it is necessary, if using standard computer hardware, to configure for asynchronous and parallel operation. A system is presented which uses several computer processors, which can support one or more operators, and which divides processing into interactive and noninteractive sections, smoothes the rate of presentation of interactions, and keeps both the operator and the computer fully employed.

A cytogeneticist's microscope.

It is demonstrated that there are a number of advantages in using a mechanised microscope for scoring a large number of metaphase cells from human blood lymphocyte preparations. Following the development of an automatic metaphase spread finding machine based upon a large motorised microscope and a synchronous closed circuit television camera and flashing light source, a much smaller machine which is more appropriate to the cytogenetics laboratory, but with a similar metaphase finding performance has been constructed. The new machine which consists of a Cambridge Instruments 1 micron stepping microscope stage, a linear diode array scanner and a computer is described in detail. Metaphase finding performance figures for various orcein stained human blood lymphocyte preparations are given.