DE3612884A1 - Method for measuring the shape or cells of cell organelles - Google Patents

Abstract

The shape of a cell or of a cell organelle is measured in order to obtain information about the shape properties of an object or a population. The cell or cell organelle is measured by generating an image of the detailed edge formation of the cell or organelle, additional information being obtained about the height by observing under the reflected light of an interferometer which consists of a reflecting material and a surface layer which causes a constant interference increment. At least one geometric descriptor of the resolved image is measured and then compared with previously determined descriptors of individual objects or populations of cells. Cells or cell organelles which appear dissimilar are classified by image analysis and sample analysis methods.

Description

Method of measuring the shape of cells or

Cell organelles The invention relates to a method for measuring the Form of cells or cell organelles, especially for measuring changes in configuration and the growth of cells and cell organelles.

Cells containing a tissue cause fine control over them their own destiny. This regulating performance is varied by a variety Mechanisms causes. One of these is a mechanism whereby a modulation of the Shape of cells directly affects their growth rate.

The study of cell configuration and growth control shows that there is a close relationship between these parameters. For example in Rous sarcoma virus-transformed chick embryo cells (Temin et al. 6 Virology 669, 1958) and spontaneously transformed mouse embryo cells (Sanford et al., 39 J. Nat'l. Cancer Inst. 705, 1967) changes can be made either by a rough analysis or by a microscopic analysis can be determined.

In a rough analysis, virus-transformed focal points are identified the darker color of the rounded and multilayered cells was noted. It were differences between spontaneously transformed cells and their normal counterparts found in the area occupied on a substrate (Cherny et al., 90 Exp. Cell Res. 317, 1975) in cytoplasmic basophils in dry weight and im Ratio of nuclear and cytoplasmic areas (Fox et al., 36, Cander Res. 1556, 1976; Handleman et al., In Vitro 526, 1977).

However, due to features of the experiments mentioned above lead increased basophils and nuclear cytoplasmic ratios directly from a reduction into a lamellar spread. With the decrease in the extent of the spread of the peripheral cytoplasms are expected to increase in height.

The transformed cells have a thicker optical cross-section, hence a denser coloration occurs. Subject to changes in cell spread also the observed cytological changes in the nuclear-cytoplasm a ratio. Regardless of the quantitative extent of expansion, the core takes a shape which approximates more to a spherical shape than that of a cell as a whole, therefore, as the extent of the spread increases, changes become less drastic are.

It is not just different that contribute to changes in cell spread cytological terms altered growth control, but the latter Changes are almost invariably present in transformed cells. That prevalence A reduced lamellar spread in vitro can be confirmed by the number of reports can be estimated in which a "rounding up" has been described. A change in shape of the cells is one of the earliest manifestations of the "pleiotropic reaction" temperature-dependent transformations of chick embryo cells by Rous sarcomavirus. Rapid morphological changes in observed the cells after reaching a temperature suitable for src gene expression is permissible (Ambros et al., 72 Proc. Natl. Acad. Sci. USA 3144, 1975; Bader, 10 J. Virol. 207, 1972). These changes may be independent of the activity of the src g en-cAMP Protein kinase coupled (Collett, 75 Proc.Natl.

Acad. Sci. USA, 2021, 1978; Levinson et al., 15 Cell 561, 1978; Collett et al., 295 nature. 167, 1980). Although a single gene product for transformation in ts 68 is responsible (Vogt, Comprehensive Virology, Vol. 10, 341, 1977), the product is an enzyme that not only has a broad specificity for substrates, but that also changed in its subcellular distribution in transformed cells can be.

In contrast to the normal form of the src gene-cAMP-independent kinase the abnormal form was activated after entrapment in cell membranes (Courtneidge et al., 79 Proc. Natl.Acad.Sci.USA, 7117, 1982). The discovery of membrane participation shows a central event in the process of retrovirus-mediated transformation in the expression of the scr gene.

~ .of. . . . .. When considering the structure ofiellier anchoring dependent cells can switch the gene product into the membrane sites extracellular Adhesion affect the upoid bilayer as such and / or the internal Ztyoskeiett macromolecular components that deal with the internal aspect of the plasma membrane integrate. Regarding the main reorganization of the -like components in cells, transformed by the src gene will both become normal actinic Network as well as the micro-channel thread network disturbed. The lack of a technical However, being able to collect and analyze cytoskeletal data meant that Pertubations short for a complete structural reorganization difficult to are marked. Therefore, there is a need for an improved method for the determination and quantitative determination of the shape of cells.

Changes in the shape of a cell that occur during a transformation are found everywhere, and cell shape is an intrinsic aspect of the transformation process. Classification systems have already been developed for the assessment of the Form intended as a global characteristic of the cells should be. So far, people such as qualified cytological technicians have performed the function of the cytological diagnosis by imaging and assessing the changes in the cells in order to then make a distinction between the cell types. There is however, the disadvantage that human error in performing the collection of Information from samples of flaking cells may appear and that often insufficient detection of the differences in the cells is carried out can be. A lack of standardization in the implementation of such Analyzes by qualified technicians and laboratories also helps ensure that false negative diagnoses can result. Furthermore, the technicians are not in able to distinguish enough information about the changes in cell shape to be able to make a forecast. There is therefore a need improved methods of mapping and measuring the shape changes in cells too create, as well as variables for an analysis of the many geometrical aspects of cell shapes to develop.

Previous attempts to quantify the cell shape by light microscope method are based on absorbance or optical measurements Density (Sychra et al., 20 Acta Cytol. 68, 1976; Holmquist et al., 26 J.

Histochem. Cytochem. 1000, 1978; Sychra et al., 21 Acta Cytol. 765, 1978).

A method currently in use for collecting such information is the automatic digitization of absorbance measurements of stained cell samples. A major problem with such cell imaging techniques used by NASA is that the limit of accuracy is 90%, which is in terms of clinical requirements for a cancer diagnosis are unsatisfactory.

Attempts have been made to better visualize the To achieve cell shape, for example with the help of interferometry, one with Tantalum sprayed slide glass was anodized in such a way that a thin layer of Tantalum oxide is formed, which isolates the metallic tantalum (Heckman et al., 9 Tissue and Cell 317 1977). The slides were taken with an AO microscope or examined with a Leitz-Ortholux microscope.

A method of classifying cultured cells on the basis of of shape properties was developed in which optical information with high Resolving power using interferometry with an anodic oxide layer (Olson et al., 77 Proc. Nat'l Acad. Sci., 1516, 1980).

37 mathematical descriptors were measured to classify individual cultured cells according to the values of their descriptors. Included however, changes in the shape configuration or changes in growth patterns or regularities of growth in these cells were not investigated. Therefore, this did not result in any improved possibilities of distinguishing between objects than is the case with visual observation of the objects.

It is therefore the object of the invention to provide an improved method for Acquisition and quantitative determination of the shapes of cells and other objects of biological interest. In particular, a method for mapping and measurement of shape changes in cells are given and variables are developed, to analyze the many geometrical aspects of cell shapes.

It is therefore an object of the invention to provide a method of collection and to indicate for an analysis of information from cells or cell organelles, to improve the diagnosis of abnormal medical conditions, including of malignant conditions and those previous conditions.

The invention therefore provides a method for determining new Interference principles and geometric descriptors in relation to biological cells or subcellular particles are created. The values of the key-value descriptors are for the suspected normal or abnormal size and shape characteristics of the specimen and are calculated using a known statistical range of values for the specific type of cells or organelles that are being examined.

The invention also provides a method for collecting data indicated, which refers to the resolution of marginal details of cells or organelles and quantitative findings on the thickness properties of cells or organelles. These data form the basis for the procedure for Differentiation between different cells or organelles of populations on a statistical basis.

The invention is to be explained in more detail, for example, with the aid of the drawing will.

They show: FIG. 1 a schematic representation of an interferometer, that of a reflective material and a surface layer for production there is a constant interference increment in the reflected light; Fig. 2 illustrations of cells grown on an interferometer with an anodic oxide layer are; Fig. 3 is a schematic view of the comparison of manually drawn Illustrations A, C with images B, D reconstructed from points in a virtual memory; Fig. 4 is a graphical representation of the evaluation of points that of the contours of a 1000 W cell were collected; 5 shows a comparison of bivariable graphical representations selected descriptor values for the 1000 W (W) and 165S (S) cell lines with a Representation of the decision limits; 6 shows a comparison of bivariable graphs Representations of selected descriptor values for early (1) and late (2) passes the 1000 W lines showing the decision limits; Fig. 7 a Table of a three-way comparison of the variance of descriptor values by the chi-square test; 8 shows frequency histograms for explaining the distribution of selected descriptor values for lines PB3-0 and WT clone 9; 9 shows a comparison of graphical representations the distribution of principal component values for 50 cell samples from five different Epithelial populations; Fig. 10 is a table showing the results of tumor formation tests with lines selected to study shape change; Fig. 11 is a table corresponding to chi-square and t-tests on a range of populations that derived from the 1000 W line; Figure 12 shows a comparison of graphical representations of analyzes corresponding to a least square mean error, which performed on early, middle, and late passes from the 1000 W line became; Figure 13 is a table of the correlation coefficients of individual shape descriptors as a function of time for three lines that are progressively carcinogenic in culture became; 14 is a table of robust descriptors obtained by stepwise discriminant analysis, exerted on IAR 6-7 cells were selected; Fig. 15 is a table showing the classification of IAR 6-7 samples by probability estimation based on rigorous and expanded descriptor sentences; 16 is a table of linear discriminant analysis for four passage levels of the IAR 6-7 line; 17 shows a comparison graphically Representations of the non-iterative probability light analysis that are performed at four pass intervals the IAR 6-7 line was performed; Fig. 18 is a schematic Representation of interference profiles obtained from cells which the affect earliest (A, B) and latest (C, D) IAR 6-7 phenotypes; 19 shows a multiple linear regression analysis showing the frequency of classification in several Shows categories of predicted time; 20 shows a logarithmic-linear regression analysis, which the frequency of classification in different categories of the predicted Time shows; Fig. 21 an example of the earliest (A to C) and latest (D to F) Phenotypes selected by linear regression analysis; and FIG. 22 an example of the earliest (A to C) and latest (D to F) phenotypes that passed through the analysis were selected for maximum likelihood.

The invention provides a method for collecting and analyzing Information on the shape of cells given and an interpretation of illustrations performed by cells attached to a substrate.

A selective interference can be carried out by im reflected Beam path of an interferometer examinations are carried out from a reflective material and a surface layer to create a constant There is interference increments. A precise interference pattern is used with certain Wavelengths due to the high index of refraction of the oxide and the possibility of one Multiple passages formed through the oxide layer. Fig. 1 shows a schematic Representation of such an interferometer, which is made of a reflective material and a surface layer. The different wavelengths of white Light result in superimposing phenomena that lead to an extinction or to a lead to maximum gain, as shown at A and B, respectively. The phase difference, which is caused by a dielectric layer deposited on the surface of the Interferometer is adsorbed, can lead to a change in a maximum attenuation lead to a partial weakening, as shown at C. Because the surface is observed in reflected light, the oxide layer becomes at least twice traversed by the beam path. This optical method is advantageous for imaging usable by cells. First, the interference pattern is created by the adsorption of thin layers of biological material changed and it is possible to have a thickness dissolve from adsorbed material of less than 3 nm. That's why they get thin The edges of cells are effectively resolved and the finest cytoplasm projections along the scope become visible.

Furthermore, the repetition results from interference orders in the Figure some information about the third dimension of the shape of the cell, viz about the height of the cell.

Metallic tantalum and its oxide are used as interference substrates Use.

Other metals such as zirconium also show a reproducible relationship of interference colors in the event of oxidation. Under certain conditions there are faint colors visible in other metals, especially aluminum, copper and iron. the Similarity of these colors can be improved by adding a layer of reflective Material like gold, silver or aluminum is applied to the surface of the oxide will. Therefore, one of the essential factors in using an interferometer, namely the reflectivity through a secondary modification of the optical system be modified. Other factors affecting the optical quality of the interferometer contribute, are the reflectivity of the substrate below the Oxide, the phase difference associated with the reflection from this material, as well as the Refractive index of the oxide layer. In particular, the refractive index of thin amorphous Oxide layers show considerable differences compared to the refractive index corresponding crystalline body. Other factors of interest in dielectric Materials are the wettability of the surface, since surfaces are easily wettable are required for the cells to adhere to the substrate.

Because of certain difficulties of a practical nature in using Tantalum, other materials suitable as substrates have also been investigated Germanium was unsuitable for the attachment of cells to the substrates examined. The following materials are listed in an order similar to the visual Contrast in the imaging of cells corresponds to: gold-palladium alloy, molybdenum, Chromium and silicon oxide. Other materials can also be used. Though many Low reflectivity compared to tantalum, silver, and gold metals Having chromium, high reflectivity is not absolutely necessary because light passing through the layer does not contribute to the imaging.

Attached to an interferometer with an anodic oxide layer Cells can be observed as geometric objects. Because of the serial Interference patterns can be topographical representations of the ones shown in FIG. 2 Kind to be preserved. Fig. 2 shows a view of images of cells 1 on grown on an anodic oxide layer of an interferometer. The resolving power the edge formation is achieved by arranging the first dark interference fringe optimized on the cell border. This interference color is dark because of it a subtraction takes place when the wave trains are superimposed with the same wavelength. The appearance corresponding to a topographic map is based on the alternating one Illustration of dark and light colors in the different orders of the Due to interference (600x. Bar = 40). The contrast of these figures will be modified by adjusting the thickness of the oxide layer. Because of the high contrast an interpretation of such images is easier than with images that are colored Cells observed in light passing through. The contours become stronger with Magnification observed. Find out all or part of the cell map a conversion to digital form by taking certain coordinates of an or several cell contours can be applied.

The recording can be done by hand or using a plotter with a video device and a computer for image analysis take place.

A comparison of the original figure with the binary figure shows that the original features of the contours are retained with high accuracy (Olson, 77 Proc. Natl. Acad. Sci. U.S.A. 1516, 1980).

3 shows a schematic view of the comparison from manual drawn images A, C with reconstructed digital images B, D. That Resolving power is about 0.5 µ and is close to the resolving power for visible light.

A number of variables representing geometric aspects of the cell shape were developed by Olson in the referenced publication.

Those used to develop the methods described below Descriptors include measurements such as area size, circumference and length of the main axis an ellipse adapted to the contour. 35 additional geometrical measurands calculated, most of which then on one of the three-dimensioned Deskri ptors are normalized. Measurements of a standard deviation are taken from dimensionless Measured variables are derived and therefore do not require normalization. The degree of curvature will normalized to the number of points obtained on the perimeter. It is possible to measure values of different descriptors and use these as the normalization values to determine. For example, the area size and the circumference of the ellipse are measured and used as a basis for expressing additional values instead of the area size and the perimeter of the contour. The length of the longest (or shortest) dimension of the Contour can be used instead of the main axis of the ellipse. Furthermore, for Time used descriptors after a normalization to the area size of the contour normalized to a differently dimensioned descriptor. For example, can the area size of the ellipse, now normalized to the area size of the contour as ARAT, can be normalized to the square of the circumference of the longest cell dimension. In addition to these descriptions, further descriptors can also be generated thereby that the relationships between simulated or mirrored descriptors calculated in different contours of the same cell. That way can have many different sets of descriptors that can be mathematically different from differ from the originally defined sentence. Although these descriptors deliver special data values as unambiguously as those obtained by the present Set of formulas are obtained, there is the common characteristic of all such Records in that they are based on information obtained from the interference pattern of the reflected light can be derived.

As the optical technology information with high resolution via the fine structure of the cell border, you can choose from a number of these Variables limiting features are evaluated. An example of how digitized Illustrations can be mathematically converted into the descriptive values in that the perimeters are "unfolded" and the points along the contour accordingly their distance from the centroid. The number of projections on the scope are determined by determining the number of "peaks" in such a Record counted. FIG. 4 shows a schematic view of the evaluation of Sites collected from contours of a 1000 W cell. In Fig.

4A the original contours are shown by dots. The the model figure Representing ellipse, the ellipse of concentration, is as a solid line shown. The origin of the investigation is in the first contour by one Arrow indicated. In Fig. 4B, the center and focal points of the first ellipse become used to "unfold" the contour. This is done by calculating the distance between these internal points and each point on the contour. The distance is calculated based on the focal points (above) and the centroid values (below) and plotted on the contour as a function of position. These distances FOCI or CENT provide information on the extent to which the contour deviates from a circular one or elliptical shape. Optionally, these points are based on their distance applied from the focal points to the effect of eccentricity in the shape of the Negate contour. The protrusions or evaginations will be along the contour considered as triangles and the dimensions of their median, height and base measured.

In the following an example is to be described, whereby by visual Observation a distinction between objects can be made. The effectiveness of the descriptor set when distinguishing between different populations of cells was evaluated. A preliminary classification test was given two Cell lines are carried out differently from each other readily using a phase microscope could be distinguished. By simple bivariate records of some of the descriptors, the objects are resolved into two populations, with misclassifications occur in fewer than five out of a hundred cases. Fig. 5 shows a comparison of such Records of selected descriptor values for the 1000 W (W) and 165S (S) cell lines, from which decision limits can be seen. In Fig. 5A enables an illustration of CENT (ordinate) against ACAV (abscissa) one Dissolve the 100 cell sample into the correct groups with four misclassifications. In Fig. 5B, a Representation of SHPF (ordinate) against PSHR (abscissa) a resolution of the cells into the correct groups with five misclassifications. In addition to the examples In Fig. 5 there are other combinations of parameters used to distinguish between the populations can be used with this level of security. which may represent normal and abnormal mechanisms of growth control however, be of identical origin, so that finer differences between these too are expected. Therefore, comparisons of numerous descriptor values were made accordingly those in Figure 5 performed using early and late passes a single cell line, 1000 W. Even the most effective combinations of descriptors only led to a separation of about 80% of these cells into discrete classes, such as shown in Figs. 6A and 6B. Figures 6A and 6B show a comparison of bivariates Representations of selected descriptor values for early (1) and late (2) passes the 1000 W line, from which the decision limits can be seen. Enabled in Fig. 6A a plot of PSHR (ordinate) versus MAXP (abscissa) that the majority of Cells are resolved, but 20 misclassifications occurred. In Fig. 6B enables a plot of PTOM (ordinate) against SDFD (abscissa) that the Majority of cells resolved, however 22 misclassifications occurred. In addition to the combinations of descriptors shown, each solves for PSHR, The combination of PTOM and MAXP plotted the populations at that level. Other Populations of descriptors, especially records of the most effective descriptors (MAXP, PSHR, PTOM and SDFD) against ASHR, LCAV or SHPF also give correct classifications from about 75% of the cells.

To explain the usefulness of these descriptors for identification of specific cell types in a mixture of heterogeneous cells was a three-way comparison performed by descriptor values generated by cell lines and sublines became. Figure 7 shows a three-way comparison of the variability of descriptor values corresponding to the chi-square test (a line shows that the populations below p = 0.05 were not significantly different).

The data contained in Fig. 7 shows relationships between the populations.

Parameters that determine the actual dimensions of the cells, tend to deviate in variance regardless of whether the populations are close or distantly related. The values of other measured variables, such as spatial and model figure comparisons, Shape features and dimensions of minor contour irregularities tend to be identical variance in closely related populations. The variables therefore the cell features which are the least malleable are related to the spatial and model figure comparisons. In the example shown in FIG these values are the same for closely related populations. The numeric values the traits measured for these populations are visualized to ensure that their distribution is consistent with the static measure. Fig. 8 shows frequency histograms to explain the distribution of selected descriptor values for the cell lines BP 3-0 and WT clone 9. For the populations concerned are their descriptors and means: OCNT-7.7, 6.1; SHPF - 4/31, 3/21; AFRN -0.364, 0.395; MINP - 0.802, 0.876. All of these values differed in distribution for the two populations (see Fig. 7). Compare as those for BP3.0 and WT clone 9 lines as shown in Fig. 8 clearly show changed distributions for the feature values corresponding to differences of the chi-square test, as shown in FIG. 7.

The mean values of the descriptors for closely and distantly neighboring Populations are compared. A comparison of closely related lines reveals that numerous descriptors with equivalent variance exist which differ in terms of differ in their mean values. Almost all of these descriptors fall into classes of form factors (SHPF), dimensions that affect the deviation from the circular shape (ARAT, FOCI) as well as dimensions of smaller irregularities of the contour (BMPS, SDMD, SDAL, SDWD and MDAL). Some of these latter characteristics also have different ones Means when comparing distantly related populations as shown in Fig. 7 is shown. In the case of the one shown in Fig.

7, however, could only be a fraction the total number of descriptors should be checked for the requirement that the Variance of their values should be equivalent. Therefore the values of the descriptors can be compared more easily in similar populations. This analysis indicates whether most of the descriptors can be expected to have a distinction between populations are appropriate. With the exception of a few spatial Compare (DCNT, ANGL) and measure of minor imperfections (ALTI, WDTH) and a single descriptor (LNNC) for a negative curvature show all descriptors a considerable difference in either variance or mean at least one of the pairwise comparisons. The results show that most Descriptors have a high information content and therefore in the eventual Sentence to be retained. Similar statistical comparisons, which for the same Lines were carried out initially in bivariate graphs (1000 W and 165 S) showed that only six descriptors exhibited equivalent variances (MEDN, SDMD, SDAL, MDAL, MINP, CAVS). Therefore appear these lines as considerably different.

Attempts to distinguish between dissimilar populations (Fig. 5) show that the classes are not completely covered by a single decision surface a bivariable graphic representation can be separated. From these attempts and show conclusions that can be derived from the statistical univariation analysis (Fig. 7), that data from a number of descriptors must be used to make the classifications to optimize. However, many multivariate pattern classification methods are based on the assumption that the populations tested are normally in a characteristic Space were distributed.

SOME M ultivari ation m etch C I assi fi ci ation proce dures require the additional assumption that the population variances were equal.

A normal multivariate distribution implies that the population samples fall into a single group with a center defined by the mean vector the descriptor values and a shape is defined by the covariation matrix is determined. To the distribution of populations in a multidimensional characteristic To illustrate space, descriptor values for cells are of five different types Cell lines derived. These multivariable data are represented by a main component channel yse transformed to get compound variables with the same variance. As can be seen from representations of four main components, the large one appeared Majority of cells tested in a line distributed normally. Figures 9A and 9B show the distribution of principal component values for samples with 50 cells that are five different Empithelial populations were taken. In Fig.

9A are all populations tested as elliptic distributions of a projected graphical representation of main components 1 and 4. The distributions of populations A (dashed line, large ellipse), B (dashed line, small Ellipse) and C (solid line) are indicated. In Fig. 9B, most are the tested populations as an elliptical distribution on a representation of the main components 1 and 2 projected. The distribution of A (solid line) and B (dashed Line) are shown. However, a few of the cells in Group A are on one Outlying position projected separately from the main group (top left). Although most of the samples contained some cells, the outlying positions on the representations, there was little evidence that that the lines contained recognizable subsets, those with a bimodal distribution consistent are. In one case, in vvelchem a clearer one Subset was resolved, the population was from a primary culture of normal Tracheal epithelium derived and probably contained a smaller portion on mesenchymal cells. Therefore the indications are consistent with an assumption that the established epithelial cell lines have a unimodal distribution.

These methods include measuring and comparing cell shapes provide information about the shape properties of individual cells and cell populations, from which values numerous conclusions can be drawn. First are a large number of different geometric properties useful for definition the form. Second, certain populations are so dissimilar that they can be identified by a Number of different combinations of descriptors can be distinguished. Although the combinations that appear maximally effective to make comparisons within or to perform between cell lines, can be different they descriptors from each class as shown in FIG. When analyzing of the 1000 W and 165 S cell lines they differ so radically that 95% of the examined individual cells linearly based on only two decision criteria are separable. Third, seemingly similar cell populations have a substantial number Overlapping of the values of most descriptors, however, can result in the appropriate ones Classes resolved with the help of a mapping analysis and pattern recognition process will. Precise quantitative data can be obtained on the extent of the differences will. Fourth, cells tested show significant differences within a line in terms of shape. The results, including those from the principal component analysis, however, are consistent with the assumption that any population is normally im n-space is distributed. For a person skilled in the art of optical analysis is it can be seen that these principles apply equally to the analysis of subcellular Particles are applicable.

The following is an example of how to distinguish between Objects are described that cannot be distinguished by visual observation can be. After elucidation of these form properties of populations of epithelial cells other studies were carried out with cell lines in which changes occurred which are not based on optical examinations with visible light of individual Cells could be carried out. In connection with the oncogenic transformations changes were found in epithelial cells from respiratory tract (Heckman and Olson, 1979) which indicated a change in the deformation properties of cells. These Changes were determined by scanning recordings with an electron microscope, which resulted from small colonies of cells, mainly due to clones was. From these results it was understood that during continuous growth in vitro in tumorigenic lines the majority of cells a subtle Experienced change in shape properties. These changes were independent of the the exact length of time the line was in the culture. To these changes to identify and finally to determine whether there is a connection with mechanisms an oncogenic transformation existed, the cell shape was examined in lines, which retained tumorigenic properties as they grew up in vitro. The criteria to select this line were set beforehand, so that the loss of growth control could be substantiated through empirical experiments, especially with regard to tumorigenic properties, and the changes were made according to one Timescale showing the collection of populations at widely spaced time intervals enables. Several of the F344 airway epithelial lines identified by Marchok and Steele (Marchok et al., supra 1978, Steele et al., 39 Cancer Res. 3805, 1979) these criteria. In addition, cell lines from IARC (International Agency of Research on Cancer! from the liver of BD-IV and BD-VI rats (Montesano et al., 1975) also examined.

The 1000 W, IAR-20 PGI and IAR 6-7 lines were used in experiments Identify shape changes associated with the development of tumorigenesis stood. Data on the tumorigenic properties of these lines are summarized in FIG.

While populations from earlier and later passages from the 1000 W line can be partially resolved in bivariate representations, the Analysis optimized by utilizing the information content of numerous descriptors will. Previous pairwise comparisons of cell lines indicate that no specific Disadvantages occur when statistical analysis of closely related populations he follows. While only a fraction of the descriptor values show a similar variance, when unequal cell lines are compared (Fig. 7), is of a greater proportion expect this to have a similar variance when comparing successive passes of a single cell line. By determining the differences in the means and the variance for multiple populations, taken from the 1000 W line as a sample is shown to be about one Thirds of the dimensionless descriptors show a different variance. A large number of these descriptors (SHPF, PTOM, NONC, FRNC, SDFD, MAXP, ASHR, PSHR and ACAV) differ in variance and mean or both for all populations considered. Of the variables involved in pairwise comparisons showed an equivalent variance, some differ in the mean. It will however, not a single variable is set that provides a basis for the resolution of provides more than two populations. For example, the differences between the earliest and latest checked passes can be based on the mean values can be recognized by CSQD, ALTI, FOCI and FINE. If populations correspond to a medium and later pass are examined, but only have the descriptors AFRN, DCNT, and CAVS means that were different while their variance stayed the same. In the case of a renewed derivation of a sample that has been separated from a adjacent passage of the middle population took place, but it was found that these differ from the late population in the values of SDNC, MINP, ASHR and CAVS differed as shown in FIG.

The extent of the summed shape changes was quantitatively determined by the linear least squares algorithm determines the variable values at the Adjusts vector of predicted properties. This algorithm LEAST (Begovich et al., A User's Manual for the Pattern Recognition Code RECDG-ORNL, ORNL / esd / TM-21 Natl.Tech.lnfo.Serv., 1977) was tested on three of those tested in the previous analysis Populations applied as shown in Fig. 11 to be able to recognize whether these could be separated by linear decision boundaries. By assigning the Values of the cumulative time the population was grown to the vector the predetermined properties was a solution of weighted vectors for each of the test classes received. The adaptation of this solution to the training rate was made recognizable by reclassifying each of the objects, although each one Was unknown. The resulting classification shows an approximately correct one Estimation of the mean times since these populations were grown elapsed as shown in FIG. Fig.

Figure 12 is a comparison that is a least mean square analysis Error shows that at an early, middle, and later pass of the 1000th W cell line was performed. The actual time that has passed since the Populations elapsed was 369, 522 and 606 days, respectively. The information presented represent the mean values recalculated for the individual cells of each group became. Although the populations were partially disbanded, the decision boundaries resulted There are still considerable areas of overlap between the classifications (arrows).

Because classifications of the training set provide a better resolution compared to those for unknown sentences, some data sets were analyzed by an additional 1000 W populations. When classifying a sentence of 50 cells from passage 39, using weighted vectors generated with known sets were used, it was found that the distribution of predicted time values wider in this example than for the passage 38 cells were 1 but the Means were similar. Attempts have been made to find the various Classify populations of passage 50 using the elapsed Time since disguising as a vector of predicted properties. In this attempt it was not possible to achieve a linear separation even with the training sets. Therefore, these analyzes1 confirm that collected over a wide range of time periods Populations differ in terms of cell shape, and that these distinctions not readily induced by trivial changes in breeding conditions for can be, for example by the time, the / an attachment and expansion was allowed for determining which factors affect the process of form revision In affecting epithelial cells, it is necessary to establish quantitative criteria for one Define change. Although the LMSE procedure used for three pass intervals the 1000 W cell line was carried out, exemplifying the type of dissolution that might be expected with a trainable pattern classifier became the first application to the 1000 W line based on the values of all of the descriptors, some of which had little impact on the analysis. Although there is no formal redundancy was present among the variables, it is likely that some of their values coordinates due to inherent properties of the measured biological Change structures. For example, depending on the relationship between structural Elements that determine the projections on the cell periphery can use the Dimensions of these projections may be different in a coordinated manner Kind of a biometric redundancy is made by the step-by-step differentiation analysis eliminated, which is based on the covariance statistics of the descriptor values. This analysis uses the F-statistic, where the variance of the means for different Groups are compared with the variance of the combined samples from all groups, to select a variable to include in the list. Using the Vector of the means for all other variables except for the List selected as a measure of the differences remaining between the samples, will the F-value recalculated for each variable. The second selected The descriptor is therefore the one who takes into account the main variance, which does not has already been corrected by removing the first descriptor. This procedure is repeated step by step until the calculated F statistic is no longer required Level of importance.

Although the classification of the populations by the LEAST algorithm shows how quantitative criteria could be supplemented result from this Algorithm certain problematic aspects. As by reassigning Vector values for individual 1000 W cells in a preliminary classification attempt is displayed (Fig. 12), the mean values did not correspond to the predicted values always the actual time elapsed in breeding. Performed with this line additional experiments indicated that there was a trend for the means between the actual time and all of the time values, due to the proportion earlier and later populations that are misclassified as cells of an intermediate phenotype occurs. A more efficient method involved preselection of complementary variables, each of which had values that are highly related to are correlated with time. Since 1000 W populations are already classifiable proved, the IAR 6-7 line was used for further analysis. By calculation the correlation coefficient of the descriptor values with time was determined, that the descriptors were in some cases identical for the two cell lines.

The correlated descriptors were evaluated for the determination whether their values over time as a result of a normalization to a one-dimensional Descriptor are correlated, which changed itself coordinated with time.

Although the dimensioned descriptors are not directly used in the analysis were used, other variables were normalized to their values. An examination of the highly correlated variables in the case of the 1000 W line shows that at least a descriptor is normalized to each dimensioned descriptor, as in Fig. 13 is shown. For the IAR 6-7 cell line, only those descriptors appear which are normalized to the length of the circumference or the main axis than with time correlated. Although this shows that the AREA descriptor is in non-uniform Way could have changed so that descriptors containing these values as well correlated less well, it was found that those actually found AREA values decrease progressively over time. In addition to determining which Factors influencing time-dependent values were determined to determine whether there was a similarity between the lists of correlated descriptors that for different cell lines were obtained. The similarity would indicate that consistent changes in cell structure in the course of oncogenic transformation occurred. Comparing the 6-7 cell lines generated for the 1000 W and IAR Lists found that seven of the time-dependent variables found were equal and that the direction of change for six variables, namely AFRN, CSQD, BMPS, WDTH, PSHR and CAVS are identical for the relevant contours of the two cell types was. The 7th descriptor FRNC showed a difference in direction the change for the comparable first contours of the cells of the two lines. This result shows that certain characteristics of the cells are uniform during long periods of time are affected in vitro.

When the method according to the invention is applied to cells corresponding to samples from the IAR 6-7 line is exercised, results in the incremental discriminator analysis a list of seven descriptors. Your identity and F-values in the eventual Iterations of the program are shown in FIG. With the exception of FINE 1 the list consists entirely of descriptors whose values move very well over time are correlated. The most significant changes in cell shape were a function the time in breeding. The exception descriptor FINE 1 had similar mean values for samples corresponding to the first, third and fourth point in time, and differentiates between all these samples and the sample corresponding to the second point in time. His Inclusion on the list is due to the effectiveness in performing this single Discrimination, which is to be seen as a characteristic of the analysis, the even with an early registration to distinguish unrelated populations was observed (Olson et al., loc. cit. 1980). Three of the descriptors what information About the areas with negative curvature encompassing the contour were considered complementary Information including selected as shown in Fig.

14 represents. This got over the difficulty of making predictions indicated the types of descriptors that might show biometric redundancy.

Although the most accurate classifications with an analysis on the Based on the values of all descriptors can be obtained, those of interest can Populations including those experimentally unknown by using the list resolved by robust descriptors. To evaluate the effectiveness of Classifications based on a strict set of descriptors was an analysis done for maximum likelihood. if the data structure is represented by nonlinear functions, then you can estimate the maximum likelihood or other nonlinear regression methods used. to adapt functions to data. Some of these algorithms collect no requirements regarding the distribution of population samples. Others are applicable to distributions that have an exponential shape, for example normal, Bionom, Piosson or gamma distributions. In probability analysis becomes the vector derived from the descriptor values together with the parameter vector used to define one of several time-independent functions, their Derivatives can be specified. In addition to specifying the leads of the functions are provided by the source programs (BMDP Statistical Software, Ed. Dixon W.J., Univ. Calif. Press Berkeley 733, 1983) iterative algorithms for improvement the estimation of the parameters. The parameter vector determined in the analysis can then used to assign each cell a value equivalent to its predicted time is in breeding. This is why the IAR 6-7 cells can get through an approximation similar to that in which vectors for prediction using the LEAST algorithm according to Fig.

12 can be used. However, when estimating the probability stopped the analysis after receiving the first estimate so that the "innate" structure of the data instead of the goodness of fit to a number of logarithmic functions can be estimated.

If the probability estimate for the IAR 6-7 cell samples on the basis of the parameter vectors that comply with the strict theorem of seven descriptors (Fig. 14) are obtained showing the mean values of the earlier and later tent estimates considerable deviations from the actual mean values, as shown in FIG. However, descriptors were emphatically emphasized, which were robust in two respects, namely when distinguishing between relative closely related populations and in making time-dependent distinctions. An extreme example of the former capability was the selection of FINE 1 by the step-by-step discriminator analysis, as shown in Fig. 16, although the Correlation coefficient of time was zero. An examination of the mean values of descriptors the strict list shows that some of the descriptors are in a non-progressive one Way changed between certain time intervals. To determine how the probability anal yse has been influenced by this, a more extensive list of descriptors is created by removing made of descriptors from the strict list, if their values don't have one progressive Showed change for more than one time interval and by adding several descriptors, which are also to a large extent related to the Time were correlated.

A second probability estimation analysis based on a The extended list of descriptors revealed few pronounced discrepancies between the predicted and actual values1 as shown in FIG.

In order to make the time course of the changes recognizable, a frequency histogram was made prepared according to FIG. 17, which shows the distribution of cells in a row of predicted time categories shows. Fig. 17 is a comparison showing the non-iterative probability analysis shows that at four pass intervals the IAR 6-7 line was performed. The progression of the shape changes were non-linear. Some of the variations were due to technical factors as populations maintained from varying periods of time in USA to an origin in France became. The predicted mean values of the time in the culture are shown in Figure 15. Individual cells that represent the earliest and latest phenotypes (arrows), were retrieved from the data sets for comparison. Although the biggest difference was found in mean values again between the intermediate points in time, corresponded the range of the predicted time values corresponds to the rank of the actual time values. Individual Cells were considered to be representative of early or late time points in a progressive manner Series identified and the earliest phenotypes, assigned time values of -378 and -66 days, became the latest with associated values of 871 and 796 days as shown in FIG. 17.

When the profiles of these cells are reproduced as shown in FIG there were differences in different respects. Fig. 18 shows a schematic representation of interference profiles obtained from cells, which represent the earliest and the latest IAR 6-7 phenotypes. The position of the cells within the entire sample is indicated in FIG. The first contour extends along a considerable distance beyond the second in the early Cells1 is restricted in the late cells. Because the second and third Contours closer together when the late cells were, the cells became progressive more rounded in their edge areas. Although a reference to the original drawing for the cells in Fig. 18 indicates that the number of contours increased from four to five, the correlation coefficient of this descriptor (OCNT) with time was 0.

Therefore, the number of contours did not show a consistent increase with time.

The changes exemplified by these extreme phenotypes relate to the time-independent changes in the descriptor values. Though through the value of the main concavities (CAVS) represented structure and fine projections (BMPS) are more prominent in the contours than early cells, they actually are less often when their number is in proportion to the actual cell dimensions is expressed. Furthermore, the shape of the structures of the first contour is changed to indicate a more continuous curvature, as by increasing the summed quadratic values of curvature (CSQD) and for all dimensions of the projections (SDTH, ALTI, MEDN) is indicated. With regard to the gradual changes in the projected cell area characteristic of the IAR 6-7 line is to be expected the descriptors normalized to the surface show progressive changes.

Taking into account that all the variables normalized on the surface refer to the dimensions of the main concavities or to model figure comparisons, shows the lack of changes in these values (Figure 13) that the lamellar structure of cells "shrunk" in relation to their projected area. With regard to on the postulated mechanism of transformation it was found that the Centers of the cells that dictate a major cell plasma organization, were probably not disturbed. The organization controls in 1000 W cells, which showing small progressive area changes can be done in different ways be influenced, since many of the area-normalized descriptors time-dependent changes in this line as shown in FIG. That line showed one little progressive change in projected area. The two "evolving" lines show in spite of originating from widely different ones Organ sources common changes of shape. A smoothing of folds and Projections in the cell periphery and rounded height dimensions are common Responses to oncogenic transformations in the lines. The difference in trends for the variables reflects a more extensive central cell organization because ASHR, PSHR, DAVS, ACAV, CVSD and LCAV indicate that the global organization in is disturbed to a greater or lesser extent.

The preceding results lead to the classification of objects based on their shape. First, all descriptors are used to a high degree are correlated with the predicted properties.

Then the descriptor list becomes gradually discriminatory or subjected to a stepwise regression analysis to exclude variables, which show a biometric redundancy. The type of analysis used depends of that depends on whether a linear or non-linear model is used to interpret the relationship between the descriptors and the predicted property is applied. The remaining small matrix of variables is finally used to determine the properties of individuals to predict.

Since it is possible to classify the two lines described above, the method according to the invention was applied to a third cell line IAR-20PC1. On the basis of a model of linear time-dependent changes were descriptors selected by stepwise regression analysis (BMDP Statistical Software, loc. cit. 1983). This analysis showed that most of the variations in the data were due to the values of the Descriptors OCNT, NONCI, AFRN3 and CSQD1 could be expressed. The regression was also of great importance. An F ratio of 29.70 was found while a value of 2.74 or more only 1% of the time due to a sampling error was observed.

19 shows a multiple linear regression analysis of the line IAR-20 PC1.

The analysis shows the frequency of classification in different Predicted Time Categories. For the four populations tested are the predicted times 195, 211, 282 and 310 days, respectively. The experimental times, too the populations examined were 142, 184, 299, and 373 days.

The regression equation used was: Y = 61 4-54 * OCNT-39 * CSQD1 -135 * NONCl + 123 * AFRN2 + 4 * OCNSQ + CSQSQa24 * AFRSQ The regression analysis resulted in that shown in Fig. 19 Classification results. Tended to the extremely early and late points in time the predicted time values to a closer approximation of the population mean as a whole as to be expected from the range of their original time values per se was. A similar trend was seen in the previous analysis of 1000W data, which are fitted to a linear regression by the LEAST algorithm.

An alternative model based on a logarithmic linear Kinetics was also used to fit the IAR-20PC1 data. In this case the descriptors were selected through a step-by-step discriminatory analysis and a list of six complementary values was used in the classification attempt used. The results of this analysis are shown in FIG. Fig. 20 shows a log linear regression analysis of line IAR-20PC1. The analysis shows the frequency of classification in different Categories the predicted time. For the four populations tested are the average predicted times 163, 150,320 and 365 days, respectively.

The experimental times at which the populations were tested were 142, 184, 299, and 373 days. The variables used in the analysis were OCNT, CSQD1, NONC1, MINP1, PTOR1, FOCR1.

The trend determined on the basis of linear models in predictions, after which predicted values converged to the mean of the entire sample unavailable. Predictions derived from the log-linear model therefore appeared more accurate than those derived by a linear model became. But both had a high value because of the logarithmic-linear regression also showed an F ratio of 25.23.

While it is important in theory, it was as predictive Tool used mathematical model of pragmatic importance only when the Results of the discrimination process were significantly affected. That's why it was determines whether the extreme cell types identified by the alternative analysis were, were similar. Such a comparison is shown in FIGS. Fig. 21 shows examples of the earliest (A to C) and latest (D to F) phenotypes, selected by linear regression analysis of line IAR-20PC1. the Predicted times assigned to these individuals are: A + 96, B + 111, C + 114, D + 421, E + 404 or F + 388 days. The identical cells are also saved as early and late considered in the logarithmic-linear analysis: A - 3, B + 81, C - 22, D + 491, E + 419 and F + 433 stretcher. Fig. 22 shows examples of the earliest (A to (=) and latest (D to F) phenotypes identified by analysis for maximum likelihood from line IAR-20PC1.

The predicted times associated with these individuals are: A. - 104, B - 99, C - 97, D + 588, E + 558 and F + 556. The these individuals through the linear analysis allotted predicted times are: A + 146, B + 207, C + 146, D + 353, E + 336 and F + 371 days.

The IAR-20PC1 cells identified as the earliest phenotypes of these analyzes identified showed some disparities. Through the linear analysis became a phenotype with a high curvature relative to the length of its girth selected, during the log-linear analysis, a phenotype with a number identified by sharp features of the contours.

In both analyzes, those selected tended to be the late phenotype Cells to show a relaxed contour and considerable asymmetry.

Those identified as early in the linear model for analysis Cells received also low values when looking at it in the logarithmic linear model as shown in FIG. In in a corresponding manner, the cells received as the last phenotypes with the linear model for the analysis were considered, high values if an analysis according to the logarithmic-linear model (and vice versa).

These results showed that this was due to the most economical methods of the variable selection based K 1 assi fication method regressions with high statistical Wertiglleit generated. Furthermore, the conceptual and theoretical context in which the analysis was assessed had little impact on the classification, in particular in relation to the selection of the abnormal phenotype, which in this example is the is late.

- blank page -

Claims (5)

Claims 1. Method for measuring the shape of a cell or Cell organelles and for generating information about the shape properties of the Cells or organelles of a population that are not indicated t that a high resolution image of details of the edge formation of the cell or organelle of the cell and information on the level of the cell population generated using reflected light from an interferometer, which consists of reflective material and a surface layer through which a constant interference increment is introduced to achieve a resolution of the image of the cell or organelle to achieve that at least one geometric descriptor the resolved image of the cell or organelle is measured that the measured geometric descriptor of the resolved mapping of the cell or organelle with predetermined Comparing descriptors for known objects or populations of cells, and that the resolved image of the cell or organelle in one of the particulars known populations of cells or organelles based on quantitative data about the differences or similarities between the resolved image of the Cell or organelle and the known populations or objects classified which creates individually resolved images of the cells or organelles the base of the geometric descriptor are separable. 2. The method according to claim 1, d a d u r c h g e k e n n z e i c h n e t that the one or more of the measured geometric descriptors dimensioned Parameters of the cell or cell organelle, form factors of the cell or Cell organelles, spatial comparisons of the cell or cell organelle, on the curvature of the cell or cell organelle, related to the negative curvature of the cell or cell organelle-related dimensions, to minor irregularities in the contour of the cell or cell organelle-related dimensions, on deviations from the circular or elliptical Shape of the cell or cell organelle-related dimensions, with model figure comparisons of the cell or cell organelles or related to larger concavities of the cell or cell organelles Dimensions are. 3. The method according to claim 2, d a d u r c h g e k e n n z e i c h n e t that the geometric descriptor of the resolved cell or organelle by Conversion of the resolved image of the cell or cell organelle into a digital one Shape is measured by taking certain coordinates on one or more of the contours of the cells or cell organelles are applied. 4. The method according to claim 3, d a d u r c h g e k e n n z e i c h n e t that a computer is used for video-based image analysis, to convert the resolved image of the cell or cell organelle into a digital form. 5.- The method according to claim 4, d a d u r c h g e k e n n z e i c h n e t that the reflective material is a metal and that the surface layer is formed by an oxide of the metal.