A pattern recognition system for use with a two-dimensional image, each point associated with a height and with contributions from various wavelengths of light. The image to be analyzed is first sensed by a sensor and stored in a memory. Signals, read from the memory, are filtered and processed to produce signals measuring each of the wavelengths and measuring the height. The resulting distributions of data are first scanned for areas containing appropriate height measurements. Each pixel of these areas is classified according to predetermined classification definitions, further reducing the areas to be processed. These areas are subjected to spatial classification according to other classification definitions. The spatial classification may be accomplished by a neighborhood processor. The areas of the image are then identified with the features to be recognized.

In a matcher mechanism for verifying a match of a set of subject points having positional parameters against a set of reference points having positional parameters, the improvement of employing a set of Group tables derived from the difference table. Each line entry on the difference table constitutes a lead entry of a separate Group table. The Group table are overlapping subsets of the entries on the difference table. The members of the Group table are limited to only one line entry representing each reference point. That line entry is the one which is closest to the lead entry on the difference plane. In addition, all line entries within a Group are limited to those that are within a predetermined window on the difference plane around the lead line entry for the group. A quality score is determined for each of the Group tables based on a weighted proximity value between each pair of line entries in the Group involved. The closer a pair of line entries are to each other, the greater will be its weighted score and its contribution to the weighted proximity value for the Group table. That Group table with the greatest quality score is deemed to provide the quality score for the matcher. A match is determined if the value of that quality score exceeds a threshold which is a function of the number of reference identification points. There may also be a requirement that the quality score exceeds some minimum value.

An image analysis system is used for the quantitation of nuclear proteins in cell populations. Particularly, the hormonal receptor content of fine needle aspirates of human breast carcinomas are evaluated. Estrogen or progesterone receptors are amplified and visualized in the specimen by a staining technique of the immunoperoxidase type. Monoclonal antibodies specific against the receptor are attached to the receptor sites and are then amplified by a bridging antibody which attaches to the monoclonal antibody and a peroxidase-antiperoxidase complex. A chromagen, diaminobenzidine is combined with the complex and treated with hydrogen peroxide to react with the peroxidase forming an insoluble brown precipitate which marks the receptor sites for optical identification. The specimen is then counterstained with another chromagen, methyl green which is specific to the nucleus of each cell. Two monochromatic filterings optionally separate the areas stained by the receptor site optical enhancer and the nuclear area optical enhancer. Measurements of the optical density values of the stained receptor areas yield an intensity value directly related to the quantity of hormonal receptor in the specimen. A comparison of the nuclear area containing hormonal receptor with the total nuclear area yields a percentage value which indicates the distribution of cells throughout of the specimen which contain receptor. These two values for intensity and distribution are then combined to yield a predictive score for an assay. The measured score when compared to an empirically derived reference score is predictive of the prognosis of endocrine therapy.

Human beings judge patterns (such as images) by complex mental processes, some of which may not be known, while computing machines extract features. By representing the human judgements with simple measurements and reducing them and the machine extracted features to a common metric space and fitting them by regression, the judgements of human experts rendered on a sample of patterns may be imposed on a pattern population to provide automatic classification.

An apparatus for classifying leukocytes comprises a sample image preprocessing unit which modifies the pattern of the nucleus of each of predetermined number of leukocytes extracted from the image of a blood sample, a unit for analyzing each of the processed nucleus patterns and broadly classifying each of the leukocytes into a typical or quasi-typical band form neutrophil, a typical or quasi-typical segmented neutrophil, or other neutrophil in an intermixture domain or a leukocyte other than a neutrophil, and a unit for calculating sample estimation indexes from the numbers of typical and quasi-typical neutrophils and the total number of neutrophils, classifying each of the neutrophils in the intermixture domain into a band form or a segmented neutrophil in accordance with the estimation indexes, and adding the numbers of intermixed band form and segmented neutrophils to the numbers of typical or quasi-typical band form and segmented neutrophils, respectively.