Image analysis – Pattern recognition – Classification
Reexamination Certificate
1999-06-17
2003-04-15
Mehta, Bhavesh (Department: 2625)
Image analysis
Pattern recognition
Classification
C382S158000, C382S133000, C382S134000, C356S039000, C356S073000
Reexamination Certificate
active
06549661
ABSTRACT:
TECHNICAL FIELD
The present invention relates to a pattern recognition technology utilizing a computer. More particularly, this invention relates to a pattern recognition apparatus and a pattern recognition method which can be applied to a method of, and an apparatus for, executing pattern recognition suitable for an automated urinary sediment examination system for classifying particles in urine, and which can accurately classify objects having great individual differences without being affected by the individual differences.
BACKGROUND ART
A urinary sediment examination is the one that examines solid components such as blood cells, tissue cells, etc., contained in urine and reports the kinds and amounts of the components. It has been customary for a laboratory expert to first centrifuge a predetermined amount of urine, then to dye the resulting sediment components, to sample them on a preparation and to microscopically observe the components. Each component is classified in accordance with its features such as the shape, dyeability, and so forth. A method of imaging the solid components in urine as a still image is described as a technology for automatically executing the urinary sediment examination in, for example, JP-A-57-500995, JP-A63-94156 and JP-A-5-296915. These technologies involve the steps of passing a sample through a flow passage (flow cell) having a specific shape, causing particles in the sample to flow through a broad imaging area, turning on a flash lamp when any solid components are detected in the sample and imaging a magnified image of the solid components in urine as a still image. The sediment components imaged as the still image in this way are automatically classified by separating the region of the sediment components from a background region on the image, determining image feature parameters in the sediment component region and classifying the sediment components on the basis of these feature parameters. An area, a perimeter, a mean color density, etc., are used as the image feature parameters. JP-A-1-119765, for example, describes a region dividing method of a blood cell image as one of the technologies of separating the region of the solid components from the background region on the image. This reference segments the region of the image in a color space by using a threshold value determined from a density histogram. JP-B-58-29872 and JP-A-3-131756, for example, describe the classification of the blood cell images as a technology of classifying objects from the image feature parameters. JP-B-58-29872 employs a discrimination logic or dicision tree constituted by statistical discriminating functions in multiple stages on the basis of the image feature parameters. JP-A-3-131756 employs a hierarchical network as a recognition logic.
DISCLOSURE OF INVENTION
The individual difference of each specimen is great in the urinary sediment examination and even those objects which ought to be classified into the same class exhibit great differences in the shape and dyeability from specimen to specimen. Therefore, the individual difference renders a great problem for automatic classification. For instance, the size and the shape of the blood cells in urine vary with pH of urine, its specific gravity and osmotic pressure. Because the white blood cell is generally greater in size than the read blood cell, it is rare that they are wrongly classified. However, there may be the case where the white blood cell of a specimen shrinks depending on the condition such as the pH, the specific gravity, the osmotic pressure, etc., and is wrongly classified as the red blood cell. When this classification is done with eye, all the specimens are first checked as a whole so as to sort out typical those objects which can be judged reliably as the white blood cell and then to judge the overall tendency of the specimens that the white blood cells are rather small as a whole, or that there are a large number of white blood cells which are deformed, for example. Thereafter, the objects which cannot be classified easily are tackled. Even though the white blood cells are so small that they are likely to be mistaken as the red blood cells, for example, they are classified as the white blood cells if the typical blood cells of the specimen are small as a whole and if the red blood cells do not appear. Therefore, the conventional pattern recognition method which decides the classification class to which a given pattern belongs from only the given pattern cannot eliminate the influences of the individual difference for each specimen, and is not free from the problem that classification accuracy drops for those specimens in which rather small white blood cells peculiarly appear. It is an object of the present invention to provide pattern recognition which reduces wrong classification of objects resulting from the individual difference in pattern recognition of those objects which exhibit different features depending on the individuals (samples) even though the objects ought to be classified into the same class.
The first construction of a pattern recognition apparatus according to the present invention comprises first pattern recognition means for inputting a set of input samples constituted by a plurality of input patterns, classifying each input pattern in the set of input samples into a classification class to which this input pattern belongs, evaluating reliability of this classification result and outputting the classification class to which this input pattern belongs, as being recognizable, for the input pattern for which a classification result having high reliability can be obtained; first storage means for storing those input patterns among the set of input samples which are evaluated as having low reliability of the classification result obtained by the first pattern recognition means and for which recognition is suspended; second storage means for storing those input patterns among the set of input samples which are evaluated as having high reliability of the classification result obtained by the first pattern recognition means and which are judged as recognizable, and for storing the classification class, to which the input patterns belong, outputted by the first pattern recognition means; second pattern recognition means being constructed by using the input pattern stored in the second memory means and the classification class to which the input pattern belongs as a training sample, for inputting the input pattern stored in the first storage means and outputting the classification class to which the input pattern belongs; and pattern recognition method construction means for constructing the second pattern recognition means by using the input pattern stored in the second storage means and the classification class to which this input pattern belongs, as a training sample.
In the first construction described above, the second construction of the pattern recognition apparatus according to the present invention includes reference pattern recognition method storage means, and wherein the second pattern recognition means is initialized by using the content of the reference pattern recognition method storage means before the pattern recognition method construction means constructs the second pattern recognition method.
The third construction of the pattern recognition apparatus according to the present invention comprises pattern recognition means being set to the initial state before one set of input samples comprising a plurality of input patterns are inputted, for classifying each input pattern in the set of input samples to a classification class to which each input pattern belongs, evaluating reliability of this classification result and outputting a classification class to which the input pattern belongs, as being recognizable, for those input patterns for which a classification result having high reliability is obtained; first storage means for storing the input patterns in the set of samples for which the classification result is evaluated as having low reliability and for wh
Kojima Yasuaki
Matsuo Hitoshi
Mitsuyama Satoshi
Motoike Jun
Desire Gregory
Mattingly Stanger & Malur, P.C.
Mehta Bhavesh
LandOfFree
Pattern recognition apparatus and pattern recognition method does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Pattern recognition apparatus and pattern recognition method, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Pattern recognition apparatus and pattern recognition method will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3004577