Image analysis – Applications – Biomedical applications
Reexamination Certificate
1998-05-01
2001-09-18
Chang, Jon (Department: 2623)
Image analysis
Applications
Biomedical applications
C382S190000, C382S305000, C707S793000
Reexamination Certificate
active
06292577
ABSTRACT:
BACKGROUND OF THE INVENTION
The present invention is related to a resemblance retrieval apparatus for retrieving stored subject data resembling subject data, with respect to a designated retrieval subject, from a saved data group, and also to a recording medium for recording such a resemblance retrieval program.
Conventional resemblance retrieval apparatus are described in, for instance, a publication “Incremental Instance-based Learning of Independent and Graded Concept Descriptions”, D. Aha, Proceedings of the Sixth International Workshop on Machine Learning, 1987”, and another publication “A Nearest Hyperrectangle Learning Method”, S. Salzberg, Machine Learning, 6, pp. 251-276, 1991.
FIG. 8
is a schematic block diagram showing an example of such a conventional resemblance retrieval apparatus.
In this drawing, reference numeral
1
is a subject image designating unit for designating an image of a retrieval subject; reference numeral
2
is a feature quantity extracting unit for extracting a feature quantity which quantitatively indicates the feature of the subject image designated by the subject image designating unit
1
; reference numeral
4
shows an attribute input unit for inputting an attribute of the subject, other than the feature quantity related to the subject image; and reference numeral
5
represents a subject vector data forming unit for forming subject vector data in which both the feature quantity extracted by the feature quantity extracting unit
2
and the attribute input by the attribute input unit
4
are used as a vector structural elements. Reference numeral
7
denotes a vector database storing a plurality of vector data formed using the feature quantities and the attributes as vector elements; reference numeral
8
is an image database for storing a plurality of images corresponding to respective subjects; reference numeral
13
is a weight vector given to each of the vector elements of the vector data to calculate resemblance degree in a resemblance retrieval engine; and reference numeral
10
represents a resemblance retrieval engine for seeking vector data resembling subject vector data in the subject vector data forming unit
5
from a plurality of vector data stored in the vector database
7
. Further, reference numeral
11
shows a retrieval result display unit for displaying the resemblance vector data retrieved by the resemblance retrieval engine
10
, and also an image corresponding to this vector data; reference numeral
14
indicates an answer instructing unit for determining whether both the resemblance vector data designated by the resemblance retrieval engine
10
and the image are correct; reference numeral
15
shows a weight vector updating unit for updating the weight vector
13
based on the result determined by the answer instructing unit
14
; and reference numeral
12
indicates a new data adding unit for newly entering vector data and images in the vector database
7
and the image database
8
, respectively.
Now, operation will be explained. For instance, as to medical information, such as electronic medical diagnostic data and a medical image database, and as to design information, such as design drawings, that are stored, when data suitable for a new purpose are selected, the following resemblance retrieval technique for the vector data is applied. These data are rearranged as vector data stored in the database. Then, calculations are made to determine the resemblance degree between the vector data sought, which express a new purpose, and data saved in the database. The data in the database that most resembles the desirable vector data is found.
One example of such a purpose is aiding diagnoses of pathological tissue. In such a case, a pathological tissue image resembling a stored pathological tissue image is retrieved with respect to a pathological tissue image under examination. The purpose is to diagnose a disease by observing biological tissue. This pathological tissue diagnosis is mainly carried out to determine whether a tumor must be removed and to determine the sort of tumor.
FIG. 9
is a flow chart for describing the operation of the conventional resemblance retrieval apparatus. First, the subject image designating unit
1
designates a subject image for which a resemblance retrieval is to be performed, for, example, pathological tissue images to be examined (step ST
1
). Next, in the feature quantity extracting unit
2
, a feature quantity for quantitatively expressing a feature of the designated subject image is extracted from the subject image (step ST
2
). Subsequently, in the attribute input unit
4
, an attribute of the subject image designating unit
1
is input (step ST
4
). Examples of attributes of the subject image include patient name, patient ID, image ID, dimension of tumor, age of the patient, diagnosis title, and the like. It should be noted that since the diagnosis title is not yet determined at this stage, no diagnosis title is input. Subject vector data are produced using both the feature quantity extracted by the feature quantity extracting unit
2
and an attribute input into the attribute input unit
4
as vector elements (step ST
101
).
FIG. 10
shows an example of subject vector data. Vector data having a high degree of resemblance to the subject vector data are retrieved from the vector database
7
by the resemblance retrieval engine
10
, employing the weight vector
13
(step ST
102
).
In other words, assuming that the dimension (namely, the number of elements) of the vector data is selected to be “n”, the subject vector data is X=(x
1
, x
2
, . . . , xn), and vector data stored in the vector database is Y=(y
1
, y
2
, . . . , yn). These data are used to calculate a degree of resemblance between the subject vector data and database vector data. The weight vector is W=(w
1
, w
2
, . . . , wn) and a resemblance degree sim(X, Y) between the vector data X and the vector data Y is calculated based on the following formula:
[
Formula
1
]
_
sim
(
X
,
Y
)
=
∑
i
=
1
n
{
wi
·
δ
(
xi
,
yi
)
}
2
,
(
1
)
where &dgr;(xi, yi) is equal to:
(xi−yi)/(section width of attribute “i”), when the attribute “i” has a continuous value;
0
, when the attribute “i” has a discrete value, and xi=yi;
and
1
, when the attribute “i” has a discrete value, and xi is not equal to yi. (2)
(It should be noted that the section width of the attribute “i” is equal to the absolute value of the difference between the maximum value of the attribute “i” and the minimum value thereof.)
In other words, the resemblance degree sim(X, Y) is equal to the weighted distances X and Y, the symbols of which are inverted.
As previously described, when all of the vector data stored in the vector database
7
are set as Y, the degree of resemblance sim(X, Y) between these vector data and subject vector data is calculated. The maximum, or highest, degree of resemblance is selected to retrieve the vector data most closely resembling the subject vector data. When several vector data have the same maximum degree of resemblance, any one of these vector data may be selected. For instance, the first selected vector data may be employed or a selection from these vector data having the maximum degree of resemblance may be made at random.
In the retrieval result display unit
11
, both a portion of the attributes of the retrieved resemblance vector data and an image corresponding to the resemblance vector data, among the images in the image database
8
, are displayed (step ST
103
).
FIG. 11
represents an example of a display screen in which six sets of resemblance images, including images, patient IDs, and diagnosis titles are displayed in the order of degree of resemblance. A user compares the subject image with the images displayed as a retrieval result. Then, the user determines which retrieved image truly resembles the subject image. There is a high possibility that the subject image is relevant
Chang Jon
Leydig , Voit & Mayer, Ltd.
Mitsubishi Denki Kabsuhiki Kaisha
LandOfFree
Resemblance retrieval apparatus, and recording medium for... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Resemblance retrieval apparatus, and recording medium for..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Resemblance retrieval apparatus, and recording medium for... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2536568