Education and demonstration – Question or problem eliciting response
Reexamination Certificate
1999-05-13
2002-12-17
Cheng, Joe H. (Department: 3713)
Education and demonstration
Question or problem eliciting response
C434S118000, C434S262000
Reexamination Certificate
active
06494720
ABSTRACT:
DESCRIPTION
The invention relates to a method for determination by consensus analysis the capability for learning systems to classify digitised signals, for instance digitised images of objects. In addition, the invention includes the utilisation of this method in networking as illustrated in
FIG. 1
for improvement of standardisation, education and certification in classifying digitised signals.
The method is applicable to determine all type of human, recognition capabilities. Moreover, as described in claim
1
, the method is useful for any type of biological or technological, learning systems. The classifying systems may be separated units, organised as group of units, or constituting a centre of several groups of units. For exemplification, the invention is explained in relation to determination of the human capability for visual recognition of objects in digitised images.
Thus, within the healthcare sectors, large numbers of visual inspections are carried out, in particular for diagnostically and scientifically purpose. Many of these visual analyses are carried out on biological samples, which are analysed automatically by image analysis for instance digitised images recorded by videomicroscopy or scanning techniques. In many cases, the image analysis can be carried out by rule-based, digital image analysis. However, visual inspections are often necessary to be carried out on a complex, knowledge-basis by technicians, medical doctors, or other trained personnel. Obviously, it is mandatory, that a visual inspection will be carried out carefully and with sufficient experience to determine details in an image. Often, these types of images are very complex, such as images of sperm cells, cancer cells, or similar objects.
So far, more experienced personnel have educated less trained persons, who then had to carry out the visual inspection analysis on their own. Moreover, by use of known methods to determine the reproducibility in visual perception, objects in samples are classified by e.g. microscopy and morphological classification by e.g. manual count of different types of human spermatozoa. Thereby, any mistaking between different types of objects can statistically not be resolved.
As mentioned, it is extremely important, that e.g. interpretation of medical images is carried out with the greatest care. It is however well known that visual interpretations of complex objects often vary considerable between persons as reflected in high intra-person and inter-person variations.
Until now, no method has yet been developed to analyse the capabilities of persons to interpret digitised images of objects, i.e. at the object level. In particular, no method has yet been practised to determine to which extent some persons or groups of persons are special gifted for recognising special types of objects.
More specifically, the invention relates to a statistical, consensus method, to identify e.g. persons or groups of persons, who are particular, qualified to classify different types of specified objects in digitised images as defined by the classification criteria. This method is embodied by implementation of a database and a statistical software system in a computer environment. Thereby, the method is enabled both by displaying objects in digitised images, by a user-interface unambiguous to connect selected object with determined classification type, and by the simultaneous displaying of the objects and the determined types of classifications. By classifying objects in the reference images several times, thereby constructing several classification sets, these sets can then be used for calculating a consensus set.
For standardisation, education and quality assurance in classifying objects in digitised images it is therefore necessary to identify and employ persons or groups of persons, who posses a reproducible high visual perception classify objects.
A reproducible, visual perception then implies that every time, e.g. a person or groups of persons are classifying the same objects in the same reference image, the objects are classified as the same types.
The invention provides by statistical, consensus analysis so-called type-specific performance profiles, in which the percentages of recognition capability are histogram functions of the percentage of repeated classifications for each defined type of object, as described in claim
1
. Thus, as shown in
FIG. 6
, the areas in percentage under each histogram represent the type-specific object recognition, and the weighted average of all the type-specific performance profiles (WAR) represent the person's performance to recognise the specified types of objects.
As shown in FIG.
2
and in
FIG. 3
, the method of the invention, as described in claim
1
is based upon procurement and use of three main resources, which are:
Procurement of the external signals, including selection of type of objects, selection and digitalisation of reference images and marking of the objects in the reference images (
2
A,
3
A).
Procurement of the computer hardware environment and the operative system to execute the developed software for providing the information about the intellectual recognition otherwise not determinable (
2
B,
3
B).
Procurement and use of human, intellectual resources. These resources are necessary for defining classification criteria for each type of object, for selecting the learning and classifying systems, and for educating and training learning systems in use of the classification criteria (
2
C,
3
C).
To improve the determination of the visual recognition capability, the type recognising capability is further complemented by calculating the area recognising capability, as described in claim
1
. This capability to recognise object is determined by calculating for each object, the ratio of the area between the intersection and the union of the outlined objects, which is one for a perfect, outlined localisation capability (
FIG. 3
(
334
).
Provided, the same classification criteria are being practised, comparison of any combinations of classification sets can be carried out between different pairs of sets that may be between single sets, between consensus sets, or between single sets against consensus sets. This comparison can then be shown as two-dimensional histograms or so-called scatter plots as shown in FIG.
7
. Thereby, any combinations of agreements in the diagonal of the scatter plot and disagreements outside the diagonal can be visualised and located, both quantitatively and qualitatively, as described in claim
2
. In case, comparison of consensus sets are being carried out, set operations on the consensus sets like AND, OR, XOR, (A NOT B) and (B NOT A) can be carried out.
Further, using the same classification criteria, the classification performance of each person can now be ranked within the group in an individual ranking analysis by use of the individual consensus sets. This individual ranking analysis is carried out by multiplying the Sum of Performance Profiles (claim
1
) with the total Classification Agreement (claim
2
) for each individual and the group of individuals, as described in claim
3
. Thus, the classification agreements between each person and the group are calculated.
By constructing group consensus classification sets from consensus sets of individual persons, it is possible to carry out a ranking analysis of several groups. This group ranking analysis is carried out by multiplying consensus of the classification sets for each group with the consensus set of all the groups, as described in claim
4
.
An expert classification set may then be selected by the ranking procedure, as described in claim
5
.
As described in claim
6
, expert classification set may then function as filtration sets to retain all similar classified objects in the source set. Thereby, all matches of classifications between the expert sets and the filtrating set will be shown with a high reproducibility.
Single pass classifications of substructures or attributes of objects can be used to define new types of objects, by cal
Cheng Joe H.
Christman Kathleen M
LandOfFree
Methods for objectification of subjective classifications does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Methods for objectification of subjective classifications, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Methods for objectification of subjective classifications will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2932682