Model-based object recognition is a well-known task in Computer Vision. Usually, one object that can be generalized by a model should be detected in an image based on this model. Biomedical applications have the special quality that one object can have a great variation in appearance. Therefore the appearance of this object cannot be generalized by one model. A set of cases of the appearance of this object (sometimes 50 cases or more) is necessary to detect this object in an image. The recognition method is rather a case-based object recognition than a model-based object recognition. Case-based object recognition is a challenging task. It puts special requirements to the similarity measure and needs a matching algorithm that can work fast on a large number of cases. In this paper we describe the chosen case representation, the similarity measure and the recent matching algorithm. Finally, we give results on the performance of the system.
This work focuses on case-based reasoning in domains where cases have complex structures with relationships to an arbitrary number of other (potentially complex and structured) entities and where case characterisations (queries) are potentially incomplete. We summarise the requirements for such domains in terms of case representation and retrieval functions. We then analyse properties of existing similarity measures used in CBR – above all symmetry – and argue that some of these properties are not desirable. By exploiting analogies with retrieval functions in the area of information retrieval – where similar functions have been replaced by new ones not exhibiting the aforementioned undesired properties – we derive a new asymmetric ranking function for case retrieval. On a generated test-bed, we show that indeed the new function results in different ranking of cases – and use testbed examples to illustrate why this is desirable from a user’s perspective.
Different spectrometer methods exist that have been developed over time to practical applicable systems. Researchers in different fields try to apply these methods to different applications especially in the chemical and biological area. One of these methods is RAMAN spectroscopy for protein crystallization or Mid-Infrared spectroscopy for biomass identification. For the applications are required robust and machine learnable automatic signal interpretation methods. These methods should take into account that not so much spectrometer data about the application are available from scratch and that these data need to be learnt while using the spectrometer system. We propose to represent the spectrometer signal by a sequence of 0/1 characters obtained from a specific Delta Modulator. This prevents us from a particular symbolic description of peaks and background. The interpretation of the spectrometer signal is done by searching for a similar signal in a constantly increasing data base. The comparison between the two sequences is done based on a syntactic similarity measure. We describe in this paper how the signal representation is obtained by Delta Modulation, the similarity measure for the comparison of the signals and give results for searching the data base.
Case-based reasoning, image or data retrieval is based on similarity determination between the actual case and the cases in a database. It is preferable to normalize the similarity values between 0 and 1 in order to be able to compare different similarity values based on a scale. Similarity is thus imparted with a semantic meaning. The main problem arises when the case base is not yet complete and contains only a small number of cases while the other cases are collected incrementally as soon as they arrive in the system. In this case the upper and lower bounds of the feature values cannot be estimated close to the real values. This paper concerns possible methods for predicting the upper and lower bounds of a feature value and the problems that arise when these values are not correctly estimated due to a limited number of samples or a parameter distribution that is not available a-priori. The aim is to develop a method for learning the upper and the lower bounds of a feature value and to develop a methodology for dealing with change in semantic meaning of the similarity.
Sea reclamation works and fish farming are increasingly common in coastal zones, and how to accurately and rapidly extract the coastal aquaculture area is important for the development of coastal zones. This paper discusses a CBR (case-based reasoning) method. Firstly, using a 10 meter resolution of a multi-spectral remote sensing image of Eastern Guangdong over ten thousands of spatial, spectral, shape and texture features were extracted based on the 1:50000 standard framing of land use thematic data and by using image analy- sis. Then nine optimized features were selected using the principal component analysis (PCA) and the construction of a case base was accomplished based on these. After that, a multi-scale image segmentation was performed on the 2.5 meter resolution of a fused image of the test area, which is located on the West- ern Guangdong coast, and CBR classification was applied on all the segmented image objects. In the end, the classification accuracy was evaluated. The CBR classifier classifies an aquaculture area within coastal belts with an accuracy of 84.6 %, in contrast to that the accuracy of the Maximum Likelihood Classifier (MLC) is 82.5 %. The CBR method outperforms the MLC by 2.2 % in predic- tion accuracy. The advantages of the CBR approach are obvious, particularly in the areas that are far away from the coastlines. In conclusion, the CBR approach could be successfully applied to the extraction of coastal aquaculture areas.
So far, Case-Based Reasoning has not become as successful in medicine as in some other application domains. One, probably the main reason is the adaptation problem. In Case-Based Reasoning the adaptation task still is domain dependent und usually requires specific adaptation rules. Furthermore, in medicine adaptation is often more difficult than in other domains, because usually more and complex features have to be considered. We have developed some programs for endocrine therapy support, especially for hypothyroidism. In this paper, we do not present them in detail, but focus on adaptation. We do not only summarise experiences with adaptation in medicine, but we want to elaborate typical medical adaptation problems and hope to indicate possibilities how to solve them.
The importance of learning distance functions is gradually being acknowledged by the machine learning community, and different techniques are suggested that can successfully learn a strong distance function in many various contexts. Nevertheless the studies in the area are still rather fragmentary; they lack systematic analysis and focus on a limited circle of application domains. In this paper, two techniques for learning discriminative distance function are evaluated and compared on biomedical data of different kind; learning from equivalence constraints and the intrinsic Random Forest similarity. Both techniques demonstrate competitive results with respect to plain learning; the Random Forest similarity exhibits a more robust behaviour and is shown to be less susceptible to missing data and noise.
Clinical research significantly contributes to the better understanding of diseases and their treatment. Physicians, however, could not take fully advantage of the resulting scientific knowledge. To improve sharing and processing of primary clinical data new generation technologies for knowledge management are needed. In this paper we discuss some of their features on the basis of a real-world system in the domain Nephropathotology. We propose several interpretations of user-aware behavior and show how they influence the understanding and assessing of the similarity in CBR theory. Moreover, we illustrate our point with several examples, inspired by the experience, gained during the development of a CBR-based search engine within the scope of the OpEN.SC project.
So far, Case-Based Reasoning has not become as successful in medicine as in some other application domains. One, probably the main reason is the adaptation problem. In Case-Based Reasoning the adaptation task still is domain dependent und usually requires specific adaptation rules. Furthermore, in medicine adaptation is often more difficult than in other domains, because usually more and complex features have to be considered. We have developed some programs for endocrine therapy support, especially for hypothyroidism. In this paper, we do not present them in detail, but focus on adaptation. We do not only summarise experiences with adaptation in medicine, but we want to elaborate typical medical adaptation problems and hope to indicate possibilities how to solve them.
Disorders of the heart valves constitute a considerable health problem and often require surgical intervention. Recently various approaches were published seeking to overcome the shortcomings of current clinical practice,that still relies on manually performed measurements for performance assessment. Clinical decisions are still based on generic information from clinical guidelines and publications and personal experience of clinicians. We present a framework for retrieval and decision support using learning based discriminative distance functions and visualization of patient similarity with relative neighborhood graphsbased on shape and derived features. We considered two learning based techniques, namely learning from equivalence constraints and the intrinsic Random Forest distance. The generic approach enables for learning arbitrary user-defined concepts of similarity depending on the application. This is demonstrated with the proposed applications, including automated diagnosis and interventional suitability classification, where classification rates of up to 88.9% and 85.9% could be observed on a set of valve models from 288 and 102 patients respectively.
Many image analysis methods need a lot of parameters that have to be adjusted to the particular image in order to achieve the best results. Therefore, methods for parameter learning are required that can assist a system developer in building a model. This task is usually called meta-learning. One problem in meta-learning is to describe the properties of the input so that it can be properly mapped to the parameters. In this paper, we consider this task for image segmentation based on the watershed transformation. We use CaseBased Reasoning to control the parameter selection process. Our previous investigation on the theoretical and implementation aspects of the watershed transformation allowed us to draw conclusions for suitable image descriptions. Four different descriptions have been considered based on: statistical and texture features; marginal distributions of columns, rows, and diagonals; similarity between the regional minima; and central moments. The two descriptions based on statistical and texture features and on central moments resulted to be the best ones for segmentation based on watershed transformation. They can best separate the cases into groups having the same segmentation parameters and work nicely also for rotated and rescaled images.
The notion of distance is the most important basis for classification. This is especially true for unsupervised learning, i.e. clustering, since there is no validation mechanism by means of objects of known groups. But also for supervised learning standard distances often do not lead to appropriate results. For every individual problem the adequate distance is to be decided upon. This is demonstrated by means of three practical examples from very different application areas, namely social science, music science, and production economics. In social science, clustering is applied to spatial regions with very irregular borders. Then adequate spatial distances may have to be taken into account for clustering. In statistical musicology the main problem is often to find an adequate transformation of the input time series as an adequate basis for distance definition. Also, local modelling is proposed in order to account for different subpopulations, e.g. instruments. In production economics often many quality criteria have to be taken into account with very different scaling. In order to find a compromise optimum classification, this leads to a pre-transformation onto the same scale, called desirability.
In this paper, a novel method for diagnosing lymphatic tissue tumors is presented. Microscopic specimen images are analyzed for extracting and characterizing malignant cells. A case-based reasoning approach is followed for classifying morphologic and densitometric cell features so as to provide a final diagnosis.
In medical practice and in knowledge-based systems too, it is necessary to consider exceptions and to deal with them appropriately. In this paper, a system is presented, which helps to explain cases that do not fit to a theoretical hypothesis. It is proposed to combine Case-Based Reasoning with a statistical model, where Case-Based Reasoning is used to explain the exceptional cases. Additionally, a method to partly solve the missing data problem was developed. This method combines general restoration techniques with domain dependent formulas provided by an expert. For the latter technique, Case-Based Reasoning is applied again.
This paper describes an approach to detect risks of procurement fraud. It was developed within the context of a European Union project on fraud prevention. Procurement fraud is a special kind of fraud that occurs when employees cheat on their own employers by executing or triggering bogus payments. The approach presented here is based on the idea to learn a similarity measure that compares an employee (or payroll) standing-data record to a creditor record, in order to detect creditors that are suspiciously similar to employees. To this ends, it combines several simple similarity measures like address similarity or spatial similarity using a weighting scheme. The weights, that is the overall similarity function, are learned from user input specifying whether a particular pair of payroll and creditor data records are similar. This leads to an adaptive, easily transferable approach for a generic class of fraud opportunities.
Intelligent analysis of heterogeneous data and information sources for efficient decision support presents an interesting yet challenging task in clinical environments. This is particularly the cas ...
Case-Based Reasoning (CBR) is a methodology that reuses the solutions of previous similar problem to solve new problems. Adaptation is one of the most difficult parts of CBR cycle, especially, when the solution space with multi-dimension. This paper discusses the adaptation of high dimensional solution space and proposes a possible approach for it. Visualisation induced Self Organising Map (ViSOM) is used to map the problem space and solution space first, then a BackPropagation (BP) network is applied to analyse the relations between these two maps. A simple military scenario is used as case study for evaluation.
