[Show abstract][Hide abstract]ABSTRACT: This paper presents the results of using type 2 fuzzy sets to assist in the pre-processing of data for use with neuro-fuzzy clustering for classification of sports injuries in the lower leg. This research is concerned with the analysis of bone scans from stress related injuries to the tibia. Of particular interest is whether neural network based clustering techniques can help the consultant in classifying the images. The work was motivated by the situation where there is a relatively small amount of relevant data and difficulties are faced by consultants in classifying the various types of injuries. For this particular problem the consultant's interpretation of the image lends itself to representation using type 2 fuzzy sets. This research sets out to address whether, with fuzzy neuro-clustering techniques some insights may be provided to the consultant that they can use along with their experience and knowledge. The results of this approach indicate that the use of neural clustering using a type 2 representation can improve the classification of shin images.
[Show abstract][Hide abstract]ABSTRACT: This chapter presents an application of plastic (adaptive) neural network models to aid in the medical pattern application task of image classification by an expert clinician. We present a detailed description of the context and use of these neural models and show that the plasticity of the model and fuzzy representation are of particular value in this application. The neural models are considered as a model of how an expert would classify images from fuzzy linguistic and numeric descriptions.
Our model of the clinician is that, given a new image to classify, a cluster of similar images is produced (recall of similar diagnosed cases) which is then used in a further matching process to find the best match. The inference is made that the new image has the same class as the best match in the selected cluster. We also assume in our model, that the classes that the expert clinician has used in diagnoses are at least partially the result of analysing images which have been seen over a long time and in a particular order.
This model allows many experiments which determine, for example, the effects of different recall ability (associated with experience) on classification of new images. We report the results of these investigations in detail and conclude that the overall approach is successful but that further research is necessary to investigate the relative effects of the various parameters on the clinical value of the classification predictions.
[Show abstract][Hide abstract]ABSTRACT: This paper is concerned with pre-processing of data for submission
to neural networks. In particular the use of type 2 fuzzy sets to assist
in this process is discussed and the results of using type 2 sets with
FuzzyART is presented for clustering of radiographic tibia images. These
results indicate that the approach out performs a type 1 approach, for
certain tibia problems, and that the type 2 solution assisted the expert
in analysing a set of images that he was unable to classify originally
[Show abstract][Hide abstract]ABSTRACT: This paper concerns the classification analysis of exercise-induced lower leg pain by applying competitive neural network clustering and mapping techniques to type 1 and type 2 fuzzy descriptions of bone scan images of the tibia. The clusters are described and compared with each other and with the experts known classes that would be expected from medical findings. The discovered clusters provide training sets for supervised learning by an ARTMAP and similar neural network. These were used to classify the previously unclassified images and hence improve the classification process. The overall conclusion is that the use of the neural clustering methods has improved the classification process of the shin images despite the paucity of data and its inherent uncertainty.
Article · Dec 1997 · Artificial Intelligence in Medicine