ABSTRACT
Medical diagnosis can be viewed as a pattern classification problem: based a set of input features the goal is to classify a patient as having a particular disorder or as not having it. Performance of medical diagnosis is typically assessed in terms of sensitivity and specificity. In this paper we introduce a pattern classification system for medical diagnosis that is based on fuzzy logic and utilises weighted training patterns. Adjusting the weights allows to focus either on sensitivity or specificity while not neglecting the other one and hence lends a degree of flexibility to the diagnostic process. A learning method is utilised that provides improved classification performance. Excellent classification results based on the University of Wisconsin breast cancer database are presented.
ABSTRACT
We introduce a method for visualising and navigating through a collection of medical infrared images. Multidimensional scaling is used to provide an overall picture of a given image database by projecting all images on a plane and arranging them so that images that are visually similar are placed close to each other. Navigation through the image set can then be performed by zooming into an area of interest which could correspond to images describing similar symptoms. Experimental results are provided on an image database of 200+ thermal images.
ABSTRACT
Several popular lossless image compression algorithms were evaluated for the application of compressing medical infrared images. Lossless JPEG, JPEG-LS,JPEG2000, PNG, and CALIC were tested on an image dataset of 380+ thermal images. The results show that JPEG-LS is the algorithm with the best performance, both in terms of compression ratio and compression speed.