Some New Results in Nearest Neighbour Classification and Lung Sound Analysis

[摘要] This thesis describes the results of the author's contribution to a collaborative research programme between the Department of Respiratory Medicine, Glasgow Royal Infirmary, and the Department of Electronics and Electrical Engineering, University of Glasgow. After the first six months of research, it was decided that the project would be limited to demonstrating the feasibility of developing a non-invasive examination system for patients exposed to asbestos dusts. The development of this system led to a growing interest in nearest neighbour (NN) classification algorithms and to the investigation of a number of interesting problems in this area. In particular, it is argued that when the size of the prototype set is finite, a weighted k-NN rule may, in some cases, has a smaller probability of error than an unweighted k-NN rule. Analytical solutions to a simple example and experimental results are presented to support this argument. A new NN classification scheme is also described in this thesis. Again a number of modifications for the case of a finite prototype set are suggested, and experimental results are given. The remainder of the work in this thesis concerns the development of the proposed non-invasive examination system which uses lung sound as its input. Due to the complexity of the proposed system and the initial small data set, only part of the system has been implemented. However, preliminary experimental results on three groups of patients, namely (a) patients with asbestosis, (b) patients who have exposed to asbestos dust, and (c) healthy subjects, have shown that it is possible to discriminate these three groups of patients. More extensive studies are required before the system can be used in clinical conditions. Suggestions for these continuing studies are made.

[发布机构] University:University of Glasgow

[关键词] Medicine, Biomedical engineering