| Diagnosis of lung diseases using lung sounds has the distinct advantages of being noninvasive, nonhazardous, inexpensive, convenient. However, it is often difficult to reach useful diagnosis based on lung sounds only due to the inability to extract enough information. There have been some efforts in past decade to improve this and more recently, signal processing methods to purify and extract information from lung sounds are being investigated. This dissertation is one such effort.;The first problem addressed is that of high fidelity acquisition. Then, the issue eliminating interferences that persist the first stage of acquisition is addressed. Heart sounds interference creates some special problems since they are quite large in amplitude compared to lung sounds and have considerable overlap of spectral information with lung sounds. A solution for this problem is developed using adaptive filtering concepts.;Having obtained 'pure' lung sounds, the next issue addressed is that of detecting abnormalities in the lung using lung sounds. Two advanced high resolution signal processing methods--Wigner time-frequency representation based crackle detection and maximum entropy spectrum based wheeze detection--are proposed. Also, spectral methods for early detection of pulmonary consolidation are discussed.;The next issue addressed is the modeling and characterization of the lung sounds source and transmission characteristics. Autoregressive signal modeling techniques are applied for this, and it is shown by examples that while the prediction error characterizes the source, the filter coefficients (autoregressive coefficients) characterize the transmission.;The dissertation first outlines the field of lung sounds and research work done in the area, and the inherent problems that hamper further development using conventional approaches. Based on this perspective, the second chapter states the specific problems addressed in this dissertation.;Results from two actual experiments, using the abovementioned techniques for analysis are then presented. The first experiment consists of sounds recorded from normal subjects along with lung volume and ECG signals. The second experiment consists of inducing pulmonary edema in mongrel dogs by intravenous infusion of Ringer lactate solution.;It is concluded from this dissertation research that signal processing applied to lung sounds can lead to very useful advances in the use of noninvasively acquirable lung sounds for diagnosis. |
