AbstractThis study investigated an automated classification and diagnosis process for Doppler ultrasound blood flow signals to support clinical decision making. This process is used in this study to diagnose the severity of vascular disease in the lower limb. The study reviewed the existing methods for clinical decision making and proposed a multi-stage process of pre-processing, feature extraction and classification. Each stage was investigated separately and then combined to an automated classification and diagnosis process. Clinical Doppler ultrasound blood flow signals from the lower limb were used to compare existing methods with the novel methods proposed in this study.
The pre-processing stage normally requires the intervention of an operator to select a noise reduction threshold. The novel approach of utilising the Wavelet-based noise reduction method in this study has removed the need for human intervention. Detailed investigations of the feature extraction stage were conducted with the aim to automate and improve the algorithms. Results from these investigations showed that the algorithms proposed in this study, not only automated, but also improved the extraction of the feature vector. Artificial Neural Networks (ANNs) were used to investigate the classification of the Doppler ultrasound blood flow signals into different classes of disease severity.
The results from the investigation into the overall process have proven the full automation of the classification and diagnosis process. This study demonstrated that the introduction of the algorithms proposed in this study significantly improved the diagnosis result over traditional manual methods.
|Date of Award||May 2003|
- Blood vessels