Professor Dosik Hwang, from Yonsei University in Korea, talks about the work behind the paper ‘Murmur-adaptive compression technique for phonocardiogram signals’, page 183. Professor Dosik Hwang Our field of research is focused on 1) biomedical signal processing, such as electro-/phono- cardiograms; and 2) tomographic imaging, such as magnetic resonance imaging (MRI), computed tomography (CT), and ultrasound (US). Our mission is to help human health by developing practical algorithms for processing human vital signals and by advancing novel medical imaging technologies. Examples include noise removal in electrocardiograms (ECG) and phonocardiograms (PCG), unifying frameworks of ECG and PCG, robust feature extractions of ECG/PCG, denoising of MR images, multi-contrast MRIs, fast acquisition of MRI data and reconstruction, susceptibility-weighted MR imaging, brain tomographic imaging, artefact corrections in CT, and motion analysis in US. The initial vision of making contributions to humans led me to keep pursuing this field of research. Mobile monitoring of human vital signals – as mobile and wearable devices evolve, health technologies are being actively incorporated into these devices enabling the constant monitoring of human vital signals and subsequent diagnosis based on it. However, the application of conventional health technologies into mobile devices needs rigorous signal processing techniques, since the data acquisitions are less optimal than those of the conventional standard devices in regular hospitals. Special techniques for denoising, robust feature extraction, data compression, and novel multi-modal signal processing will expedite the successful applications of the current health technologies into the mobile world. I am also interested in multi-contrast medical imaging. As the current medical imaging systems are advanced to produce multiple contrast images, novel techniques to handle these multiple datasets need to be developed with new insights for better image quality, better diagnostic outcome, and better imaging efficiency. We are actively involved in these special applications. We have reported an effective way of compression for the phonocardiogram signals (PCG) emphasising their diagnostic purpose. In the several previous attempts to compress PCG, their main interest of compression was to reduce the data size while maintaining a certain level of overall sound quality without considering a diagnostic purpose. The natural consequence of it was to lose some of valuable diagnostic information when compressed. As an example of this study, the murmur (noise-like abnormal heart sound) information in PCG is diagnostically important, but conventional compression techniques would treat some of this murmur sound as noise, and distort it during compression. We had successfully incorporated a murmur estimate process during the compression and were able to adaptively compress PCG without significant loss of the murmur information in the compressed PCG. The effectiveness of this technique was validated in quantitative metrics and in subjective quality assessment as well. Two main challenges were 1) effective incorporation of the murmur estimate process into the overall compression procedure, and 2) evaluation of the compressed PCG. In order to seamlessly incorporate the murmur estimate in the compression, we developed a murmur estimate technique on the several levels of wavelet decomposition and combined it with the wavelet compression technique, which resulted in the murmur-adaptive compression performance. The conventional evaluation method for compressed sounds was not effective to evaluate the murmur-containing PCGs. In order to evaluate the quality of the murmur-related sound in the compressed PCG, we adopted both the quantitative metrics such as the compression ratio (CR) and the percent root-mean-square difference (PRD) and a subjective assessment method, MUSHRA (multi-stimulus test with hidden reference and anchor). We are working on how to combine multi-modal vital signals acquired at the same time in terms of signal acquisition, noise removal, feature extraction, signal enhancement, and diagnosis. In addition, we are also actively working on multi-contrast and multi-modal medical imagings, which will be eventually combined with the vital signals for thorough analysis and guidance for health. This field of research will help accumulate massive diagnostic information on a specific person and the general population as well, which will significantly contribute to accurate diagnosis, effective monitoring of health, and guiding personal health life. The exciting challenges would be how to incorporate multi-modal vital signals to fulfil these purposes.
Read full abstract