Audio-based tissue classification - preliminary investigation for a needle procedure

Abstract

Image-guided and minimally invasive procedures still require confirmation on having reached a target. Intraoperative imaging is not always sufficient or conclusive as it comes with artifacts that can come with a certain amount of ambiguity and inaccurate location information. As an alternative to imaging, we want to explore sounds produced by the biopsy needle tip while advancing and interacting with tissue. In this paper, we show that by analyzing vibroacoustic signals acquired at the proximal end of the needle we are able to differentiate the tissue type. In total, 419 audio samples of 5 tissues were acquired and converted to spectrograms used as input to a convolutional neural network. Using this experimental setup we were able to differentiate the tissue types with an F1 score of 71.64%. Based on these results we were able to demonstrate the feasibility of our approach, as well as the importance of further experiments to ensure that vibroacoustic sounds produced by the needle tip can be a new navigation method.

Keywords

• vibroacoustic signal processing
• convolutional neural network
• tissue characterization
• interventional therapy
• audio guidance
• minimal-invasive procedures