Tissue Classification Using Data from Vibroacoustic Signals Produced from Needle-Tissue Interaction

In many clinical procedures, precise needle localization is crucial for avoiding organ damage and ensuring accurate target placement. Most imaging systems come with some limitations and produce artifacts in combination with the devices that hinder accurate needle localization, particularly with respect to visualizing the tip of the needle. We propose to use vibroacoustic signals generated during needle movement through human tissue, in combination with advanced data processing and deep learning techniques for precise localization. To validate this concept, we designed a specialized phantom using animal tissues submerged in gelatine to collect the vibroacoustic data. This paper summarizes initial experiments, involving data acquisition and preprocessing, conversion into Mel and continuous wavelet transform spectrograms, as well as their use as inputs for two distinct deep learning models: NeedleNet and ResNet-34. The aim of this research was to prove that vibroacoustic signals can be used to identify tissue types during needle insertion and serve as a baseline for further investigations.Clinical relevance: These are the first phantom studies that show the usefullness of vibroacoustic signals that are generated during device-tissue interaction as a support tool for localization information.