Spectral analysis of ultrasound r.f. image data to monitor cavitation and thermal bubble formation in HIFU treatment

The high acoustic intensity in HIFU often results in bubble production, either through acoustic cavitation or boiling. Both these are believed to be a primary contributor to tissue necrosis. Some HIFU protocols rely on the evidence of cavitation as a strong indicator of tissue lesions. Cavitation is normally associated with hyperechoic regions (dasiabright uppsila) in ultrasound B-mode image feedback but other events may cause hyperechoicity as well, so spectral detection methods are becoming of increasing interest in early and robust detection of cavitation activity. It is also useful to distinguish it from thermal generation of bubbles. To estimate the spectrum whilst still retaining good spatial localization requires methods which work on short time series of data, so methods based on the Fourier transform are not applicable,. The method in this paper uses parametric statistical methods, (collectively described as ARMA models) to analyse the spectral data at high spatial resolution so local changes can be more easily identified. It applies an adaptive method which estimates both the best model type and order. The method is assessed using a simulation of data from cavitation, and the applied to detect cavitation following HIFU in ex-vivo porcine liver. Thermal bubble generation is monitored using a lower frequency hydrophone.