Quantification of Histotripsy Dosage Using Machine Learning-Enhanced Ultrasound Imaging Analysis: Correlation with...

Histotripsy is a noninvasive ultrasound therapy that mechanically disrupts target tissue through controlled acoustic cavitation. Clinically, a fixed histotripsy dose (number of pulses) is used for treating liver tumors, which does not account for tumor heterogeneity. Since histotripsy-induced damage can vary based on the tumor's mechanical properties, there is a clinical need for a reliable and noninvasive method to measure the extent of cellular damage. Here, we present a quantitative, noninvasive, and image-based approach to evaluate histotripsy-induced tumor cellular damage by combining ultrasound texture analysis with machine learning and correlating these results with histology.

Immunocompetent rats (n = 20) bearing orthotopic liver tumors were treated with varying histotripsy doses (20, 50, 100, and 200 pulses per location), covering the spectrum from sparse treatment to overtreatment. Pre- and post-histotripsy B-mode ultrasound images were obtained, and texture analysis was performed followed by feature reduction. Tumors were harvested post-treatment to quantify cellular damage (area covered by nuclear debris, intact nuclei, and necrosis scoring) via histology.

Strong correlation was observed between area covered by nuclear debris and first-order kurtosis (R² = 0.9) and autocorrelation (R² = 0.8). A random forest classifier trained on these features produced dose predictions that closely matched the administered values (R² = 0.9, p < 0.05).

Our results show that subtle texture changes in post-treatment vs. pre-treatment ultrasound images can serve as reliable indicators of dose-dependent tumor cellular disruption generated by histotripsy, as evidenced by their strong correlations with histological analysis. Moving forward, integrating real-time quantitative imaging feedback into clinical practice could help clinicians tailor histotripsy dosing more precisely for each patient.