Improving automatic segmentation of tissue-targeted nanoparticles on echographic images

In recent years, intensive investigations have been undertaken to develop nanoparticle-based cancer targeting agents for various imaging modalities, including ultrasound. Thus, diagnostic paradigms are needed to correctly detect the presence of nanoparticles (NPs) in the anatomic districts. Furthermore, it would be desirable to have algorithms for the automatic recognition of areas where NPs are localized. In this work an experimental optimization of an algorithm for automatic segmentation of nanoparticle-containing tissues is presented and is based on time-frequency processing of the radiofrequency (RF) signals derived from conventional echographic acquisitions. The employed prototypal software (RULES, Radiofrequency Ultrasonic Local Estimator, developed by ELEN SpA, Florence, Italy) correlates spectral parameters to the mechanical and physical properties of the object examined. The effectiveness of the algorithm was evaluated for different configurations of the spectral parameters and tested for different NP size (330 and 660 nm). Accuracy of the algorithm has been quantified through two parameters: sensitivity and specificity. Specifically, the possibility to improve selective identification of NPs disperse in tissue-mimicking layer was investigated. Through subsequent refinement, the most promising results were obtained with algorithm parameter configuration for 330-nm nanoparticles. In particular, it was found that an increase in sensitivity up to 13.9% (from 63% to 76.9%) is achievable by accepting a decrease of 1.5% in specificity (from 99.6 % to 98.1%).