3D multiparametric ultrasound of spontaneous murine tumors for non-invasive tumor characterization

<i>Objective.</i>Non-invasive tumor diagnosis and characterization is limited today by the cost and availability of state of the art imaging techniques. Thanks to recent developments, ultrasound (US) imaging can now provide quantitative volumetric maps of different tissue characteristics. This study applied the first fully concurrent 3D ultrasound imaging set-up including B-mode imaging, shear wave elastography (SWE), tissue structure imaging with backscatter tensor imaging (BTI), vascular mapping with ultrasensitive Doppler (uDoppler) and ultrasound localization microscopy (ULM)<i>in-vivo</i>. Subsequent analysis aimed to evaluate its benefits for non-invasive tumor diagnosis.<i>Approach.</i>A total of 26 PyMT-MMTV transgenic mice and 6 control mice were imaged weekly during tumor growth. First-order statistics and radiomic features were extracted from the quantitative maps obtained, and used to build predictive models differentiating healthy from cancerous mammary pads. Imaging features were also compared to histology obtained the last week of imaging.<i>Main results.</i>High quality co-registered quantitative maps were obtained, for which SWE speed, BTI tissue organization, ULM blood vessel count and uDoppler blood vessel density were correlated with histopathology. Significant changes in uDoppler sensitivity and BTI tissue structure were measured during tumor evolution. Predictive models inferring the cancerous state from the multiparametric imaging reached 99% accuracy, and focused mainly on radiomics measures of the BTI maps.<i>Significance.</i>This work indicates the relevance of a multiparametric characterization of lesions, and highlights the strong predictive power of BTI-derived parameters for differentiating tumors from healthy tissue, both before and after the tumor can be detected by palpation.