Motion model ultrasound localization microscopy for preclinical and clinical multiparametric tumor characterization

Abstract

Super-resolution imaging methods promote tissue characterization beyond the spatial resolution limits of the devices and bridge the gap between histopathological analysis and non-invasive imaging. Here, we introduce motion model ultrasound localization microscopy (mULM) as an easily applicable and robust new tool to morphologically and functionally characterize fine vascular networks in tumors at super-resolution. In tumor-bearing mice and for the first time in patients, we demonstrate that within less than 1 min scan time mULM can be realized using conventional preclinical and clinical ultrasound devices. In this context, next to highly detailed images of tumor microvascularization and the reliable quantification of relative blood volume and perfusion, mULM provides multiple new functional and morphological parameters that discriminate tumors with different vascular phenotypes. Furthermore, our initial patient data indicate that mULM can be applied in a clinical ultrasound setting opening avenues for the multiparametric characterization of tumors and the assessment of therapy response.

Conflict of interest statement

The authors declare no competing interests.

Figures

Fig. 1

Motion model ultrasound localization microscopy (mULM): Principle, examples, and assessable parameters. a Sketch illustrating the principle of mULM. The filled circles mark the positions of detected MB. The red circles indicate detected MB supposed to be false alarms. The colors (blue/green) indicate the association of the MB to different tracks. One possible association of MB tracks is shown in the left diagram, another one in the right. The lighter ellipses indicate the probability density functions for the positions predicted by a linear motion model. From these, the likelihoods of the detected positions for an association are determined. The Markov Chain Monte Carlo Data Association (MCMCDA) algorithm searches for the association that maximizes the posterior probability. This also accounts for prior probabilities, like, e.g., the probability of false alarms. Taking these analyses into account, in this example, the left association is more probable than the right one. b Super-resolution ultrasound images of an A431 tumor provide detailed information on the microvascular architecture including insights into vascular connectivity and the number of vascular branching points (see arrows in magnifications). Functional information such as MB velocities (left image) and MB flow directions (right image; color-coding illustrating the direction of flow according to the colored circle) can be determined for each individual vessel and evaluated together with the morphological characteristics. Scale bar = 1 mm. c Overview of the parameter classes obtained with mULM and their accessibility with standard contrast-enhanced ultrasound methods (turquoise dot: quantitative and robust assessment of a parameter is possible; yellow dot: the information is available but its assessment is less robust, less accurate, or not quantitative; magenta dot: the parameter cannot be obtained with the respective method)

Fig. 2

mULM-based parameter maps of tumors with different vascular phenotypes. The color-coded maps indicate a the detected positions of MB overlaid on the B-mode images, representing the relative blood volume, b individual MB velocities, and c directions of MB flow. Scale bars = 1 mm. The three tumor models can be distinguished based on their different vascular patterns and quantitative textural analysis can be performed based on the super-resolution parameter maps

Fig. 3

Comparison of mULM parameters. While rBV (a), local flow direction entropy (b), and MB velocities (c) did not differ significantly between A431, MLS, and A549 tumors, the tumor models could be distinguished using the parameters of distances to the closest vessel (d), and the parameters that combined velocity and distance information, i.e., distances to vessel with low (e) and high velocities (f). Only parameters that could distinguish all three tumor models were used for further analysis. For all bar plots shown, data are expressed as the mean ± s.d. (n = 4 per tumor model; **p < 0.01; *p < 0.05 by one-way ANOVA with Bonferroni post-hoc analysis)

Fig. 4

Capability of mULM parameters to distinguish tumors with different vascular phenotypes. a Results of the inter-group comparison of all parameters using the one-way ANOVA and Bonferroni post-hoc test. Differences between parameters with p < 0.01 are highlighted in dark green. Differences with p < 0.05 are indicated in light green. Only the parameters which could discriminate all three tumor models were used to generate confusion matrices. b Confusion matrices were generated to assess the capability of the parameters to correctly assign individual tumors to their according group. The numbers in the diagonal elements of the matrix represent correct classifications (highlighted in green), the remaining numbers indicate false assignments (highlighted in pink; see explanatory example in the upper row). Confusion matrices of the maximum of distances to the closest vessel and of the maximum of distances to vessel with low velocities reveal a correct classification in all cases (100%). For the variance of distances to the closest vessel, 83% correct classification is achieved. c Although several parameters alone already allowed a correct assignment of all tumors, parameter combinations may be required when investigating more heterogeneous tumor populations. Therefore, a correlation matrix (Pearson’s correlation coefficient (r)) of all mULM parameters was generated to indicate the parameters providing complementary information. The highly discriminating distance parameters strongly correlated and thus, their combination may not be advantageous. However, the parameter local flow direction entropy showed a low correlation with the distance parameters and could be selected as a potential candidate for a multi-parameter readout

Fig. 5

Comparison of mULM parameters with reference methods. rBV values in A431, MLS, and A549 tumors were obtained by mULM (a), Maximum Intensity over Time (MIOT) US analysis (b), micro-computed tomography (µCT) (c), and immunohistochemistry (IHC) (d). Scale bars = 1 mm. All methods show a similar trend, with A431 tumors having the highest and A549 tumors the lowest level of vascularization. While MIOT clearly overestimates the rBV, µCT and mULM provide comparable values, which are in line with the data from histology (e). f Mean MB velocity values were either obtained from an exponential fit of a MB replenishment curve after a destructive US pulse from a ROI covering the entire tumor or from the mULM velocity maps. Both postprocessing procedures indicate that there are no significant differences in mean MB velocities between the tumor models. However, the mean velocity values are significantly lower in the replenishment analysis than mean velocities calculated by mULM. In g distance parameters determined by IHC analysis in the tumors with different vascular phenotypes are shown. Mean, variance, and maximum of the distance to the closest vessel determined by mULM had the same trend as their counterparts determined by IHC. For all bar plots shown, data are expressed as the mean ± s.d. (n = 4 per tumor model; *p < 0.05, **p < 0.01 by one-way ANOVA with Bonferroni post-hoc analysis)

Fig. 6

Preliminary mULM results from breast cancer patients. CEUS measurements were performed with a conventional US device and phospholipid MB. Scale bars = 10 mm. In a B-mode and mULM images of the patient with the HER2 positive breast carcinoma are shown. mULM visualizes in detail the tumor vascular pedicle on the bottom side of the tumor that branches into smaller vessels which are distributed heterogeneously throughout the tumor, slightly denser on the right-hand side. In b B-mode and mULM images of a triple negative breast carcinoma in a patient treated with neoadjuvant chemotherapy are presented. Measurement were performed before (first row), and after the first (second row), the second (third row), and the third cycle of chemotherapy (fourth row). The first column shows B-mode images with the tumor borders highlighted by the red polygon, the second column displays the mULM velocity maps and the third column indicates the mULM direction maps. At the baseline measurement, the tumor vascularization was present mainly in the peripheral areas, and only modestly in the tumor core without showing any dominant direction. After the first cycle of treatment, the tumor shrinked and vascularization appeared more homogeneous with only one avascular part at the bottom side of the tumor. After the second and third cycle of treatment, the tumor volume further decreased, while the level of vascularization remained stable