Super‐resolution ultrasound localization microscopy for the non‐invasive imaging of human testicular microcirculation and its differential diagnosis role in male infertility

Testicular microcirculation is closely related to spermatogenic function and seminiferous tubular function. The diagnosis and monitoring of testicular diseases can be associated with testicular microcirculation; however, there are currently no effective non‐invasive methods for super‐resolution imaging of testicular microcirculation. In this study, we introduced state‐of‐the‐art graph‐based tracking with the Kalman motion model algorithm to non‐invasively image human testicular microcirculation for the first time with a regular frame‐rate clinical ultrasound imaging system (37 Hz). Two distinct testicular vessels with an 81 µm separation were resolved in the testicular vasculature, surpassing all other imaging modalities. In a retrospective study, we performed contrast‐enhanced ultrasound examinations(CEUS) and ultrasound localization microscopy (ULM) processing on the included 76 infertile patients and 15 healthy controls from August 2021 to May 2023 and obtained super‐resolution images of testicular microcirculation with sub‐diffraction resolution. Through the results of one‐way analysis of variance tests and receiver operating characteristic analyses, we found that the ULM‐based parameters hold promise as clinical guidance for differentiating between non‐obstructive and obstructive male infertility. The mean vessel diameter achieved an area under the curve (AUC) of 0.920 (95% confidence interval [CI]: 0.847–0.994, p < .001) with a cut‐off value of 170.9 µm in oligoasthenospermia, and an AUC of 0.952 (95% CI: 0.875–1.000, p < .001) with a cut‐off value of 169.9 µm in azoospermia patients, respectively, addressing a significant clinical challenge.


INTRODUCTION
Azoospermia or oligoasthenospermia in the male partner (male infertility) is one of the most common causes of infertility and affects an estimated 30 million men worldwide. [1]Based on the etiologies, azoospermia or oligoasthenospermia can be divided into two categories: one is the non-obstructive cause with impaired spermatogenesis within the testes, and the other is the obstructive cause with impaired communication of sperm from the testes into the ejaculate.It is clinically important to distinguish between them because the management of obstructive and non-obstructive causes is completely different.Sperm abnormalities caused by obstruction may be amenable to surgical or interventional correction, while assisted reproductive techniques such as in vitro fertilization/intracytoplasmic sperm injection are suitable for non-obstructive patients.Despite its importance, it is clinically challenging to distinguish obstructive from non-obstructive infertility in 60% of infertile patients, especially when follicle-stimulating hormone (FSH) levels are not unequivocally elevated, in which diagnostic testicular biopsy is indicated for differential diagnosis.However, a biopsy is potentially harmful due to its invasiveness and can lead to complications such as bleeding, wound infection, persistent pain, and possibly permanent damage to the testis; developing a non-invasive method could help select the most effective treatments for male infertility within individual men who have been affected. [2]everal non-invasive imaging modalities are available to evaluate infertile patients.Ultrasound and transrectal ultrasound (TRUS) have conventionally been used as firstline diagnostic modalities for the evaluation of distal male reproductive system obstruction.Additionally, ultrasound provides information about testicular volume, texture, and elasticity for better differentiation. [3]1a,4] Ultrasound localization microscopy (ULM), also known as super-resolution ultrasound (SRUS) imaging, or ultrasound super-resolution microcirculation imaging (USRmi), is an emerging technology that was inspired by optical super-resolution imaging techniques.ULM detects and tracks the movement of individual microbubbles (MBs), enabling it to bypass the diffraction limit and the inherent compromise between resolution and penetration in conventional ultrasound imaging.ULM can produce super-resolution images at the scale of micrometers with typically 10× better spatial resolution than that of contrast-enhanced ultrasound (CEUS). [5]Moreover, the principle of localizing and tracking the displacement of MBs makes it possible for ULM to extract and obtain reliable quantitative information for subsequent calculation.
Testicular blood supply is closely associated with spermatogenesis and seminiferous tubular function.Insights into testicular microcirculation hold the potential for diagnosing and monitoring testicular diseases. [6]The most commonly used CEUS is limited in its ability to evaluate testicular microcirculation due to spatial resolution constraints.Up till now, there has been significant in vivo experimental research on ULM reported, including both animal and human studies. [7]However, there is currently no relevant research on the application of ULM in testicular microcirculation imaging.Consequently, there is no effective way to validate the relationship between testicular microcirculation and testicular diseases.No previous research has been conducted from the perspective of testicular microcirculation to distinguish between obstructive and non-obstructive cases.Therefore, this study presents the first application of ULM in human testicular imaging.We hypothesized the feasibility of imaging human testicular microcirculation using ULM and further evaluated its diagnostic value in the differential diagnosis of non-obstructive and obstructive male infertility.

Structural imaging of testicular vasculature at super-resolution with ULM
Figure 1 illustrates the microvascular architecture of the testis of a normozoospermia subject.Figure 1A,B provides functional information about the direction of MB movement and flow velocity of individual MBs.A maximumintensity projection image offered a rough assessment of the overall vascular morphology (Figure 1C).A map of super-localized MBs indicated the superposition of the positions of detected MBs, corresponding to the relative blood volume (Figure 1D).A higher magnification of the same rectangular region illustrated a significant improvement in the resolution of ULM technology over CEUS.A branched vessel profile was clearly identified and distinguished from another branched vessel in the super-resolution image (Figure 1F).In contrast, it is impossible to identify a distinct vascular structure in the same magnified region in the CEUS image (Figure 1E).From the image intensity cross-sectional profiles, apparent microvessels that were 81 µm apart can be clearly resolved in the ULM image but were not separable in the CEUS image, and the vessel we showed was measured to be 76 µm (Figure 1G).The superiority of superresolution images in demonstrating testicular microcirculation was evident when compared with other imaging methods.Compared to conventional B-mode ultrasound, color Doppler flow imaging (CDFI), CEUS, and magnetic resonance imaging (MRI), ULM provided detailed information about testicular microvasculature, including branching and connections between microvessels (Figure 2).
One patient was excluded due to recent exogenous hormone administration.The two undefined patients, with normal or elevated FSH levels, absence of signs of obstruction, and absence of genetic abnormalities, underwent testicular sperm aspiration (TESA) for confirmation, and both were ultimately diagnosed with NOA (Figure S1).
In NOO patients, four patients had microlithiasis.In OO patients, one patient had microlithiasis.In NOA patients, one patient had microlithiasis, and two exhibited tumorlike proliferations of Leydig cells after biopsy.In OA patients, one patient had microlithiasis.
One patient in the OO group presented with unilateral testicular obstruction, while one patient in the OA group underwent surgical removal of the right testis.Patient clinical characteristics are summarized (Table 1).When comparing the testicular volumes in each patient group to the control group, it was observed that only in the NOA group, the left (p < .05)and right (p < .01)testicular volumes were significantly smaller than those in the control group.All other patient groups showed no statistically significant differences compared to the control group.

2.3
Vascular architecture visualization of testes in NOO, OO, NOA, OA, and N using ULM With the application of ULM, structural and functional imaging of testicular vasculature in different groups were visualized in fine detail.The vasculature of the normozoospermia subject was characterized by abundant blood supply, with blood vessels extending and elongating regularly and continuously.The testicular vasculature formed an extensive branched, interconnected, and dense meshwork of microvessels that distributed homogeneously throughout the entire parenchyma (Figure 4A,B).Different testicular parenchymal regions demonstrated either fast or slow blood flow velocities, and the overall testicular blood velocity in the normozoospermia group was higher than all the other patient groups (Figure 4C).When comparing NOO to OO, it can be observed that OO displayed several larger blood vessels in the central part, characterized by greater thickness, and the relative blood volume of OO was higher than that of NOO (Figure 4A,B).Maps of super-localized MBs with arriving times in NOO demonstrated certain black areas (Figure 4E).OO was characterized by higher entropies of blood flow than NOO, as manifested by the angle of MB flow in blood vessels (Figure 4D).When comparing NOA to OA, a smaller blood vessel diameter was observed in NOA, along with a decrease in the extent of homogenous vascularization in the testicular parenchyma.The microvascular meshwork contained fewer branches and connections.NOA was characterized by zonal differences in vessel density reflecting inferior or superior perfusion instead of homogenous testicular vascularization (Figure 4A,B).More black areas were observed in NOA (Figure 4E).OA was characterized by higher entropies of blood flow than NOA (Figure 4D).

DISCUSSION
Up till now, there has already been much ULM in vivo experimental research reported, including both animal and human studies. [7]Nevertheless, to our knowledge, it is the first time ULM has been applied in human testicular imaging.The data were acquired with a widely available clinical ultrasound imaging system without high frame rates.Conventionally, high frame rates are generally required to track MB movements in ULM imaging but are not available for most of the clinical ultrasound imaging systems.Therefore, the start-of-art graph-based tracking with the Kalman motion model algorithm was introduced in this study to deal with the limited frame rate of our clinical machine.The accuracy of the tracking algorithm has been evaluated at a frame rate of 20 Hz. [9] In this study, we demonstrated the feasibility of imaging human testicular microvasculature using ULM and evaluated its clinical value in the differential diagnosis of obstructive and non-obstructive male infertility.We achieved a spatial resolution sufficient to resolve testicular microvessels 81 µm apart, exceeding all the other existing imaging methods.Through the super-resolution images generated by ULM, we have excavated more quantitative and qualitative information about testicular microcirculation.
To investigate the clinical utility of ULM, we recruited azoospermia and oligoasthenospermia patients, characterized their features, and observed a significant reduction in testicular volume among NOA patients compared to the control group, consistent with findings reported in the literature. [10]Then, we conducted an analysis using ULM imaging technology.6a] Through one-way ANOVA tests, we found that the infertile patients were characterized by decreased blood flow velocity and increased vessel tortuosity compared to the controls, consistence with the previous histologic descriptions by Kumamoto [11] and De Ia Baize. [12]They reported the sclerosis and increased tortuosity of intertubular capillaries in infertile patients.Additionally, obstructive patients were characterized by a larger mean vessel diameter and higher vascular density than non-obstructive patients.NOO also exhibited higher mean vessel tortuosity than OO.We further verified the discrimination performance of obstruction and non-obstruction using these ULM-derived parameters through ROC analyses.Mean vessel tortuosity, mean diameter, and vascular density exhibited great diagnostic value in NOO and OO.The mean diameter and vascular density exhibited great diagnostic value in NOA and OA.As the best-performing indicator, mean vessel diameter revealed a remarkable ability to discriminate non-obstructive from obstructive causes.Strong diagnostic performances were exhibited with an AUC of 0.920 (95% CI: 0.847-0.994,p < .001) in oligoasthenospermia and 0.952 (95% CI: 0.875-1.000,p < .001) in azoospermia patients, respectively.Vascular density is another great indicator to discriminate obstruction and non-obstruction, with an AUC of 0.852 (95% CI: 0.743-0.961,p < .001) in oligoasthenospermia and 0.882 (95% CI: 0.759-1.000,p < .001) in azoospermia patients, respectively.
The results showed encouraging facts that the parameters of mean diameter and density have a great discriminative effect on the etiologic classification of male infertility and have great potential for further exploration.The potential mechanism lies in the negative feedback loop of sperm production. [13]A low sperm count induces more active spermatogenesis, leading to a compensatory increase of blood supply and thickening of the vascular network, which is supported by the histological findings of testicular tissue in OA showing vascular dilation and congested peritubular capillaries. [14]evertheless, there are still some limitations in our study.First, we used a single testicular ultrasound scan plane (the maximum transverse plane) in all subjects to ensure accordance with the plane of sperm collection by testicular sperm extraction (TESE) if necessary.However, it cannot comprehensively reflect the holistic picture of testicular vasculature.This restriction can be overcome by 3D reconstruction after the contiguous capture of 2D US images plane by plane or a 3D transducer.Second, given that this was a preliminary study, due to the limited sample size, we refrained from histological classification of the  The discrimination performance comparison of different ULM-based parameters in patients.Sensitivity, specificity, and 95% confidence intervals were presented as percentages.95% confidence intervals were included in parentheses.Abbreviations: AUC, area under the curve; CI, confidence intervals; NOA, non-obstructive azoospermia; NOO, non-obstructive oligoasthenospermia; OA, obstructive azoospermia; OO, obstructive oligoasthenospermia.
enrolled NOA patients.Literature has reported the presence of dilated vessels in the intertubular spaces in cases of normal-sized testes with mixed atrophy in NOA, [14] which was not observed in our current study.This discrepancy may be attributed to the insufficient sample size, potentially resulting in the omission or overshadowing of such results.Hence, we intend to augment our sample collection in our future work to elucidate potential disparities in super-resolution images among different NOA subtypes.In summary, we conclude that ULM is a promising non-invasive technology that opens new avenues for depicting human testicular microvasculature with details at an unprecedented limit resolution and providing robust quantitative vascular parameters.We also consider ULM as an innovative approach to assisting the differential diagnosis of obstructive and non-obstructive azoospermia or oligoasthenospermia.

Ultrasound localization microscopy processing pipeline
ULM processing was performed using the pipeline shown below (Figure 8).A two-stage image registration algorithm [7a] was employed to detect motions in the region of interest (ROI) of the testes induced by breathing or probe pressure in the B-mode sequence and to correct motions in the CEUS sequence.MBs in the CEUS sequence were localized and tracked using a framework proposed in our previous work. [9]MBs that were identified to exist for less than 3 frames during tracking were discarded to further reduce noise.
The trajectory of MB movement was generated by linking each paired MB with lines.The super-resolution MB density map was generated by accumulating trajectories on the corresponding pixels and blurred by a 2D Gaussian filter, whose SD was defined by the localization uncertainty of MBs.Flow velocity on each pixel was calculated by the vector mean of passed trajectories.The flow magnitude and direction were plotted into two images with the MB density map respectively.
Super-resolution images were also quantified by parameters, including flow velocity, tortuosity, diameter, and vessel density.The mean and SD of flow velocity were calculated across the non-zero values on the flow magnitude map.Tortuosity was defined by the ratio of total length to the distance between the two ends of each track.The binary vessel structure was obtained by thresholding the MB density with half the maximum value of the 2D Gaussian filter that was used for blurring.Vessel density was calculated by the area ratio of the vessel structure to the total ROI.Centerlines of vessel structures were estimated using the 'bwmorph' function, and the vessel diameter was twice the nearest distance of each point on the centerlines to the boundary of the vessel structure.

Patients
This multicenter retrospective study was conducted in accordance with the declaration of Helsinki and approved by the institutional review board (ChiCTR2100048361), and all participants were provided with informed consent.This study consecutively included 76 infertile outpatients with azoospermia/oligoasthenospermia from Shanghai Sixth People's Hospital and the International Peace Maternity and Child Health Hospital from August 2021 to May 2023.The included patients were diagnosed in accordance with the WHO criteria. [15]Exclusion criteria included infertility caused by active genital tract infection or malignancy and a history of recent exogenous hormone administration.Eligible patients underwent a systematic evaluation including a detailed medical history acquisition, semen analyses, physical examination, serum hormonal profile, and scrotal ultrasound or TRUS.Additionally, azoospermia patients underwent genetic analyses for karyotype and Y chromosome microdeletion.Percutaneous epididymal sperm aspiration (PESA) or TESA was performed if necessary.Based on the results, patients were categorized into NOO, OO, NOA, and OA groups (Figure 1).Age-matched normozoospermia volunteers recruited from the community were considered healthy controls for comparison.

Ultrasound examination protocol
Ultrasound examination was performed by an experienced sonographer with over 20 years of clinical experience in scrotal ultrasound imaging.Transducers were placed to ensure all testes were sectioned through the maximum transverse plane, in consistency with the section during TESE in the subsequent assisted reproductive technology treatment if necessary.Firstly, representative static images of bilateral testes were obtained with a L14-3 MHz (depth: 3.5 cm; gain 65) linear transducer (Resona R9, Mindray) after a B-mode scanning of the targeted region.Secondly, the CDFI mode was used to roughly visualize the main testicular blood flow.Thirdly, CEUS was performed with the same transducer (MI: 0.083; depth: 3.5 cm; gain 60), to measure the overall distribution of testicular vasculature.During CEUS, 2.4 mL of contrast medium (SonoVue) was administrated into a radial vein intravenously immediately after drawing into the syringe and was followed by a 0.9% saline flush (5 mL).After a duration of about 20 seconds, the contrast agent reached the testes and MBs were visible for 2-3 min.The MB suspension was prepared strictly according to the instructions.We acquired side-by-side Bmode and CEUS images in the testes at a frame rate of 37 Hz continuously for 30 s and the data were recorded for further analyses.For safety concerns, all subjects were kept under close medical supervision during and for at least 30 min following the administration of SonoVue to monitor the risk of serious hypersensitivity reactions.

Statistical analysis
ULM processing was done using MATLAB (2021a, Math-Works, MA).Data were presented as mean ± standard deviation and analyzed using GraphPad Prism 9.0.The differences in testicular volumes among patient groups and the control group were assessed using the Student's t-test.We extracted parameters from the ULM images and performed quantitative analyses.The evaluated parameters were as follows: (1) mean blood flow velocity, (2) mean blood vessel tortuosity, (3) mean blood vessel diameter, and (4) vascular density.The Shapiro-Wilks test and Levene's test were used to confirm the assumptions of normality of the data and homogeneity of variances, respectively.A One-way ANOVA test followed by a Bonferroni post-hoc test was conducted to assess the statistical significance of differences among independent groups and make multiple comparisons.The discrimination performance of the four ULM parameters was assessed by the area under the ROC curve.Using the optimal cut-off value based on the maximum Youden index, sensitivity and specificity were obtained.A probability level of p < .05 was considered a statistically significant difference, and a probability level of p < .01 or p < .001was considered highly statistically significant for all the aforementioned tests.

C O N F L I C T O F I N T E R E S T S TAT E M E N T
The authors declare no conflict of interest.

F I G U R E 1
Images of a normozoospermia testis with the application of ULM.(A) A super-resolution map of super-localized MBs with direction generated by ULM.Red represents MB flow toward the transducer, whereas blue indicates flow away from the transducer.(B) A super-resolution map of super-localized MBs with velocity generated by ULM, showing the corresponding speed of MB flow through the vessels, with the dynamic range of flow demonstration set from 0 to 10 mm/s.(C) An MIP image reconstructed by projecting the pixel with the highest value onto a 2D image extracted from the CEUS sequence.(D) A super-resolution map of super-localized MBs generated by ULM.(E) A magnified image of the rectangular region in (C).The vascular morphology was indistinguishable.(F) The same magnified region in (D), provides insights into the details of microvascular connectivity and branching at the same 35× magnification.(G) Vessel cross-section intensity profiles of ULM and CEUS images.Scale bars = 10 mm.ULM, ultrasound localization microscopy; MB, microbubble; MIP, maximum intensity projection; CEUS, contrast-enhanced ultrasound.

F I G U R E 2
Different imaging modalities in the evaluation of the testis of an OA patient in the same maximum transverse planes.Appearances of the testis on (A) B-mode ultrasound, (B) CDFI, (C) MIP of CEUS sequence, (D) MRI T1-weighted image, (E) and (F) super-resolution images from ULM. Scale bars = 10 mm.OA, obstructive azoospermia; CDFI, color Doppler flow imaging; MIP, maximum intensity projection; CEUS, contrast-enhanced ultrasound; MRI, magnetic resonance imaging; ULM, ultrasound localization microscopy.

TA B L E 1
Baseline characteristics of study patients.

=
Except where indicated, continuous variables were presented as mean ± standard deviation, with a 95% confidence interval in parentheses.#Data were numbers of patients, with percentages in parentheses.* = All data were 0, not described as normally distributed data.Abbreviations: FSH, follicle-stimulating hormone; LH, luteinizing hormone; N, normozoospermia; NOA, non-obstructive azoospermia; NOO, non-obstructive oligoasthenospermia; OA, obstructive azoospermia; OO, obstructive oligoasthenospermia; T, testosterone.Testicular volumes (in mL) were calculated using the formula of Lambert: length × height

F
I G U R E 4 ULM-based super-resolution parameter maps of testes with different vascular patterns among N, NOO, OO, NOA, and OA groups.(A) Maps of super-localized MBs, representing the relative blood volume in the testes.(B) Maps of super-localized MBs with direction.Red represents MB flow toward the transducer, whereas blue indicates flow away from the transducer, providing information about arterial and venous supply.(C) Maps of super-localized MBs with velocity magnitude, showing the corresponding speed of MB flow through the vessels, with the color dynamic range set from 0 to 10 mm/s.(D) Maps of super-localized MBs with angle, an indicator of entropies of blood flow, demonstrated using a color wheel.(E) Maps of super-localized MBs with arriving time, indicating the relative arrival time of MBs to the entire testicular contrast-enhanced area from the first frame when MB appeared.Scale bars = 10 mm.N, normozoospermia; NOO, non-obstructive oligoasthenospermia; OO, obstructive oligoasthenospermia; NOA, non-obstructive azoospermia; OA, obstructive azoospermia; MB, microbubble; ULM, ultrasound localization microscopy.

F I G U R E 7
Graphs of the ROC curves.Performance of various ULM-based parameters in the form of ROC curves and AUC scores based on mean diameter, density, mean tortuosity, and mean velocity for (A) NOO and OO patients and (B) NOA and OA patients.The ROC curves of models are represented by lines of different colors, and the legend indicates the AUC values for each parameter.ROC, receiver operating characteristic; NOO, non-obstructive oligoasthenospermia; OO, obstructive oligoasthenospermia; NOA, non-obstructive azoospermia; OA, obstructive azoospermia; AUC, area under the curve; ULM, ultrasound localization microscopy.TA B L E 2 The discrimination performance comparison of different parameters in patients.

F I G U R E 8
ULM processing diagram.(A) Pipeline for generating super-resolution images from the acquired two-column image sequence.(B) Sparsity-based deconvolution shrank MB images to smaller regions to isolate MBs from overlapped images.Image features could be preserved by deconvolution.(C) Graph-based assignment found the globally optimized pairing between two frames by considering motion model and image feature difference between MBs.MBs paired with dummy MBs were considered as disappearing or appearing.ULM, ultrasound localization microscopy; MB, microbubble.
This study was supported by the Key Program of the National Natural Science Foundation of China (No. 82030050, T2394534), National Key Research and Development Program of China (No. 2023YFC2410800), Key Project of the Shanghai Committee of Science and Technology in 2021 (No. 21Y21901100).