Increased Junctional Zone Stiffness and Serum Small Extracellular Vesicle Proteomic Signatures in Adenomyosis-Associated Infertility: An Exploratory SWE and Proteomic Study

Adenomyosis is an estrogen-dependent benign gynecological disorder characterized by the presence of ectopic endometrial glands and stroma within the myometrium. 1 Although adenomyosis has traditionally been associated with older multiparous women, increasing evidence suggests that it is also observed in women with subfertility and may adversely affect natural fertility and assisted reproductive technology (ART) outcomes. 2 , 3 However, the relationship between adenomyosis and reproductive outcomes remains controversial, partly because of heterogeneity in diagnostic criteria, disease phenotypes, and study populations. 4 The revised Morphological Uterus Sonographic Assessment (MUSA) criteria have provided a standardized ultrasound-based framework for diagnosing adenomyosis by classifying sonographic features into direct and indirect signs. 5 Using the MUSA criteria, Alson et al 6 reported that approximately one in ten women scheduled for ART exhibited direct ultrasound features of adenomyosis, whereas Dason et al 7 suggested that these features may not significantly affect reproductive outcomes. These inconsistent findings indicate that structural diagnosis alone may be insufficient to explain the reproductive consequences of adenomyosis. From a reproductive perspective, adenomyosis-associated infertility (AM+INF) should not be simply equated with adenomyosis itself. Although adenomyosis has been associated with impaired fertility, not all women with adenomyosis experience infertility, and the mechanisms linking adenomyosis to reproductive dysfunction are likely multifactorial, involving impaired endometrial receptivity, abnormal uterine peristalsis, chronic inflammation, fibrosis, and extracellular matrix (ECM) remodeling. 8 , 9 The uterine junctional zone (JZ), located between the endometrium and the outer myometrium, is highly relevant to implantation and reproductive function. Alterations in the JZ may disrupt coordinated uterine peristalsis, impair embryo transport and implantation, and contribute to reduced endometrial receptivity. 10 In adenomyosis, JZ thickening, irregularity, and architectural distortion have been associated with impaired fertility and poorer ART outcomes. 11 However, morphological abnormalities alone may not fully capture the functional and biomechanical alterations of the JZ, such as fibrosis, smooth muscle dysfunction, and ECM-related tissue remodeling. Therefore, beyond morphological assessment, evaluating the biomechanical properties of the JZ may provide additional insight into the functional uterine alterations associated with AM+INF. Ultrasound elastography, particularly shear wave elastography (SWE), allows non-invasive quantitative assessment of tissue stiffness and has been increasingly applied in gynecological disorders. 12 Previous ultrasound elastography studies in adenomyosis have mainly focused on adenomyotic lesions, global or regional myometrial stiffness, or other reproductive tract tissues. 13 Recent work by Kurt et al 14 further suggested that cervical elastography parameters may reflect adenomyosis-related biomechanical alterations, supporting the potential value of SWE as a non-invasive tool for assessing tissue stiffness in adenomyosis. However, compared with ultrasound elastography studies focusing on adenomyotic lesions, myometrial stiffness, or cervical stiffness, the potential relevance of JZ-specific SWE parameters in AM+INF remains insufficiently explored. In this context, SWE-based assessment of the JZ may provide additional, non-invasive information regarding functional uterine alterations associated with AM+INF. Small extracellular vesicles (sEVs) are membrane-bound vesicles that carry proteins, lipids, and nucleic acids and mediate intercellular communication. Accumulating evidence suggests that sEVs are involved in reproductive processes, including gamete maturation, fertilization, embryo development, and implantation, and that altered sEV-mediated communication may be associated with reproductive disorders. 15 Previous proteomic work has also identified adenomyosis-associated proteins in circulating sEVs, suggesting that serum-derived sEVs profiling may provide complementary molecular information on adenomyosis-related biological alterations. 16 However, because circulating sEVs may originate from multiple tissues, their proteomic signatures should be interpreted cautiously and should not be assumed to directly reflect uterine tissue-derived changes. Moreover, evidence specifically linking serum sEV proteomic alterations to infertility in women with adenomyosis remains limited. Therefore, studies focusing on serum proteomics of sEVs in women with AM+INF are still needed. Because fibrosis and ECM remodeling may contribute to altered uterine tissue stiffness, while serum sEV proteomics may provide exploratory circulating molecular information related to ECM and cell-adhesion pathways, integrating JZ-SWE with serum sEV profiling may help characterize complementary biomechanical and molecular features associated with AM+INF. In this study, we investigated JZ stiffness using SWE and explored serum sEV proteomic alterations in women with AM+INF compared with healthy controls. Given the study design, our aim was to characterize imaging and circulating sEV proteomic features associated with AM+INF, rather than to definitively distinguish infertility-specific mechanisms from adenomyosis-related changes. We hypothesized that AM+INF may be associated with altered JZ stiffness and corresponding changes in serum sEV protein profiles. By integrating JZ SWE parameters with serum sEV proteomic profiling, this study aimed to explore potential biomechanical and molecular alterations associated with AM+INF and to generate preliminary evidence for future studies.

Endometrial elasticity measurements (ENSWVmean, ENSWVmax, and ENSWVmin) and JZ elasticity measurements (JZSWVmean, JZSWVmax, and JZSWVmin) demonstrated excellent intra- and inter-observer reproducibility with ICCs ranging from 0.940 to 0.998 (all ICCs >0.90, Table S1 . The Bland–Altman analysis showed good agreement between the repeated SWE measurements of both ENSWV and JZSWV. Most observations fell within the 95% limits of agreement for intra- and inter-observer comparisons, with no significant systematic bias (Bland–Altman plots for JZ and endometrial SWE measurements are shown in Figure 2 and Figure S1 , respectively, with the corresponding agreement statistics summarized in Table S2 ). Figure 2 Bland–Altman analysis of repeated SWE measurements for JZ shear wave velocity parameters. Bland–Altman plots showing intra-observer agreement by the same physician ( A – C ) and inter-observer agreement by different physicians ( D – F ) for JZSWVmin, JZSWVmax, and JZSWVmean. The x-axis represents the mean of two paired measurements, and the y-axis represents their difference. The solid line indicates zero difference, and the dotted lines indicate the 95% limits of agreement. Mean bias and 95% limits of agreement are provided in Table S2 . Most measurements fell within the 95% limits of agreement, supporting good reproducibility of JZ SWE measurements. The image A showing a Bland Altman scatter plot titled, Same physician, JZSWVmin. The x axis label is Average with range 0 to 4. The y axis label is Difference with range negative 0.10 to 0.10. Points appear near Average about 0.8 at Difference about 0.07; around Average 1.2 to 3.2 mostly at Difference 0.00; and two points near Average about 2.2 to 2.6 at Difference about negative 0.06. A solid horizontal line is at Difference 0.00 and dotted horizontal lines are near Difference about 0.07 and about negative 0.07. The image B showing a Bland Altman scatter plot titled, Same physician, JZSWVmax. The x axis label is Average with range 3 to 7. The y-axis is labeled Difference and ranges from −0.15 to 0.10. Points cluster mainly around Average 3.6 to 5.2 with differences near 0.00, with positive points at approximately 0.05 to 0.06 and lower points at approximately −0.06, −0.09, and −0.11. One point lies near Average 6.8 with a difference close to 0.00. A solid horizontal line is at approximately 0.00, with dotted horizontal lines near 0.08 and −0.10. The image C showing a Bland Altman scatter plot titled, Same physician, JZSWVmean. The x axis label is Average with range 2.5 to 4.5. The y axis label is Difference with range negative 0.10 to 0.10. Points span Average about 2.6 to 4.1, with most near Difference 0.00. Higher points include about Average 2.9 at Difference about 0.05 and about Average 3.2 at Difference about 0.03. Lower points include about Average 2.8 at Difference about negative 0.04 and about Average 3.6 at Difference about negative 0.05. A solid horizontal line is at Difference 0.00 and dotted horizontal lines are near Difference about 0.05 and about negative 0.05. The image D showing a Bland Altman scatter plot titled, Different physicians, JZSWVmin. The x axis label is Average with range 0 to 4. The y axis label is Difference with range negative 0.10 to 0.10. Points lie mainly between Average about 1.0 and 3.2 with Difference near 0.00, with two higher points near Difference about 0.06 around Average about 1.8 and 2.6 and one lower point near Difference about negative 0.06 around Average about 2.1. A solid horizontal line is at Difference 0.00 and dotted horizontal lines are near Difference about 0.07 and about negative 0.07. The image E showing a Bland Altman scatter plot titled, Different physicians, JZSWVmax. The x axis label is Average with range 3 to 7. The y axis label is Difference with range negative 0.15 to 0.10. Points cluster around Average about 3.6 to 5.2 with Difference near 0.00, with several negative points near Difference about negative 0.03 to negative 0.06, one low point near Difference about negative 0.10 around Average about 4.8 and one point near Average about 6.6 at Difference about 0.02. A solid horizontal line is at Difference 0.00 and dotted horizontal lines are near Difference about 0.06 and about negative 0.08. The image F showing a Bland Altman scatter plot titled, Different physicians, JZSWVmean. The x axis label is Average with range 2.0 to 4.5. The y axis label is Difference with range negative 0.08 to 0.04. Points lie between Average about 2.6 and 4.1. Several points are near Difference 0.00 around Average about 3.2 to 3.5. Positive points include about Average 2.9 at Difference about 0.01 and about Average 3.2 at Difference about 0.03. Negative points include about Average 2.7 to 3.1 at Difference about negative 0.04 to negative 0.06 and a point near Average about 4.0 at Difference about negative 0.04. A solid horizontal line is at Difference 0.00 and dotted horizontal lines are near Difference about 0.04 and about negative 0.07. A scatter plot set showing Bland Altman agreement for JZSWVmin, JZSWVmax and JZSWVmean. Abbreviations : SWE, shear wave elastography; JZ, junctional zone; JZSWV, junctional zone shear wave velocity. Bland–Altman analysis of repeated SWE measurements for JZ shear wave velocity parameters. Bland–Altman plots showing intra-observer agreement by the same physician ( A – C ) and inter-observer agreement by different physicians ( D – F ) for JZSWVmin, JZSWVmax, and JZSWVmean. The x-axis represents the mean of two paired measurements, and the y-axis represents their difference. The solid line indicates zero difference, and the dotted lines indicate the 95% limits of agreement. Mean bias and 95% limits of agreement are provided in Table S2 . Most measurements fell within the 95% limits of agreement, supporting good reproducibility of JZ SWE measurements. A total of 24 women were included, comprising 16 patients with AM+INF and eight HC volunteers. Age, BMI, reproductive hormone levels, AMH, and serum CA125 were comparable between groups. In the AM+INF group, five patients had primary infertility and 11 had secondary infertility, with a median infertility duration of 1.50 years. All AM+INF patients presented at least one direct ultrasound feature of adenomyosis, including myometrial cysts, hyperechogenic islands, or echogenic subendometrial lines and buds, whereas none of these features were observed in controls. Indirect ultrasound features were also recorded but were not used as inclusion criteria. Participant characteristics are summarized in Table 1 . Patients with AM+INF showed larger uterine size than HC ( P = 0.001) and lower endometrial and JZ volumes ( P = 0.023 and P = 0.032), whereas endometrial thickness was comparable between groups. ENSWVmean, ENSWVmin, JZSWVmean, and JZSWVmin were significantly higher in the AM+INF group (all P <0.05), while ENSWVmax and JZSWVmax showed no significant differences. Detailed results are presented in Table 2 . Table 2 Comparison of Ultrasound and SWE Parameters Between AM+INF Patients and Healthy Controls Parameter Group HL Difference (95% CI) Z Effect size r P AM+INF (n=16), Median (IQR) HC (n=8), Median (IQR) Uterine size 15.40 (14.70, 16.96) 12.85 (12.55, 13.75) 2.63 (1.63, 3.70) 3.246 0.663 0.001* Endometrial thickness 6.55 (5.25, 9.50) 7.85 (6.10, 11.22) −1.20 (−3.80, 1.25) −0.827 0.169 0.408 Endometrial volume 2.25 (1.58, 4.55) 4.75 (3.08, 7.10) −1.95 (−5.00, −0.45) −2.267 0.463 0.023* JZvolume 2.90 (2.45, 3.93) 4.55 (3.52, 4.73) −0.95 (−2.05, −0.10) −2.144 0.438 0.032* ENSWVmean 2.97 (2.59, 3.33) 2.10 (1.83, 2.33) 0.87 (0.48, 1.31) 3.185 0.650 0.001* ENSWVmax 4.13 (4.00, 4.45) 3.81 (3.52, 4.21) 0.41 (−0.31, 0.84) 1.776 0.363 0.076 ENSWVmin 1.85 (1.44, 2.61) 0.96 (0.51, 1.33) 1.09 (0.47, 1.73) 2.694 0.550 0.007* JZSWVmean 3.09 (2.86, 3.38) 2.65 (2.55, 2.74) 0.58 (0.25, 0.76) 3.369 0.688 0.001* JZSWVmax 4.35 (4.07, 4.61) 4.25 (3.88, 4.62) 0.14 (−0.46, 0.62) 0.551 0.113 0.582 JZSWVmin 2.11 (1.73, 2.55) 0.96 (0.73, 1.41) 1.04 (0.50, 1.68) 3.062 0.625 0.002* Notes : Data are presented as median (interquartile range). Between-group comparisons were performed using the Mann–Whitney U -test, and standardized Z values are reported. Effect size r was calculated as |Z|/√N. Unadjusted nominal P values are presented because the ultrasound and SWE comparisons were used as exploratory analyses of predefined imaging parameters. * P < 0.05 was considered statistically significant. Abbreviations : AM+INF, adenomyosis-associated infertility; ENSWV, endometrial shear wave velocity; JZSWV, junctional zone shear wave velocity; mean, max, and min indicate the mean, maximum, and minimum values, respectively. Comparison of Ultrasound and SWE Parameters Between AM+INF Patients and Healthy Controls Notes : Data are presented as median (interquartile range). Between-group comparisons were performed using the Mann–Whitney U -test, and standardized Z values are reported. Effect size r was calculated as |Z|/√N. Unadjusted nominal P values are presented because the ultrasound and SWE comparisons were used as exploratory analyses of predefined imaging parameters. * P < 0.05 was considered statistically significant. Abbreviations : AM+INF, adenomyosis-associated infertility; ENSWV, endometrial shear wave velocity; JZSWV, junctional zone shear wave velocity; mean, max, and min indicate the mean, maximum, and minimum values, respectively. Collinearity analysis showed that several SWE parameters had high variance inflation factors. ENSWVmean, ENSWVmin, and JZSWVmin showed substantial multicollinearity, with VIF values of 23.846, 45.028, and 32.303, respectively. JZSWVmean showed borderline collinearity, with a VIF of 9.249. In contrast, uterine size, endometrial volume, and JZ volume showed lower VIF values. Given the marked collinearity among SWE parameters, alternative feature-selection approaches were used instead of simultaneously including these variables in a conventional multivariable regression model. Detailed results are presented in Table S3 . Based on the univariate Mann–Whitney U -test results, seven ultrasound and SWE-derived variables with significant between-group differences were included as candidate input features for the exploratory machine-learning pipeline. Using L1-regularized logistic regression within the training pipeline, three variables retained non-zero coefficients: uterine size, ENSWVmean, and JZSWVmin. The remaining variables, including endometrial volume, JZ volume, JZSWVmean, and ENSWVmin, were shrunk to zero and excluded from the final feature set ( Figure 3A ). These selected features were then used for SVM classification. Figure 3 Exploratory machine-learning analysis and SHAP-based interpretation. ( A ) LASSO path plot for feature selection among ultrasound and SWE-derived parameters. Three features retained non-zero coefficients: uterine size, ENSWVmean, and JZSWVmin. ( B ) ROC curve of the final SVM model in the internal hold-out test set, with an AUC of 0.933 (95% CI: 0.625–1.000; n = 8). ( C ) SHAP summary plot showing the impact of the selected features on the model output. Feature values are colored from low (blue) to high (red). ( D ) SHAP global importance plot based on mean absolute SHAP values, ranking JZSWVmin as the most influential feature, followed by ENSWVmean and uterine size. Image A: Line graph with x-axis Alpha (10 superscript -3 to 10 superscript -1) and y-axis Coefficients (-0.4 to 0.4). Legend: uterine size, ENvolume, JZvolume, JZSWVmean, JZSWVmin, ENSWVmean, ENSWVmin. Curves trend toward 0 as Alpha increases. Image B: ROC curve with x-axis False Positive Rate and y-axis True Positive Rate (0.0 to 1.0). Diagonal reference line from (0.0, 0.0) to (1.0, 1.0). .The ROC curve includes points at approximately (0.0, 0.8), (0.33, 0.8), (0.33, 1.0), and (1.0, 1.0). Text: ROC curve (AUC = 0.933), 95% CI: 0.625 to 1.000, n = 8. Image C: SHAP summary dot plot with x-axis SHAP value (impact on model output) and y-axis categories JZSWVmin, ENSWVmean, uterine size. X-axis ticks at -0.2, -0.1, 0.0, 0.1, 0.2, with vertical line at 0.0. Side scale labeled Feature value (High to Low). Image D: Horizontal bar chart with x-axis mean(|SHAP value|) (average impact on model output magnitude) and y-axis categories JZSWVmin, ENSWVmean, uterine size. X-axis ranges 0.00 to 0.14. Bar lengths: JZSWVmin 0.14, ENSWVmean 0.13, uterine size 0.11. Different plots showing LASSO path, ROC curve, SHAP summary and SHAP importance for three features. Abbreviations : LASSO, least absolute shrinkage and selection operator; SWE, shear wave elastography; ROC, receiver operating characteristic; AUC, area under the curve; SVM, support vector machine; SHAP, Shapley additive explanations; JZSWVmin, minimum junctional zone shear wave velocity; ENSWVmean, mean endometrial shear wave velocity; JZ, junctional zone. Exploratory machine-learning analysis and SHAP-based interpretation. ( A ) LASSO path plot for feature selection among ultrasound and SWE-derived parameters. Three features retained non-zero coefficients: uterine size, ENSWVmean, and JZSWVmin. ( B ) ROC curve of the final SVM model in the internal hold-out test set, with an AUC of 0.933 (95% CI: 0.625–1.000; n = 8). ( C ) SHAP summary plot showing the impact of the selected features on the model output. Feature values are colored from low (blue) to high (red). ( D ) SHAP global importance plot based on mean absolute SHAP values, ranking JZSWVmin as the most influential feature, followed by ENSWVmean and uterine size. The optimized SVM classifier was constructed using the three selected features: uterine size, ENSWVmean, and JZSWVmin. Hyperparameter tuning was performed by stratified five-fold cross-validation within the training set, and the final model used a linear kernel with C = 0.1. In the internal hold-out test set of eight samples, the model yielded 3 true negatives, 0 false positives, 1 false negative, and 4 true positives. The model achieved an AUC of 0.933 (95% CI: 0.625–1.000) ( Figure 3B ), accuracy of 0.875 (95% CI: 0.625–1.000), sensitivity/recall of 0.800 (95% CI: 0.333–1.000), specificity of 1.000 (95% CI: 1.000–1.000), precision of 1.000 (95% CI: 1.000–1.000), and F1-score of 0.889 (95% CI: 0.500–1.000). Permutation testing showed that the cross-validated true-label AUC was higher than the null distribution generated by random label permutation, with a mean permuted AUC of 0.545 and a permutation P value of 0.0099 ( Figure S2 ). Given the small hold-out sample size and wide confidence intervals, these results should be interpreted as exploratory rather than clinically diagnostic. SHAP analysis was performed to provide exploratory interpretation of the final SVM model based on the three retained features. The SHAP summary plot and global importance plot showed that JZSWVmin, ENSWVmean, and uterine size contributed to the model output, with JZSWVmin showing the highest mean absolute SHAP value, followed by ENSWVmean and uterine size ( Figure 3C and D ). Higher values of these features generally tended to increase the model output toward AM+INF classification. However, because SHAP values were derived from a very small hold-out test set, these interpretability results should be considered hypothesis-generating and should not be interpreted as confirmatory evidence of independent clinical predictors. Serum sEVs isolated from patients with AM+INF and HC volunteers were characterized using nanoFCM and TEM. NanoFCM showed that vesicles from both groups were mainly distributed within the typical sEV size range of 30–150 nm, as shown by particle size distribution analysis in the AM+INF group and HC group, respectively ( Figure 4A and Figure S3A ). The mean particle diameter was 96.47 nm in the AM+INF group and 92.25 nm in the HC group, with comparable size profiles. The absolute sEV concentration did not differ significantly between groups (AM+INF: 8.18 × 10 9 particles/mL vs HC: 4.74 × 10 9 particles/mL, P = 0.40). TEM revealed round- to cup-shaped membrane-bound vesicles consistent with sEV morphology in both groups ( Figure 4B and Figure S3B ). NanoFCM further confirmed the expression of the EV-associated tetraspanins CD9, CD63, and CD81 on the vesicles ( Figure 4C–E and Figure S3C – E ), with minimal signal in the blank control ( Figure 4F and Figure S3F ). Figure 4 Characterization of serum sEVs from patients with AM+INF. ( A ) Representative nanoFCM particle size distribution of serum sEVs from the AM+INF group. ( B ) Representative TEM image showing round- to cup-shaped membrane-bound vesicles consistent with sEV morphology; scale bar = 100 nm. ( C – E ) Representative nanoFCM analysis showing positive expression of the EV-associated tetraspanins CD9, CD63, and CD81, respectively. ( F ) Blank control showing minimal background signal. Red events indicate marker-positive vesicles, and blue events indicate marker-negative/background events. Image A: Histogram of particle sizes (30-150 nm) with concentration per ml (0-1.2e+8). Peaks at 65-75 nm near 1.0e+8, declines to 150 nm. Image B: TEM micrograph of vesicle, scale bar 100 nm. Image C: nanoFCM plots labeled CD9. Top histogram: Events (0-60). Middle scatter: SS-A (1-10M), FITC-A (1-1M). P1: 527/20.5%, P2: 2047/79.5%. Right histogram: Events (0-300). Image D: nanoFCM plots labeled CD63. Top histogram: Events (0-60). Middle scatter: SS-A (1-10M), FITC-A (1-1M). P1: 697/22.1%, P2: 2462/77.9%. Right histogram: Events (0-300). Image E: nanoFCM plots labeled CD81. Top histogram: Events (0-60). Middle scatter: SS-A (1-10M), FITC-A (1-1M). P1: 171/8.4%, P2: 1858/91.6%. Right histogram: Events (0-200). Image F: nanoFCM plots labeled Blank control. Top histogram: Events (0-40). Middle scatter: SS-A (1-10M), FITC-A (1-1M). P1: 12/0.3%, P2: 4276/99.7%. Right histogram: Events (0-800). A mixed figure showing one histogram, one micrograph and four flow cytometry scatter plots for serum sEVs. Abbreviations : AM+INF, adenomyosis-associated infertility; sEVs, small extracellular vesicles; nanoFCM, nano-flow cytometry; TEM, transmission electron microscopy; sEV, small extracellular vesicle. Characterization of serum sEVs from patients with AM+INF. ( A ) Representative nanoFCM particle size distribution of serum sEVs from the AM+INF group. ( B ) Representative TEM image showing round- to cup-shaped membrane-bound vesicles consistent with sEV morphology; scale bar = 100 nm. ( C – E ) Representative nanoFCM analysis showing positive expression of the EV-associated tetraspanins CD9, CD63, and CD81, respectively. ( F ) Blank control showing minimal background signal. Red events indicate marker-positive vesicles, and blue events indicate marker-negative/background events. Peptide length distribution showed that most identified peptides were short, mainly ranging from 7 to 15 amino acids, with the highest abundance in the 7–9 amino acid range ( Figure 5A ). Molecular weights of identified proteins were predominantly distributed between 20 and 70 kDa ( Figure 5B ). Venn analysis identified 1,739 shared proteins between the AM+INF and HC groups, with 16 proteins unique to AM+INF and 21 unique to HC ( Figure 5C ). GO classification showed that the identified proteins were mainly associated with cellular process, biological regulation, cellular anatomical entity, protein-containing complex, binding, and catalytic activity ( Figure 5D ). KEGG pathway annotation indicated enrichment in metabolic pathways, regulation of actin cytoskeleton, endocytosis, PI3K-Akt signaling, platelet activation, and focal adhesion-related pathways ( Figure 5E ). Figure 5 Overall annotation of the serum sEV proteome in patients with AM+INF and HC. ( A ) Peptide length distribution, with the highest abundance in the 7–9 amino acid range. ( B ) Molecular weight distribution of identified proteins. ( C ) Venn diagram showing 1,739 shared proteins, 16 AM+INF-specific proteins, and 21 HC-specific proteins. ( D ) GO functional classification of identified proteins. ( E ) KEGG pathway annotation of identified proteins. A) Peptide Length Distribution: Histogram showing peptide lengths from 5 to 35 amino acids, with highest abundance in 7 to 9 range. B) Molecular Weight Distribution: Histogram and line graph showing protein weights from 0 to 180 kDa, with cumulative percentage. C) Venn Diagram: 1,739 shared proteins, 16 AM-specific, 21 HC-specific. D) Function Classification (GO): Bar graph showing protein numbers across biological processes, cellular components and molecular functions. Categories include cellular anatomical entity, protein-containing complex, binding and catalytic activity. E) KEGG Pathway: Bar graph showing number of proteins in pathways like metabolic, neurodegeneration, Alzheimer disease and PI3K-Akt signaling. Each panel provides detailed insights into the serum sEV proteome analysis. Infographic: serum sEV proteome - peptide length, weight, Venn, GO, KEGG. Abbreviations : sEVs, small extracellular vesicles; AM+INF, adenomyosis-associated infertility; HC, healthy controls; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes. Overall annotation of the serum sEV proteome in patients with AM+INF and HC. ( A ) Peptide length distribution, with the highest abundance in the 7–9 amino acid range. ( B ) Molecular weight distribution of identified proteins. ( C ) Venn diagram showing 1,739 shared proteins, 16 AM+INF-specific proteins, and 21 HC-specific proteins. ( D ) GO functional classification of identified proteins. ( E ) KEGG pathway annotation of identified proteins. Differentially abundant proteins between the AM+INF and HC groups were screened using a fold-change threshold of FC > 1.2 or FC < 0.83 together with a nominal P value < 0.05. FDR-adjusted P values were calculated using the Benjamini–Hochberg method. Using the predefined exploratory criteria, 100 candidate differentially abundant proteins were identified between the AM+INF and HC groups, including 63 upregulated and 37 downregulated proteins in the AM+INF group ( Figure 6A and B ). These findings were interpreted as exploratory because no protein remained statistically significant after FDR correction ( Table S4 ). Enrichment analyses were performed for all differentially abundant proteins, upregulated proteins, and downregulated proteins. The analyses of all differential proteins and upregulated proteins showed heterogeneous biological themes and are provided in Figure S4 . Because the downregulated proteins were more directly enriched in cell–matrix adhesion, integrin-related terms, ECM–receptor interaction, and PI3K-Akt signaling, subsequent interpretation focused on the downregulated proteins ( Figure 6C and D ). PPI analysis and cytoHubba screening further identified growth factor receptor-bound protein 2 (GRB2) as the hub protein with the highest node degree ( Figure 6E and F ). Figure 6 Candidate differentially abundant serum sEV proteins in AM+INF versus HC. ( A ) Volcano plot of proteins screened using the exploratory criteria of FC > 1.2 or FC < 0.83 with nominal P < 0.05. Increased and decreased proteins in the AM+INF group are shown in red and blue, respectively; gray dots indicate proteins not meeting the screening criteria. No protein remained statistically significant after FDR correction. ( B ) Heat map of the 100 candidate differentially abundant proteins, including 63 increased and 37 decreased proteins in the AM+INF group. ( C ) GO enrichment analysis of the decreased proteins. Red labels indicate GO terms considered potentially relevant to AM+INF, whereas black labels indicate other enriched terms. ( D ) KEGG enrichment analysis of the decreased proteins. Red labels indicate pathways considered potentially relevant to AM+INF, whereas black labels indicate other enriched pathways. ( E ) STRING protein–protein interaction network of the decreased proteins. ( F ) cytoHubba network analysis identifying GRB2 as the top-ranked hub protein. A volcano plot shows proteins screened using criteria of fold change greater than 1.2 or less than 0.83 with nominal P value less than 0.05. Increased proteins in AM plus INF are on the right, decreased on the left. A heat map displays 100 candidate proteins, 63 increased and 37 decreased in AM plus INF. GO enrichment analysis highlights terms relevant to AM plus INF, such as mesoderm development and cell-substrate adhesion. KEGG enrichment analysis shows pathways like PI3K-Akt signaling and ECM-receptor interaction. A STRING protein-protein interaction network identifies GRB2 as a central hub. CytoHubba network analysis ranks GRB2 as the top hub protein; the network includes PTPRA, CALD1, WIPF1, BTK, ERBIN, UBE2N, and STIP1. Protein analysis: volcano plot, heat map, enrichment, interaction, cytoHubba network. Abbreviations : AM+INF, adenomyosis-associated infertility; HC, healthy controls; sEVs, small extracellular vesicles; FC, fold change; FDR, false discovery rate; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; STRING, Search Tool for the Retrieval of Interacting Genes/Proteins; GRB2, growth factor receptor-bound protein 2. Candidate differentially abundant serum sEV proteins in AM+INF versus HC. ( A ) Volcano plot of proteins screened using the exploratory criteria of FC > 1.2 or FC < 0.83 with nominal P < 0.05. Increased and decreased proteins in the AM+INF group are shown in red and blue, respectively; gray dots indicate proteins not meeting the screening criteria. No protein remained statistically significant after FDR correction. ( B ) Heat map of the 100 candidate differentially abundant proteins, including 63 increased and 37 decreased proteins in the AM+INF group. ( C ) GO enrichment analysis of the decreased proteins. Red labels indicate GO terms considered potentially relevant to AM+INF, whereas black labels indicate other enriched terms. ( D ) KEGG enrichment analysis of the decreased proteins. Red labels indicate pathways considered potentially relevant to AM+INF, whereas black labels indicate other enriched pathways. ( E ) STRING protein–protein interaction network of the decreased proteins. ( F ) cytoHubba network analysis identifying GRB2 as the top-ranked hub protein. MSigDB C2 enrichment analysis of the downregulated proteins highlighted several ECM- and cell adhesion-related pathways, including Reactome platelet adhesion to exposed collagen, Reactome cell surface interactions at the vascular wall, KEGG ECM–receptor interaction, Reactome ECM proteoglycans, and Reactome non-integrin membrane–ECM interactions ( Figure 7A ). GRB2, ANGPT1, ITGA5, and ITGB1, which were associated with these pathways, showed lower abundance in the AM+INF group than in HC ( Figure 7B–E ). Peptide-level supporting evidence for candidate proteins identified by DIA proteomics ( Table S5 ). Exploratory ELISA analysis of three selected proteins, GRB2, ITGA5, and ITGB1, further showed lower levels in AM+INF than in HC, with a significant decrease observed for ITGB1 and non-significant downward trends for GRB2 and ITGA5 ( Figure S5 ). These ELISA findings provide preliminary support for the proteomic results, particularly for ITGB1. However, given the limited and imbalanced sample size, these results should be interpreted cautiously and require further validation in larger independent cohorts. Figure 7 MSigDB C2 enrichment analysis and selected downregulated proteins. ( A ) MSigDB C2 enrichment analysis of downregulated serum sEV proteins. Red labels indicate pathways considered potentially relevant to ECM remodeling and cell–matrix interactions, whereas black labels indicate other enriched pathways. ( B–E ) GRB2, ANGPT1, ITGA5, and ITGB1 were decreased in the AM+INF group compared with HC. Data are shown as log2-transformed intensity values. Data are shown as log2-transformed intensity values. Statistical significance was indicated as * P < 0.05 and ** P < 0.01. The image A showing a bubble plot with y axis pathway names and x axis label, GeneRatio, with tick labels 2 over 37, 3 over 37, 4 over 37. A size legend labeled, Count, shows 2.0, 2.5, 3.0, 3.5, 4.0. A color scale labeled, minusLog10FDR, ranges 1.2 to 2.0. Pathway labels listed: WP VEDOLIZUMAB THERAPY FOR INFLAMMATORY BOWEL DISEASE, WP ARRHYTHMOGENIC RIGHT VENTRICULAR CARDIOMYOPATHY, KEGG ARRHYTHMOGENIC RIGHT VENTRICULAR CARDIOMYOPATHY ARVC, REACTOME PLATELET ADHESION TO EXPOSED COLLAGEN, REACTOME CELL SURFACE INTERACTIONS AT THE VASCULAR WALL, KEGG HYPERTROPHIC CARDIOMYOPATHY HCM, KEGG ECM RECEPTOR INTERACTION, KEGG DILATED CARDIOMYOPATHY, REACTOME ECM PROTEOGLYCANS, REACTOME NON INTEGRIN MEMBRANE ECM INTERACTIONS. Points appear at GeneRatio near 2 over 37, 3 over 37 and 4 over 37. The image B showing a bar chart titled GRB2. X axis categories: AM plus INF and HC. Y axis label, log2 open parenthesis Intensity close parenthesis, range 17.0 to 19.0. A bracket annotation reads asterisk p equals 0.0103. Bar heights: AM plus INF about 18.0; HC about 18.35. Dots show individual values around each bar. The image C showing a bar chart titled ANGPT1. X axis categories: AM plus INF and HC. Y axis label, log2 open parenthesis Intensity close parenthesis, range 0 to 20. A bracket annotation reads asterisk p equals 0.0223. Bar heights: AM plus INF about 10.5; HC about 13.0. Dots show individual values. The image D showing a bar chart titled ITGA5. X axis categories: AM plus INF and HC. Y axis label, log2 open parenthesis Intensity close parenthesis, range 15.0 to 16.0. A bracket annotation reads two asterisks p equals 0.0053. Bar heights: AM plus INF about 15.2; HC about 15.6. Dots show individual values. The image E showing a bar chart titled ITGB1. X axis categories: AM plus INF and HC. Y axis label, log2 open parenthesis Intensity close parenthesis, range 18.0 to 20.0. A bracket annotation reads two asterisks p equals 0.0084. Bar heights: AM plus INF about 19.1; HC about 19.45. Dots show individual values. A bubble plot and four bar charts showing enrichment and log2 intensity for GRB2, ANGPT1, ITGA5, ITGB1. Abbreviations : AM+INF, adenomyosis-associated infertility; HC, healthy controls; sEVs, small extracellular vesicles; MSigDB, Molecular Signatures Database. MSigDB C2 enrichment analysis and selected downregulated proteins. ( A ) MSigDB C2 enrichment analysis of downregulated serum sEV proteins. Red labels indicate pathways considered potentially relevant to ECM remodeling and cell–matrix interactions, whereas black labels indicate other enriched pathways. ( B–E ) GRB2, ANGPT1, ITGA5, and ITGB1 were decreased in the AM+INF group compared with HC. Data are shown as log2-transformed intensity values. Data are shown as log2-transformed intensity values. Statistical significance was indicated as * P < 0.05 and ** P < 0.01.

Women aged 20–42 years who presented with infertility were consecutively recruited from the Assisted Reproductive Center and Department of Ultrasound Medicine between January 2024 and August 2025. Infertility was defined as failure to get pregnant after ≥12 months of regular unprotected intercourse. 17 Inclusion criteria were: (1) primary or secondary infertility; (2) premenopausal status with regular or spontaneous menstrual cycles; (3) adenomyosis diagnosed using two-dimensional (2D) and three-dimensional (3D) transvaginal ultrasound (TVUS) according to the revised MUSA criteria, defined by the presence of at least one direct feature (myometrial cysts, hyperechogenic islands, or echogenic sub-endometrial lines and buds); and (4) provided written informed consent. 18 Exclusion criteria were tubal factor infertility, endometriosis (imaging-confirmed or previously diagnosed), ovarian insufficiency (anti-Müllerian hormone [AMH] levels of <1.0 ng/mL and/or antral follicle count of 40 kg/m 2 , malignant tumors, polycystic ovary syndrome, autoimmune or rheumatic diseases, and use of hormonal contraceptives or intrauterine devices within the preceding 3 months. All ultrasound images were independently reviewed by two experienced gynecological sonographers, and discrepancies were resolved by consensus. Images of inadequate quality were excluded following a standardized quality assessment. To provide non-disease reference measurements for JZ SWE parameters and serum sEV-related analyses, eight healthy women aged 20–42 years with sonographically normal uteri, no clinical or sonographic evidence of adenomyosis, and no known history of infertility were recruited as HC women. The HC women had regular menstrual cycles (28 ± 2 days), no dysmenorrhea, and no abnormal uterine bleeding, including menometrorrhagia. All HC women were asymptomatic and were recruited during routine gynecological checkups or pre-intrauterine device placement visits. This control group was used for an initial exploratory comparison between women with AM+INF and a healthy reference population; however, the study was not designed to separate the independent effects of adenomyosis from those of infertility. Given the exploratory nature of this study and the lack of prior data on JZ SWE parameters and sEV proteomic profiles in women with AM+INF, no formal a priori sample-size calculation was performed. The sample size was determined by the feasibility of recruiting strictly defined AM+INF cases and eligible HC women during the study period. Therefore, the imaging, proteomic, and machine-learning analyses were intended to be hypothesis-generating rather than confirmatory. All participants underwent TVUS in the lithotomy position using a Mindray Nuewa R9 system (Mindray Bio-Medical Electronics, Shenzhen, China) with a 2–9 MHz intracavitary transducer (Model: DE10-3WU). Both 2D- and 3D-TVUS were performed by a single experienced sonographer (>1,000 gynecologic examinations/year) who was blinded to the clinical data. The cervix, myometrium, and endometrium were systematically evaluated, and adenomyosis was assessed according to the revised MUSA criteria. 18 The uterus was divided into three layers: JZ layer, inner myometrium (between the JZ and uterine arcuate artery), and outer myometrium (between uterine arcuate artery and serosal layer including the serosa). After 2D TVUS examination, 3D TVUS was used to acquire volumetric data including endometrial volume ( Figure 1A, E ) and the total volume of the endometrium and JZ ( Figure 1B, F ). The adnexa and pelvis were examined for signs of endometriosis, and the antral follicle count was defined as the total number of follicles measuring 2–10 mm in both ovaries. SWE was performed in accordance with EFSUMB recommendations, with the probe stabilized for 5–10 s before acquisition. 19 SWE images were acquired in the longitudinal uterine plane, with the target tissue depth maintained within 5 cm from the transducer surface. The ultrasound elastographic image was stabilized for 3–5 s before measurement, and excessive probe compression was avoided. Image quality was assessed using the vendor-provided M-STB image-quality index; only measurements with a five-star M-STB index were accepted ( Figure 1C, G ). Images were reacquired if motion artifacts, poor color filling, signal dropout, acoustic shadowing, unstable ultrasound elastographic maps, or an M-STB index below five stars were present. SWE values were reported as SWV in m/s rather than Young’s modulus in kPa, because kPa conversion requires assumptions regarding tissue density, homogeneity, isotropy, and linear elasticity that may not be fully applicable to layered and heterogeneous JZ tissue. The endometrial ROI was manually traced along the outer endometrial contour ( Figure 1D, H ; white outline), and the minimum, maximum, and mean endometrial shear wave velocity values were recorded as ENSWVmin, ENSWVmax, and ENSWVmean, respectively. For JZ assessment, a 1-mm shell ROI was generated adjacent to the traced endometrial contour using the shell function ( Figure 1D, H ; pink outline), and the minimum, maximum, and mean junctional zone shear wave velocity values were recorded as JZSWVmin, JZSWVmax, and JZSWVmean, respectively. Because tracing and shell ROIs were used instead of a fixed circular Q-box, ROI size varied with uterine anatomy. Each measurement was repeated three times, and the average value was used for analysis. Reproducibility was assessed using intraclass correlation coefficients (ICCs). Reproducibility was assessed in a subset of 16 patients from the AM+INF group, which represented the target population for JZ-SWE assessment. For intra-observer reproducibility, the primary examiner repeated the SWE measurements twice. For inter-observer reproducibility, the measurements were independently reassessed by a second experienced sonographer. All the 2D TVUS images, 3D TVUS volumes, and SWE data were stored in a dedicated database for subsequent analysis. All SWE examinations and peripheral blood sampling were performed during the same study visit in the proliferative phase of the menstrual cycle. Peripheral blood was collected immediately before or after the ultrasound examination according to the clinical workflow, ensuring that blood sampling and imaging were synchronized on the same days. Figure 1 Representative 3D-TVUS and SWE assessment of endometrial and junctional zone parameters. Representative images from a healthy control ( A – D ) and a patient with adenomyosis-associated infertility ( E – H ) are shown. 3D-TVUS was used to assess endometrial volume and the combined volume of the endometrium and JZ. In panels ( A and E ) V(Orig) represents ENvol. In panels ( B and F ), V(Out) represents EJvol. JZvol was calculated as V(Out) − V(Orig). SWE was used to assess endometrial and JZ elasticity in the healthy control ( C and D ) and in the patient with adenomyosis-associated infertility ( G and H ). The white outline indicates the endometrial ROI, and the pink outline indicates the JZ shell ROI. In the measurement tables shown in panels ( D and H ) the white “A” column corresponds to endometrial SWE values, whereas the green “Shell” column corresponds to JZ SWE values. Representative volume and shear wave velocity values are displayed in the corresponding panels. Eight ultrasound images (A–H) showing representative 3D-TVUS and SWE assessments in a healthy control (A–D) and a patient with adenomyosis-associated infertility (E–H), including 3D volume renderings, elastography maps, ROI outlines, scales, and measurement tables. Abbreviations : 3D-TVUS, three-dimensional transvaginal ultrasound; SWE, shear wave elastography; JZ, junctional zone; ROI, region of interest; ENvol, endometrial volume; EJvol, total volume of the endometrium and JZ; JZvol, junctional zone volume; ENSWV, endometrial shear wave velocity; JZSWV, junctional zone shear wave velocity. Mean, max, and min indicate mean, maximum, and minimum values, respectively. Representative 3D-TVUS and SWE assessment of endometrial and junctional zone parameters. Representative images from a healthy control ( A – D ) and a patient with adenomyosis-associated infertility ( E – H ) are shown. 3D-TVUS was used to assess endometrial volume and the combined volume of the endometrium and JZ. In panels ( A and E ) V(Orig) represents ENvol. In panels ( B and F ), V(Out) represents EJvol. JZvol was calculated as V(Out) − V(Orig). SWE was used to assess endometrial and JZ elasticity in the healthy control ( C and D ) and in the patient with adenomyosis-associated infertility ( G and H ). The white outline indicates the endometrial ROI, and the pink outline indicates the JZ shell ROI. In the measurement tables shown in panels ( D and H ) the white “A” column corresponds to endometrial SWE values, whereas the green “Shell” column corresponds to JZ SWE values. Representative volume and shear wave velocity values are displayed in the corresponding panels. For discovery proteomic analysis, four AM+INF samples were randomly selected from the 16 eligible AM+INF participants, and four HC samples were selected from the eight eligible HCs. The selected samples were not individually matched, although the overall AM+INF and HC groups showed no significant differences in age, BMI, or hormone profile ( Table 1 ). The selected serum sEV samples were processed and analyzed in the same experimental batch using the same standardized workflow to minimize technical variability; therefore, formal batch randomization was not performed. sEVs were isolated from human blood samples through a series of centrifugation steps. After isolation, the sEVs were characterized using nanoparticle tracking analysis (NTA) and fluorescent labeling followed by nanoparticle flow cytometry (NanoFCM) detection. Transmission electron microscopy (TEM) was employed for further imaging of the sEVs. For proteomic analysis, mass spectrometry (MS) was performed, beginning with sample preparation and tryptic digestion, followed by MS analysis. Exploratory enzyme-linked immunosorbent assay (ELISA) analysis was additionally performed for selected serum sEV proteins using the remaining samples, including 12 AM+INF and 4 HC samples. Detailed experimental procedures are provided in Supplementary File 1 : Detailed experimental procedures. The data were processed and analyzed using the Proteome Discoverer database search. Protein-level quantitative data were processed using DIA-NN 1.8.1, Perseus 2.0.7.0, and R 4.2. DIA identification results were filtered using Lib.Q.Value ≤ 0.01 and Lib.PG.Q.Value ≤ 0.01. Protein abundance values were log2-transformed before downstream comparative analysis. Proteins quantified in more than 50% of samples in at least one group were retained, and remaining missing values were imputed using a left-shift Gaussian distribution method. Table 1 Demographic and Baseline Characteristics of the Study Participants Characteristic Group t/Z P AM+INF, N =16 Control, N =8 Age(years) 32.94±4.39 35.75±3.32 −1.591 0.126 BMI (kg/m 2 ) 21.60±2.52 20.06±2.10 1.483 0.152 Type of infertility  Primary 5 (31.3%) NA NA NA  Secondary 11 (68.8%) NA NA NA Duration of infertility, years 1.50 (1.50–2.75) NA NA NA Laboratory measurements TESTO (ng/mL) 0.40±0.17 0.29±0.08 1.702 0.103 LH (mIU/mL) 4.61±1.37 3.87±1.24 1.286 0.214 FSH (mIU/mL) 6.25±2.57 5.61±1.22 0.655 0.519 AMH (ng/mL) 2.11±1.03 2.76±1.36 −1.316 0.202 CA125 (U/mL) 26.30 (18.98–47.23) 26.42 (21.44–32.69) 0.938 0.359 Ultrasound features Direct features  Myometrial cysts 6 (37.5%) 0 (0.0%) NA NA  Hyperechogenic islands 10 (62.5%) 0 (0.0%) NA NA  Echogenic subendometrial lines and buds 7 (43.8%) 0 (0.0%) NA NA Indirect features  Globular uterus 10 (62.5%) 0 (0.0%) NA NA  Asymmetrical thickening of myometrial walls 6 (37.50%) 0 (0.0%) NA NA  Fan-shaped shadowing 7 (43.8%) 0 (0.0%) NA NA  Translesional vascularity 2 (12.5%) 0 (0.0%) NA NA  Irregular junctional zone 11 (68.8%) 0 (0.0%) NA NA  Interrupted junctional zone 6 (37.5%) 0 (0.0%) NA NA Notes : Data are presented as mean ± SD, median (IQR), or n (%), as appropriate. Serum FSH, LH, and TESTO were measured on menstrual cycle days 2–6. Continuous variables were compared using Student’s t -test or Mann–Whitney U -test, as appropriate. The slash (/) in units denotes “per”. Abbreviations : AM+INF, adenomyosis-associated infertility; BMI, body mass index; TESTO, testosterone; LH, luteinizing hormone; FSH, follicle-stimulating hormone; AMH, anti-Müllerian hormone; CA125, cancer antigen 125; NA, not applicable. Demographic and Baseline Characteristics of the Study Participants Notes : Data are presented as mean ± SD, median (IQR), or n (%), as appropriate. Serum FSH, LH, and TESTO were measured on menstrual cycle days 2–6. Continuous variables were compared using Student’s t -test or Mann–Whitney U -test, as appropriate. The slash (/) in units denotes “per”. Abbreviations : AM+INF, adenomyosis-associated infertility; BMI, body mass index; TESTO, testosterone; LH, luteinizing hormone; FSH, follicle-stimulating hormone; AMH, anti-Müllerian hormone; CA125, cancer antigen 125; NA, not applicable. Statistical analyses were performed using SPSS software (version 26.0; IBM Corporation, Armonk, NY, USA), R language (version 4.2.0), and Python 3.12.0 in JupyterLab. Continuous variables with normal distribution are expressed as mean ± standard deviation and were compared using the independent-samples t -test. Non-normally distributed data are presented as median (interquartile range) and were compared using the Mann–Whitney U -test. Categorical variables are expressed as frequencies (n [%]) and compared using the chi-square test, with Fisher’s exact test applied when the expected cell counts were 10 considered to indicate multicollinearity. A two-sided P value <0.05 was considered statistically significant. Measurement reproducibility was assessed using ICCs and Bland–Altman analysis. ICCs were calculated using a two-way mixed-effects model based on absolute agreement, and single-measure ICCs with 95% confidence intervals were reported. ICC values were interpreted as follows: 0.90, excellent reproducibility. Machine-learning analysis was performed as an exploratory, hypothesis-generating analysis. The dataset was divided into a training set and an internal hold-out test set using a 7:3 stratified split with a fixed random seed, resulting in 16 training samples and 8 hold-out test samples. To reduce the risk of data leakage, all preprocessing and model-development steps were implemented within an imbalanced-learn pipeline consisting of standardization, L1-regularized logistic regression-based feature selection, SMOTE oversampling, and support vector machine (SVM) classification. During stratified five-fold cross-validation within the training set, the scaler, feature selector, SMOTE procedure, and SVM classifier were fitted only on the training folds, while validation folds were used only for performance estimation. The hold-out test set was not used during preprocessing, feature selection, oversampling, hyperparameter optimization, or model selection. Feature selection was restricted to the training data, and a maximum of three features was retained to reduce model complexity. SMOTE was applied only to the training data within the pipeline, with k_neighbors set to 2 because of the small number of minority-class samples. SVM hyperparameters were optimized using stratified five-fold cross-validation with ROC AUC as the optimization metric. Candidate hyperparameters included C values of 0.1, 0.5, 1, 5, and 10, linear and radial basis function kernels, and gamma settings of “auto” and “scale”. The final optimized model was refitted on the full training set and evaluated once on the internal hold-out test set. Model performance was assessed using area under the receiver operating characteristic curve (ROC AUC), accuracy, sensitivity, specificity, precision, F1-score, and the confusion matrix. The 95% confidence intervals for hold-out test-set metrics were estimated using 1,000 bootstrap resamples, with ROC AUC resamples containing only one class excluded. A permutation test using the same leakage-controlled pipeline was performed to assess whether model performance exceeded that expected under random label assignment. SHAP analysis was used only for exploratory interpretation of the final model. Given the small sample size and limited hold-out test set, the machine-learning results were interpreted cautiously and not considered evidence of a validated diagnostic model.

In this study, increased shear wave velocities in the JZ reflected altered biomechanical properties and increased stiffness of the uterine endometrial–myometrial interface in women with AM+INF, with JZSWVmin emerging as the most informative variable in the exploratory model for group discrimination. When interpreted together with the serum sEV proteomic findings, these results indicated co-occurring imaging and circulating proteomic alterations associated with AM+INF. Pathway analyses implicated ECM–receptor interaction and PI3K–Akt signaling, both of which are involved in tissue remodeling, cell adhesion, and cellular signaling. In addition, MSigDB C2 enrichment analysis suggested the involvement of ECM- and cell adhesion-related pathways, with reduced levels of candidate proteins such as GRB2, ANGPT1, ITGA5, and ITGB1 in the AM+INF group. Collectively, these observations suggest that increased JZ stiffness and altered serum sEV protein profiles may represent complementary features associated with ECM-related cellular interactions and signaling processes in AM+INF, although causal or tissue-specific relationships cannot be inferred from the present data. Ultrasound elastography has been increasingly used for the evaluation of adenomyosis, with most studies focusing on lesion, myometrial, or cervical stiffness for diagnosis or differential diagnosis. 20 Previous studies have shown that ultrasound elastography may improve the diagnostic performance of ultrasound for adenomyosis and help distinguish adenomyosis from uterine fibroids. 21 In addition to diagnostic applications, ultrasound elastography may also provide information on symptom-related biomechanical alterations in adenomyosis. Ren et al, used ultrasound elastography to assess tissue stiffness and found a correlation between tissue stiffness and fibrosis, suggesting that ultrasound elastography can provide an objective complement to subjective pain assessment tools such as the numerical rating scale and visual analog scale in evaluating dysmenorrhea severity in patients with adenomyosis. 13 Recent studies have also extended elastographic assessment beyond the uterine lesion itself. For example, cervical elastography studies have suggested that adenomyosis may be associated with altered cervical biomechanical properties, supporting the broader concept of adenomyosis-associated changes in reproductive tract stiffness. 14 In addition to uterine fibrosis, disordered uterine contractility, particularly involving the JZ, may be relevant to impaired embryo implantation and infertility. 22 Ultrasound elastography has also been used to assess uterine functional characteristics related to contraction. One study reported that lower uterine contraction frequency (1.7) were associated with higher pregnancy rates after intrauterine insemination. 23 However, many previous studies used strain elastography, which is operator-dependent and does not provide absolute tissue stiffness measurements. To our knowledge, the present study is the first to specifically quantify JZ stiffness using SWE in women with AM+INF under strictly controlled inclusion criteria. The increased shear wave velocity observed in the JZ suggested increased stiffness of the endometrial–myometrial interface, and the measurements showed good consistency and reproducibility. In this exploratory dataset, an SVM model based on SWE-derived features achieved 87.5% accuracy and showed favorable classification performance for distinguishing AM+INF patients from HCs. SHAP analysis further identified JZSWVmin as the most influential variable in this exploratory group-discrimination model; however, this parameter was retained as a data-driven feature in the machine-learning pipeline and should not be considered a stand-alone diagnostic biomarker. These findings suggest that SWE-derived JZ stiffness may represent a candidate imaging marker associated with AM+INF. However, because the model was developed using healthy controls rather than infertile women without adenomyosis or adenomyosis patients without infertility, these results should not be interpreted as evidence that SWE can specifically diagnose or predict adenomyosis-associated infertility in clinical practice. Further validation in larger cohorts with appropriate disease-control groups is required. Serum sEVs are membrane-enclosed vesicles released by various cell types and can be internalized by recipient cells to modulate their phenotype by transferring bioactive molecules, including lipids, proteins, DNA, mRNA, microRNAs, and other non-coding RNAs. Their molecular composition may vary depending on the cell of origin and disease state. 24 Accumulating evidence indicates that sEVs are widely present in reproductive tissues and biological fluids such as semen, follicular fluid, tubal fluid, and uterine fluid, suggesting that sEVs-mediated intercellular communication constitutes an important regulatory mechanism in the reproductive tract, in addition to the classical autocrine, paracrine, and endocrine signaling pathways. 25 Several studies have demonstrated that sEVs and their cargo play pivotal roles in the initiation and progression of adenomyosis. 26 In this context, sEVs contribute to pathological processes by modulating immune responses and remodeling the ECM, thereby promoting lesion infiltration, fibrosis, and local inflammation. Consequently, sEVs have emerged as promising biomarkers and potential therapeutic targets for adenomyosis. 27 Notably, recent studies have shown that sEV microRNAs such as miR-92a-3p, derived from adenomyotic lesions can enhance the migration and infiltration of endometrial cells, dorsal root ganglion neurons, and endothelial cells, thereby providing mechanistic insight into the development of adenomyosis-associated pain, particularly dysmenorrhea. 28 Despite the growing evidence supporting the role of sEVs in adenomyosis-related pain and disease progression, their potential value in predicting AM+INF is unknown. In this context, proteomic-based analyses of sEVs-derived proteins may offer novel insights into the molecular mechanisms underlying infertility in adenomyosis. In the present study, GO enrichment analysis of the 37 downregulated proteins revealed these proteins were involved in integrin-mediated signaling pathways, cell–substrate adhesion, and ECM-related biological processes. These functional categories are critical for maintaining normal endometrial–ECM interactions and transmitting mechanical and biochemical signals essential for endometrial receptivity. 29 The downregulation of proteins associated with integrin complexes and collagen binding suggested a compromised capacity of endometrial cells to sense and respond to ECM cues, which may impair cell–matrix adhesion and downstream signaling cascades. Several enriched biological processes were closely associated with embryonic development, gastrulation, and mesodermal formation. Although these processes are classically associated with early embryogenesis, increasing evidence indicated that maternal endometrial signaling and ECM–integrin interactions play pivotal roles in supporting early development and implantation. 30 Therefore, the reduced abundance of proteins related to these developmental pathways in patients with AM+INF may indicate circulating proteomic alterations associated with a less favorable reproductive phenotype. Taken together, these findings suggest that downregulated serum sEV proteins in AM+INF may be associated with ECM-related and integrin-dependent signaling processes. However, because serum sEVs may originate from multiple tissues and cell types, these alterations cannot be assumed to directly reflect molecular events at the uterine JZ, endometrium, or maternal–fetal interface. Thus, the potential relationship among adenomyosis-related structural abnormalities, altered circulating sEV protein profiles, impaired endometrial receptivity, and infertility should be regarded as hypothesis-generating and requires further validation. Furthermore, KEGG pathway analysis of the downregulated proteins highlighted ECM–receptor interactions and the PI3K–Akt signaling pathway. Given that ECM–integrin communication is essential for endometrial receptivity, the suppression of ECM-related signaling may indicate impaired cell–matrix crosstalk and a reduced capacity to sustain implantation-supportive signaling. 31 , 32 Notably, several KEGG disease pathways (eg, cardiomyopathy- and cancer-related terms) were also enriched, which likely reflects shared core modules, such as ECM remodeling, cytoskeletal organization, and PI3K–Akt signaling, rather than disease-specific processes. Collectively, these KEGG findings suggest that the downregulated serum sEV proteins in AM+INF were enriched in ECM-associated and signaling pathways, but they do not establish a direct mechanism of local uterine remodeling. To further explore the biological relevance of these downregulated proteins, PPI analysis combined with cytoHubba module screening was performed, identifying GRB2 as the hub protein with the highest node degree. GRB2 is a key adaptor protein that links integrin-mediated adhesion and growth factor signaling to downstream pathways, such as PI3K–Akt, which are involved in cell survival, migration, and tissue remodeling. 33 , 34 Its central position in the PPI network suggests that reduced GRB2 abundance may be associated with altered signaling-related protein interactions in AM+INF. However, because these findings were based on serum sEV proteomics and GRB2 has not yet been independently validated, its potential role in uterine JZ or endometrial dysfunction should be interpreted cautiously and requires further investigation. Finally, disease enrichment analysis using the MSigDB C2 database revealed that the downregulated proteins were predominantly enriched in five ECM- and adhesion-related pathways: Reactome platelet adhesion to exposed collagen, Reactome cell surface interactions at the vascular wall, KEGG ECM–receptor interaction, Reactome ECM proteoglycans, and Reactome non-integrin membrane–ECM interactions. These pathways were broadly consistent with the KEGG enrichment results, which highlighted ECM–receptor interaction and PI3K–Akt signaling, suggesting that ECM-mediated cell adhesion and signaling processes may be altered in the AM+INF group. Notably, candidate proteins involved in these enriched pathways, including GRB2, ANGPT1, ITGA5, and ITGB1, were downregulated in patients with AM+INF. Endometriosis and adenomyosis are thought to affect infertility through distinct mechanisms. Endometriosis primarily involves pelvic adhesions, inflammation, and immune cell recruitment, whereas adenomyosis has been associated with local endometrial inflammation, fibrosis, altered uterine contractility, and impaired implantation. 35 In a study by Tran et al, which excluded adenomyosis to focus on endometriosis, GRB2 was identified as an important regulator of endometrial receptivity and decidualization, and its loss was associated with infertility through progesterone resistance and dysregulation of steroid hormone and FOXA2 signaling pathways. 36 In the present study, GRB2 was also reduced in the serum sEV proteomic profile of patients with AM+INF, suggesting that GRB2 may be a candidate protein associated with the reproductive phenotype of AM+INF. However, because the present analysis was based on circulating serum sEVs rather than uterine tissue, uterine fluid, lesion-derived EVs, or paired endometrial/JZ proteomics, this finding should not be interpreted as direct evidence of altered GRB2 signaling within the uterine microenvironment. In addition, the decreased abundance of ITGA5, ITGB1, and ANGPT1 in the AM+INF group may be biologically relevant because these proteins are involved in cell–ECM adhesion, mechanotransduction, vascular stability, and tissue remodeling, processes that are important for endometrial receptivity and embryo implantation. 37–39 Reduced ITGA5 and ITGB1 abundance may indicate altered adhesion-related signaling, while decreased ANGPT1 abundance may be related to vascular or tissue-remodeling pathways. Nevertheless, these interpretations remain exploratory, as the candidate proteins have not yet been independently validated and serum sEV proteins may originate from multiple tissues and cell types. Taken together, these findings suggest that downregulated serum sEV proteins in AM+INF are associated with ECM-related, integrin-related, and vascular-remodeling pathways. Rather than establishing a mechanistic link between adenomyosis-related structural abnormalities and infertility, these results provide hypothesis-generating evidence that requires further validation in studies incorporating appropriate disease-control groups and tissue-level molecular analyses. This study had several limitations. This study had several limitations. First, the sample size was small, particularly for a study integrating SWE assessment, machine-learning analysis, serum sEV proteomics, and exploratory ELISA analysis. Therefore, the present findings should be considered preliminary and hypothesis-generating rather than confirmatory. Although SVC modeling, SMOTE, internal cross-validation, and permutation testing were used to reduce class imbalance, overfitting, and data-leakage risks, insufficient statistical power may still affect the robustness of the results. Accordingly, the machine-learning model was not intended for clinical prediction and requires external validation in larger independent cohorts. Second, the control group consisted only of healthy women. The absence of disease-control groups, such as infertile women without adenomyosis and women with adenomyosis without infertility, limits the ability to distinguish the independent effects of adenomyosis, infertility, and their combination. Thus, the observed imaging and proteomic alterations should be interpreted as features associated with AM+INF compared with healthy controls, rather than as AM+INF-specific findings. Third, adenomyosis was diagnosed using predefined ultrasonographic criteria without histopathological confirmation, which may have introduced diagnostic and clinical heterogeneity related to disease severity, lesion extent, and symptom profiles. Fourth, serum sEVs represent heterogeneous circulating vesicle populations that may originate from multiple tissues and cell types. Therefore, the observed serum sEV proteomic alterations cannot be assumed to directly reflect changes in the uterine JZ or endometrium, and their precise cellular or tissue origin could not be determined. Fifth, although ultrasound examinations and serum collection were performed during the proliferative phase, early and late proliferative phases were not further distinguished. Residual intra-phase variability in endometrial thickness, JZ morphology, tissue elasticity, uterine contractility, and circulating sEV profiles therefore cannot be excluded. In addition, proteomic discovery samples were not stratified by infertility duration or infertility type, which may have introduced additional biological heterogeneity. Future studies should include larger, independent, and clinically stratified cohorts with additional disease-control groups to validate the robustness, specificity, and clinical relevance of these candidate imaging and proteomic features. Despite these limitations, this study has several strengths. By integrating SWE-based imaging with serum sEV proteomic profiling, we observed co-occurring biomechanical and circulating proteomic alterations associated with AM+INF. Enrichment analyses indicated that the downregulated serum sEV proteins were related to ECM–receptor interaction and integrin-mediated signaling. SWE further provided a quantitative, noninvasive assessment of JZ stiffness, supporting its potential value for exploring biomechanical alterations at the endometrial–myometrial interface.

This exploratory study suggests that AM+INF is accompanied by increased JZ stiffness and serum sEV proteomic alterations involving ECM-related and PI3K–Akt signaling pathways. These findings may provide a basis for future studies investigating the relationship between uterine biomechanics and molecular dysregulation in AM+INF, but they do not establish causality. Larger, cycle-controlled, clinically stratified cohorts with independent validation are required before SWE parameters or serum sEV proteins can be considered diagnostic or clinically applicable biomarkers.