Development of a predictive model and nomogram in sperm retrieval rate based on testicular pathological morphometric parameters in non-obstructive azoospermia patients: a multi-center study

Abstract

Background

Microdissection testicular sperm extraction (micro-TESE) is an effective method to retrieve sperm from non-obstructive azoospermia (NOA) patients. However, the predictive factors for sperm retrieval rate (SRR) remain confused. The goal of our study was to identify the role of testicular pathological morphometric parameters, including diameter of tubule (DT), height of spermatogenic epithelium (HSE), and thickness of basement-membrane (TBM) in NOA patients, and to develop a predictive model and nomogram to predict SRR based on these morphometric parameters.

Methods

This study involved two cohorts including 406 men with NOA. A retrospective cohort of 313 males with NOA who underwent micro-TESE at Northwest Women’s and Children’s Hospital (Xi’an, China) were included to build a prediction model of SRR. Then, another retrospective cohort of 93 males with NOA from Ren Ji Hospital (Shanghai, China) were recruited to validate the prediction model. The measurement of testicular morphometric parameters as well as the assessment of Johnsen score and pathological diagnostic types were performed by at least two pathologists. Testicular volumes as well as level of serum hormones including follicle-stimulating hormone (FSH), luteinizing hormone (LH), and testosterone (T) were also measured. Logistic regressions were used to test potential predictors of SRR. Area under curve (AUC) estimates was used to evaluate the predictive accuracy. The validation datasets were used to validate the prediction model by prediction accuracy.

Results

Our study demonstrated that DT and HSE were significantly longer in successful sperm retrieval group than in failed sperm retrieval group. In addition, DT and HSE were positively correlated with Johnsen score, testicular volume, and serum T, while, were negatively correlated with serum FSH and serum LH. On the contrary, TBM demonstrated exact opposite results. Moreover, univariate logistic analyses illustrated that longer DT and HSE was associated with a high SRR, respectively. Further multivariate logistic analyses constructed multi-variables models with better predictive abilities compared with single-variables models. A multi-variables model (predicting score = -0.612–0.018 × DT + 0.040 × HSE + 0.097 × Johnsen score-0.004 × serum FSH) was finally constructed with the best predictive ability (AUC = 0.839, sensitivity = 71.4% specificity = 77.5%, cut-off value = 0.489). A higher predicting score indicated a higher possibility of successful sperm retrieval. The predictive accuracy was 89.25% in the external validation dataset.

Conclusion

We report for the first time that DT and HSE have pretty ability to predict SRR in NOA patients.

Introduction

The relevance of male infertility has progressively grown [1, 2], with male factors accounting for approximately 30%−50% of infertility cases [3]. Azoospermia, defined as the complete absence of spermatozoa in the ejaculate, even after centrifugation on at least two occasions, is the most severe form of male infertility [4]. Among the various types of azoospermia, about 60% of patients are diagnosed with non-obstructive azoospermia (NOA), which results from inactive spermatogenesis despite an unobstructed genital tract [5]. NOA affects approximately 1% of the general male and 10% of infertile population [6].

Microdissection testicular sperm extraction (micro-TESE) combined with intracytoplasmic sperm injection (ICSI) offers NOA patients a chance to achieve biological parenthood. Micro-TESE is currently the first-line treatment for NOA, since this gold-standard technique could improve sperm yield with minimal tissue excision [3, 7, 8]. However, the success rate of sperm retrieval is only approximately 50% [9, 10]. A failed micro-TESE procedure can impose significant financial and emotional burdens on infertile couples. Therefore, identifying predictive factors for successful sperm retrieval is crucial to guiding clinical decision-making.

Notably, some studies have explored clinical predictors of sperm retrieval rate (SRR), such as serum follicle-stimulating hormone (FSH), serum luteinizing hormone (LH), testicular volume, body mass index (SRR) [11–13]. However, the predictive ability of these individual clinical parameter is limited due to the low sensitivity and specificity [14].

Recently, apart from the clinical characteristics, testicular histology/pathology parameters, which more directly reflect spermatogenic function, were confirmed to be more reliable predictors of SSR in NOA patients. One of the most common pathological parameters is testicular pathological diagnosis. Previous studies have indicated that patients with hypospermatogenesis (HS) generally have higher SRRs, while those with Sertoli cell-only syndrome (SCOS) have significantly lower SRRs [15–18]. Johnsen score, another widely used pathological parameter, was also verified to be a predictive factor for SRR [19, 20]. However, both testicular pathological diagnostic type and Johnsen score are subjective and may vary among pathologists or laboratories, introducing potential bias. Hence, some objective testicular pathological morphometric parameters have garnered attention, such as the diameter of tubule (DT), the height of spermatogenic epithelium (HSE) and the thickness of basement-membrane (TBM).

DT had been confirmed to be a useful factor in predicting SRR. Amer et al. [21] reported that when the intraoperative length of DT is ≥ 300 µm, a single tubule biopsy is usually sufficient to harvest sperm. Yu et al. [22] indicated that the best cutoff of the intraoperative DT is 100 µm in NOA patients. However, in these surgeries, only a maximum six tubules were assessed in per testis, which limited responsibility to reflect the spermatogenic function of the whole testis.

Our study focused on investigating the possible role of testicular pathological morphometric parameters in NOA as well as on developing a predictive model and nomogram to estimate SRR based on these objective morphological parameters.

Materials and methods

Study population

We performed this study with a retrospective data collected from NOA patients who underwent a micro-TESE procedure from April 2015 to December 2020 in Northwest Women’s and Children’s Hospital (Xi’an, Shaanxi, China). These data were collected to build the prediction model. Thereafter, the model was then validated using another prospective dataset collected from March 2021 to February 2022 in Renji Hospital (Shanghai, China). All the patients were informed that their tissues and related medical information would be used anonymously for medical research and were requested to provide a written informed consent. The present study complied with the Standards for the Reporting of Diagnostic Accuracy Studies (STARD) guidelines and was approved by the Institutional Medical Ethics Committee of Xi’an Jiaotong University (Protocol number: 2021133).

Each patient had a history of infertility of ≥ 12 months and was confirmed to represent as the complete absence of sperm in no less than two semen analyses with centrifugation according to the 6th edition of the WHO Laboratory Manual for the Examination and Processing of Human Semen [23]. Patients with known hypothalamic/pituitary defects, previous testicular tumors, or any other known reason for genital tract obstruction were excluded.

Serum hormone determination

Venous blood samples were drawn from each patient in the morning. FSH, LH, and T (testosterone) were measured by chemiluminescent enzyme immunometric assays using commercial kits from Roche.

Testicular sperm retrieval

In our study, standard micro-TESE surgery was used to retrieval sperm performed by the same trained urologist in each hospital. A median incision was made on the scrotum under general anaesthesia to adequately expose the testis. The length, width, and height of testis were measured to estimate the volume of testis. Thereafter, a transverse equatorial plane incision on the tunica albuginea of the testis was cut to reveal the seminiferous tubules under the professional operating microscope. After rinsing the testis surface with a balanced salt solution, tubules with full appearing and opaque, which have possible sperm production, were gently excised and placed in a petri dish. The same experienced embryologist in each hospital dissected the seminiferous tubules and assessed the presence of sperm under magnification. The test of sperm retrieval was considered successful when at least one viable spermatozoon was available for future use in ICSI.

Meanwhile, in the same position of the excised tubules, a large fragment of testicular tissue of about 8 × 4 × 3 mm³ was cut out, no matter whether the sperm was harvested successfully. The tissue was immediately fixed in Bouin’s solution and then sent for pathological analysis.

Testicular pathological morphometry

The testicular tissues, fixed in Bouin’s solution for 24 h, were sent to the pathologist. Each fixed sample was immersed in paraffin, sectioned with 4um thickness, and stained with haematoxylin and eosin (H&E).

The histological evaluation was performed by two independent pathologists using Olympus microscope BX51 (X400 magnification). Any discrepancies were addressed by the third re-evaluators and discussed to reach a consensus on all details of each slide. The pathologists were blinded to sperm retrieval outcomes. Each slide was scanned at least 50 different tubules. Firstly, typical pathological diagnostic types were determined defined as follows: (i) hypospermatogenesis (HS), defined as a severely reduced population of spermtagenesis with all germ cell types present; (ii) maturation arrest (MA), characterized with a premature arrest of spermatogenesis; and (iii) Sertoli cell–only syndrome (SCOS), defined as a complete absence of germ cells. Secondly, pathological scoring of human spermatogenesis was also preformed according to the criteria suggested by Johnsen [20].

Moreover, the last and the most important characteristics were testicular pathological morphometric parameters, including DT, HSE and TBM. Based on the protocol of measurement in previous literature [24, 25], 20 circular or similar-circular tubules were randomly selected to measure these parameters. As shown in Fig. 1, DT should be measured from the outer edge of the basement-membrane to the opposite outer edge of the basement-membrane. HSE should be measured from the inner edge of the spermatogenic epithelium to the ipsilateral outer edge of the spermatogenic epithelium. TBM should be measured from the outer edge of the basement-membrane to the ipsilateral inner edge of the basement-membrane. The mean value of the 20 tubules of each patient was calculated as the final data of this individual.

Fig. 1.

The pattern of measurements

Statistical analysis

Normally distributed continuous variables were expressed as mean ± standard deviation (SD) and were analyzed using Student’s t-test and one-way analysis of variance (ANOVA) test. Non-normally distributed continuous variables were expressed as median (IQR) and were analyzed using non-parametric tests.

Univariate and multivariate logistic regression analysis were conducted to assess the influencing factors of sperm retrieval in micro-TESE. Influencing factors used in the present study included testicular pathological morphometric parameters, Johnsen score, pathological diagnostic types, age, sex hormone levels, and testicular volume. All the variables were selected using the stepwise backward method. Subsequently, the receiver operating characteristic (ROC) curve analysis was conducted to calculate the area under the curve (AUC) and the optimal cutoff value (Youden’s index) was predicted for the likelihood of successful sperm retrieval.

Models were further externally validated in another study cohort in Ren Ji Hospital (Shanghai, China). Each individual in the validation datasets calculated the predicting outcomes of sperm retrieval using the final predicting model and its cut-off value. The accuracy was calculated to describe the generalizability of our final model.

All statistical analyses were performed in SPSS statistical software version 22.0 (IBM Corp) and R version 4.2.0 (R Project for Statistical Computing). A 2-tailed P value < 0.05 was considered to be statistically significant.

Results

The overall workflow pattern of this study is shown in Fig. 2. A total of 406 NOA patients were finally included in this study, including 313 participants in modelling dataset and 93 participants in validation dataset.

Fig. 2.

Flowchart showing an overview of our study. AUC: area under the curve; DT: diameters of tubule; FSH: follicle-stimulating hormone; HSE: height of spermatogenic epithelium; JS: Johnsen score; LH: luteinizing hormone; micro-TESE: microdissection testicular sperm extraction; ROC: receiver operating characteristic; TBM: thickness of basement-membrane; T: testosterone

Participant characteristics

In modelling dataset, the detail of clinical characteristics was summarized in Table 1.

Table 1.

Comparison of baseline characteristics, clinical features and pathological parameters in NOA patients stratified according to outcome of sperm retrieval

Variable	Overall	Successful Sperm Retrieval	Failed Sperm Retrieval	P value
	(N = 313)	(N = 72)	(N = 241)
Testicular Pathological Morphometric Parameters
DT (μm), median (IQR)	113 (84, 138)	119 (82, 165)	112 (88, 127)	0.001*
HSE (μm), median (IQR)	25 (17, 38)	32 (25, 68)	21 (16, 28)	< 0.001*
TBM (μm), median (IQR)	11 (6, 19)	12 (6.5, 21)	10 (6, 16)	0.074
Testicular Traditional Pathological Parameter
Johnsen score, median (IQR)	2 (2, 6)	5.5 (2, 7.5)	2 (2, 3)	< 0.001*
Pathological type
HS, number (percentage)	45	31 (68.89%)	14 (31.11%)	0.001*
MA, number (percentage)	55	11 (20.0%)	44 (80.0%)
SCOS, number (percentage)	213	30 (14.08%)	183 (85.92%)
Clinical Characteristics
Age (y), median (IQR)	30 (28, 32)	30 (28, 33)	30 (27, 32)	0.865
Testicular volume (ml), median (IQR)	8 (6, 10)	8 (6, 10)	8 (6, 10)	0.199
Serum FSH (mIU/ml), mean ± SD	23.67 ± 15.92	20.58 ± 14.94	24.82 ± 16.16	0.035*
Serum LH (mIU/ml), median (IQR)	8.48 (5.99, 11.95)	8.17 (5.82, 12.245)	8.63 (6.025, 11.875)	0.235
Serum T (ng/ml), mean ± SD	338.27 ± 242.93	371.55 ± 359.21	325.88 ± 181.18	0.137

DT diameter of tubule, FSH follicle-stimulating hormone, HS hypospermatogenesis, HSE height of spermatogenic epithelium, LH luteinizing hormone, MA maturation arrest, SCOS Sertoli cell–only syndrome, T testosterone, TBM thickness of basement-membrane

^*: P < 0.05 is significant

The median (IQR) of age, testicular volume, and serum LH was 30 (28, 32) year, 8 (6, 10) ml, 8.48 (5.99, 11.95) mIU/m, respectively. The mean ± SD of serum FSH and serum T was 23.67 ± 15.92 mIU/m and 338.27 ± 242.93 ng/ml, respectively.

As for the pathological examinations, participants could be classified into HS (45 individuals, 14.38%), MA (55 individuals, 17.57%), and SCOS (213 individuals, 68.05%). The median (IQR) of Johnsen score, DT, HSE, and TBM was 2 (2, 6), 113 (84, 138) um, 25 (17, 38) um, and 11 (6, 19) um, respectively.

Performance of characteristics between successful and failed sperm retrieval groups

Based on the result of sperm retrieval surgery, 72/313 (27.48%) individuals were in successful sperm retrieval group and 241/313 (72.52%) males were in failed sperm retrieval group.

In order to assess the predictive value of SRR, testicular pathological morphometric parameters, Johnsen score, pathological diagnostic types, patient age, testicular volume, and serum hormone levels (including FSH, LH, and T) were compared between successful and failed sperm retrieval groups.

DT (P = 0.001, Fig. 3, Table 1) and HSE (P < 0.001, Fig. 3, Table 1) were significantly longer in successful sperm retrieval group compared to that in failed sperm retrieval group. However, TBM did not showed any meaningful differences between the two groups. Johnsen score was significantly higher in successful sperm retrieval group than that in failed sperm retrieval group (P < 0.001, Fig. 3, Table 1). In HS, MA, and SCOS groups, the number of patients with successful sperm retrieval was 31 out of 45 (68.89%), 11 out of 55 (20.00%), and 30 out of 213 (14.08%), respectively. The distribution of three pathological type in different sperm retrieval outcomes was statistically significant (P = 0.001, Table 1).

Fig. 3.

The differences of clinical/pathological parameters between successful and failed sperm retrieval groups in NOA patients. ****: P < 0.001, ***: P < 0.005, and *: P < 0.05 were statistic significant. DT: diameters of tubule; FSH: follicle-stimulating hormone; HSE: height of spermatogenic epithelium; JS: Johnsen score

The level of serum FSH was lower in successful sperm retrieval group (P = 0.035, Fig. 3, Table 1). While, no statistically differences of age, testicular volume, serum LH, and serum T were seen between the two groups (Table 1).

Associations of testicular pathological morphometric parameters with clinical characteristics in NOA patients

The correlations between testicular pathological morphometric parameters and clinical characteristics of NOA patients were exhibited in Table 2.

Table 2.

The associations between testicular pathological morphometric parameters and clinical-pathological parameters of NOA patients

Variable	DT (μm)		HSE (μm)		TBM (μm)
	rho	P value	rho	P value	rho	P value
Testicular Traditional Pathological Parameter
Johnsen score	0.805	< 0.001*	0.752	< 0.001*	−0.562	< 0.001*
Clinical Characteristics
Age (y)	−0.660	0.249	0.024	0.678	0.038	0.506
Testicular volume (ml)	0.370	< 0.001*	0.339	< 0.001*	−0.214	0.028*
Serum FSH (mIU/ml)	−0.431	< 0.001*	−0.326	< 0.001*	0.261	< 0.001*
Serum LH (mIU/ml)	−0.392	< 0.001*	−0.280	< 0.001*	0.299	< 0.001*
Serum T (ng/dl)	0.198	0.001*	0.074	0.203	−0.146	0.012*

DT diameter of tubule, FSH follicle-stimulating hormone, HSE height of spermatogenic epithelium, rho spearman correlation coefficient, LH luteinizing hormone, T testosterone, TBM thickness of basement-membrane

^*: P < 0.05 is significant

DT and HSE were positively correlated with Johnsen score (P_DT < 0.001, rho_DT = 0.805; P_HSE < 0.001, rho_HSE = 0.752) and testicular volume (P_DT < 0.001, rho_DT = 0.370; P_HSE < 0.001, rho_HSE = 0.339) as well as negatively correlated with serum FSH (P_DT < 0.001, rho_DT = −0.431; P_HSE < 0.001, rho_HSE = −0.326) and serum LH (P_DT < 0.001, rho_DT = −0.392; P_HSE < 0.001, rho_HSE = −0.280). Inversely, TBM demonstrated exact opposite results. In addition, serum T showed positive association with DT (P < 0.001, rho = 0.198), negative association with TBM (P = 0.012, rho = −0.146), and no significant association with HSE (P = 0.203). Moreover, no notable association was seen with age.

The predictive value of testicular pathological morphometric parameters in testicular sperm retrieval in NOA patients

Univariate logistic analysis showed that longer DT (P = 0.001; OR = 1.010; AUC = 0.601, Fig. 4a), longer HSE (P < 0.001; OR = 1.026; AUC = 0.727, Fig. 4a), and higher Johnsen score (P < 0.001; OR = 1.339; AUC = 0.633, Fig. 4a) had statistically significant effect on successful sperm retrieval (Table 3). Additionally, compared with SCOS, HS group had significant effect on successful sperm retrieval (Table 3). On the other hand, lower level of serum FSH (P = 0.033; OR = 0.978; AUC = 0.600, Fig. 4a) were risk factors for SRR (Table 3). TBM, age, testicular volume, serum LH, and serum T exhibited little statistical significance in univariate logistic analysis.

Fig. 4.

Receiver operating characteristic (ROC) curves of the prediction models in NOA patients. a: univariate prediction models; b: multivariate prediction model. AUC: area under the curve; DT: diameters of tubule; FSH: follicle-stimulating hormone; HSE: height of spermatogenic epithelium; JS: Johnsen score

Table 3.

Univariate analysis of the contributing factors for successful sperm retrieval in NOA patients

Variable	Univariate analysis
	OR	95% CI		P value	AUC	Sensitivity	Specificity	Cut-off
		low	high
Testicular Pathological Morphometric Parameters
DT (μm)	1.010	1.004	1.016	0.001*	0.601	0.616	0.848	0.232
HSE (μm)	1.026	1.017	1.035	< 0.001*	0.727	0.802	0.576	0.378
TBM (μm)	0.972	0.941	1.003	0.074	-	-	-	-
Testicular Traditional Pathological Parameter
Johnsen score	1.339	1.204	1.489	< 0.001*	0.633	0.570	0.873	0.311
Pathological type
HS	8.656	4.244	17.654	< 0.001*	-	-	-	-
MA	0.977	0.467	2.046	0.951	-	-	-	-
SCOS	Reference				-	-	-	-
Clinical Characteristics
Age (y)	1.005	0.948	1.066	0.865	-	-	-	-
Testicular volume (ml)	1.075	0.963	1.003	0.198	-	-	-	-
Serum FSH (mIU/ml)	0.978	0.959	0.998	0.033*	0.600	0.612	0.766	0.255
Serum LH (mIU/ml)	0.976	0.937	1.016	0.235	-	-	-	-
Serum T (ng/ml)	1.000	0.999	1.002	0.564	-	-	-	-

DT diameter of tubule, FSH follicle-stimulating hormone, HS hypospermatogenesis, HSE height of spermatogenic epithelium, LH luteinizing hormone, MA maturation arrest, SCOS Sertoli cell–only syndrome, T testosterone, TBM thickness of basement-membrane

^*: P < 0.05 is significant

Multivariate logistic analysis was performed using factors which had significant associations with SRR in the univariate analysis. Considering that Johnsen score and pathological diagnostic types existed significant colinearity (IVF = 17.658), we decided to included the Johnsen score (ranked data) but not pathological diagnostic types (ordered categorical data) in further multivariate logistic regression models. Hence, the multivariate prediction model finally included DT, HSE, Johnsen score, and serum FSH. The four variables were randomly combined to construct multi-variables models. 11 multi-variables models were finally applied, including six possible two-variables models, four possible three-variables models, and a possible four-variables models (Fig. 4b). ROC analysis indicated that the combined model of four-variables showed the best evaluation indicators in predicting performance (AUC = 0.839, the cutoff value was 0.489, with a sensitivity of 71.4%, and a specificity of 77.5%). Multivariate logistic regression model developed a derived formula of this combined model of four-variables as follows: predicting score = −0.612—0.018 × DT + 0.040 × HSE + 0.097 × Johnsen score—0.004 × serum FSH. When the predicting score > 0.489, the male had a high possibility of successful sperm retrieval (Fig. 5).

Fig. 5.

Distribution of predicting score and the value of parameters. All the data were log-transformed. DT: diameters of tubule; FSH: follicle-stimulating hormone; HSE: height of spermatogenic epithelium; JS: Johnsen score

A nomogram was constructed to visualize the model as shown in Fig. 6. The longer the length of the line, the greater the impact of variables on the presence of SRR. Our model indicated that HSE had the greatest impact on the presence of SRR.

Fig. 6.

Nomogram for predicting the probability of the successful sperm retrieval outcomes. DT: diameters of tubule; FSH: follicle-stimulating hormone; HSE: height of spermatogenic epithelium; JS: Johnsen score

External validation

We validated the predicting model as described above in an external cohort of 93 NOA patients in Ren Ji Hospital (Shanghai, China). The prediction results in validation dataset showed an overall accuracy of 89.25% (83 out of 93, Fig. 7). This finding suggested that our prediction model generally worked well in the external validation dataset from different center, and both the prediction accuracy and generalizability of the model were validated.

Fig. 7.

Predicted and observed outcomes of sperm retrieval in the validation sets

Discussion

To our knowledge, this is the first study to identify the testicular pathological morphometric parameters (DT, HSE, and TBM) in pathological sections of testis from NOA patients. We also the first to assess the predictive value of these morphometric parameters in SRR.

Our results firstly indicated that DT and HSE were dramatically longer in successful sperm retrieval group compared to the failed group. Secondly, DT and HSE were positively correlated with Johnsen score, testicular volume, and serum T, while, negatively correlated with serum FSH and serum LH. Thirdly, univariate logistic regression models demonstrated that patients with longer DT or HSE will have a higher SRR. Notably, HSE had the strongest influence on SRR. These findings illustrated that testicular pathological morphometric parameters play crucial roles in predicting SRR in our study. Fourthly, multivariate logistic regression constructed a multi-variables model with best predictive ability (formula: predicting score = −0.612—0.018 × DT + 0.040 × HSE + 0.097 × Johnsen score—0.004 × serum FSH). This model showed an AUC of 0.839, a sensitivity of 71.4%, a specificity of 77.5%, and a cut-off of 0.489. Fifthly, an external cohort validated our multi-variables model with a pretty predictive accuracy (89.25%).

The spermatogenic epithelium comprises a large amount of spermatogenic cells and the height of spermatogenic epithelium can reflect the condition of spermatogenesis directly. In animals, the seminiferous epithelium increased from peripuberty to reproductive maturity [26]. Exposure to reproductive toxins could decrease the height of seminiferous epithelium [27], while certain dietary supplements could increase it [28]. Hence, HSE had potential to be a unique and independent predictive factor for SRR, which was confirmed by our findings. In addition, DT could also reflect the function of spermatogenesis and could be identified as a potential predictor for SRR [22]. Our study verified this point, yet. Our findings also emphasized that HSE have greater impact on SRR when compare with DT. This may result from that seminiferous tubules are hollow-tubular structure. Some tubules may include large luminal cavities inside (Fig. 1), and DT contains these cavities, which could limit the ability to reflect the true condition of spermatogenesis. Although TBM did not reach significance, it showed an inverse correlation with male fertility. Interstitial fibrosis could thicken the basement-membrane, and the presence of focal interstitial fibrosis was reported to be associated with a high incidence of infertility [29]. In addition, the increase in interstitial area was correlated with Johnsen scores [30]. Apart from human testis, in dog testis, an expanded interstitial area correlates with senescence-related abnormal spermatogenic function [26]. In the present study, the lack of significance for TBM might be attribute to the small sample size, and further studies with larger sample size are needed to evaluate the predictive value of TBM in SRR.

In recent five years, several predictive models for SRR in NOA patients have been reported, utilizing both traditional and novel factors (Table 4). For instance, Liu et al. [31], Zhang et al. [32], and Majzoub et al. [33] developed predictive models based on traditional factors including hormone levels and clinical parameters. However, their predictive performances (AUC) inferior to that of our model. Other models targeting specific etiologies (such as cryptorchidism [34] and varicocele [35]) achieved higher predictive performance (AUC for cryptorchidism = 0.907, AUC for varicocele = 0.850), likely due to the focused nature of the populations studied. These inspire future refinements of our model for idiopathic NOA. Bachelot et al. [36] employed machine learning to build a predictive model. However, its internal validation performance was modest (AUC = 0.780), possibly due to a limited sample size. In addition, although external validation was included in this study, the dataset for external validation were obtained from the same hospital, which might weaken the model’s generalizability. Beyond traditional factors, recent studies have explored innovative predictors, such as of DNA/RNA expression [37–39], agglutinin (galectin [40]), and magnetic resonance imaging [41]. While these approaches show promise, they require costly infrastructure and training, including molecular labs, specialized imaging equipment, and technician expertise. Additionally, differences among various molecular assay kits and subjective interpretation of imaging data may introduce bias, limiting generalizability. In contrast, as for DT and HSE, the morphometric parameters used in our model, are achieved only from the most basic pathological HE staining under a microscope and microimaging system. This approach minimizes inter-laboratory variability, reduces initial costs. Moreover, DT and HSE are objective and stable data, meaning that the subjective bias from different technicians could be negligible.

Table 4.

The details of models for predicting sperm retrieval rate in NOA patients by studies published in recent five years

Number	Type of surgery	Sample size	Included factors	AUC	Region	External validation
The present study	micro-TESE	406	diameter of tubule + height of spermatogenic epithelium + Johnsen score + FSH	0.839	Northwestern China	√
Liu et al. [31]	micro-TESE	294	FSH	0.630	Central China	×
	TESA		FSH	0.700
Zhou et al. [37]	micro-TESE	62	Beclin-1 expression + Johnsen score	0.848	Northwestern China	×
Chen et al. [34]	micro-TESE	162	position of cryptorchidism + unilateral and bilateral cryptorchidism + age + volume	0.907	Southeastern China	×
Din et al. [40]	surgical microscope	48	Gal-1 in seminal plasma	0.858	Egypt	×
Kaltsas et al. [35]	micro-TESE	78	varicocele grade + histopathological findings + age + preoperative hormonal levels (FSH, LH, T) + and the presence of bilateral varicocele	0.850	Greece	×
Gao et al. [41]	micro-TESE	70	mean diffusivity from Diffusion tensor imaging	0.865	Southeastern China	×
Zhang et al. [32]	TESE	3093	FSH	0.625	Turkey	×
			LH	0.597
			Testosterone	0.550
			Length of infertility	0.506
			Used testis volume	0.609
Bachelot et al. [36]	TESE	175 (Internal validation)	age + BMI + tobacco consumption + hormonal assessment (FSH, LH, T, inhibin B, and prolactin) + genetic exploration (karyotype and search for Y-chromosome microdeletion), and urogenital history (cryptorchidism, infection, trauma, gonadotoxic therapy, urogenital surgery, and varicoceles)	0.780	France	√
		26 (External validation)		0.900
Ji et al. [38]	micro-TESE	58	3 circRNAs (hsa_circ_0000277, hsa_circ_0060394 and hsa_circ_0007773) in seminal plasma	0.958	Southeastern China	×
Xie et al. [39]	micro-TESE	30 (Training set)	9 exlncRNAs (including LOC100505685, SPATA42, CCDC37-DT, GABRG3-AS1, LOC440934, LOC101929088, LOC101929088, LINC00343 and LINC00301) in seminal plasma	0.986	Southern China	×
		66 (Validation set)		0.960
Majzoub et al. [33]	micro-TESE	297	FSH + testis size	0.742	Qatar	×

Furthermore, previous studies [21, 22] reported DT as a useful predictor for SRR. However, the measurement data used in these studies were obtain during the operation using a micrometer fixed to one eyepiece of the operating microscope. The barriers of instruments and technology in microsurgery limited the promotion of their findings. Our study performed the measurement based on pathological section using the microscope. It can significantly mitigates the limitations mentioned upon and enables more efficient subsequent verification.

Remarkably, manual measurement of testicular pathological morphometric parameters is expected to evolve into automated or Artificial intelligence (AI)-based measurement systems and harvesting prediction value of SRR through mechanized assembly line system can be expected. AI has emerged as a powerful tool for medical images analysis and computer-aided histopathological diagnosis [42–46]. Most applications in pathology are currently focused on tumor diagnosis. For example, Huang et al. developed models using AI based on HE-stained pathological slides to diagnose gastric cancer and predict its outcome [47]. Wang et al. devised a Clinical Histopathology Imaging Evaluation Foundation model, a general-purpose weakly supervised machine learning framework to extract pathology imaging features for systematic cancer evaluation [48]. Recently, few studies concentrated on AI-assisted pathology to the diagnosis or prognosis of male infertility. The present study indicated that pathological morphometric parameters play crucial roles in predicting SRR in NOA patients. These findings lay a theoretical foundation for further clinical translation of AI-based histopathological diagnosis. Based on our findings and applying Wang’s general-purpose weakly supervised machine learning framework, further study is in need to develop an AI model for improved prediction of SRR.

However, the clinical application of pathological parameters for SRR prediction has been questioned due to some criticisms, since these data are obtained through invasive testicular surgery [17]. In clinical practice, some patients with failed sperm retrieval pursue salvage micro-TESE at referral centers before opting for donor insemination. Prior pathological assessments could provide more valuable information to guide decision-making, helping clinicians and patients determine whether to undergo another surgical attempt or seek help from sperm bank. Moreover, our findings were also applicable to patients undergoing a testicular biopsy before attempting micro-TESE. This can offer guidance for subsequent treatment, potentially reducing unnecessary invasive procedures and lowering diagnostic and treatment costs. And our model is engaged to be reproduced in the future using percutaneous core biopsies results. Furthermore, only one external validation cohort from southeastern China were included in the present study. Some more prospective cohorts from the whole China should be added to develop a nationally applicable model. Benefiting from its generalizability, the future model even has opportunity to be incorporated into the clinical decision-making workflows.

Several limitations should be acknowledged. Firstly, some certain SRR related factors, such as bad lifestyle habits (e.g., smoking [49], higher body mass index [50]) and novel hormonal markers [e.g., Inhibin B [51], anti-Müllerian hormone (AMH) [52, 53]] were not included. Further studies with better design and abundant data are worth expecting. Secondly, the sample size in the present study was slightly below the threshold recommended by Riley’s standard [54, 55], underscoring the need for larger studies to improve model reliability. Thirdly, considering intra-testicular heterogeneity, single-point measurement of DT may not represent the overall condition of the testis, potentially overestimating spermatogenic capacity. Hence, future studies should employ multi-point sampling from micro-TESE or sampling from fine-needle aspiration to minimize selection bias. Fourthly, apart from tubules size, other histological features, such as tubule count and the proportion of fibrotic-hyalinosic tissue per unit area, are also important for evaluating spermatogenic function. Additionally, apart from traditional H&E staining, some other pathological technique (such as immunostaining) can further aid in accurate identification of germ cells and improve diagnostic precision. These novel parameters may contribute to the development of more robust predictive models. Fifthly, the present study focused solely on sperm retrieval outcomes and did not track downstream outcomes such as fertilization or live birth. Future research should include follow-up data on pregnancy rates and live births following micro-TESE.

Conclusion

Our study was the first to report the significant predictive value of testicular pathological morphometric parameters—specifically DT and HSE—for sperm retrieval outcome in NOA patients. Longer DT and longer HSE were associated with a higher likelihood of successful sperm retrieval. Based on these findings, we developed a muti-variables model to predict SRR with a pretty predictive performance, which was further validated by an external cohort. Our findings would offer valuable insights to support clinical decision-making of NOA patients.

Authors’ contributions

Hong-xiang Wang: Funding acquisition, Resources. Jia-xi He: Investigation. Yi-min Guo: Investigation. Liang Zhou: Resources. Si-xuan Li: Data curation. Zi-tong He: Data curation. Qi-ya Jing: Data curation. Pei-quan Wang: Data curation. Liu-qing Qu: Data curation. Jun-cheng Gao: Software. Guan-chen Liu: Software. Hai-xu Wang: Resources. Yan-qi Yang: Formal analysis. Pan Ge: Formal analysis. Jian Zhang: Methodology. Xiao-ting Wang: Conceptualization, Resources, Supervision. Mo-qi Lv: Conceptualization, Resources, Methodology, Formal analysis, Funding acquisition, Data Curation, Writing—original draft. Hai-ge Chen: Funding acquisition, Resources, Supervision. Dang-xia Zhou: Conceptualization, Resources, Funding acquisition, Supervision, Writing—review & editing.

Funding

This research was financially supported by the National Natural Science Foundation of China (No. 81673224, 81273018, and 82173076), the China Postdoctoral Science Foundation (No. 2022M722540), the Natural Science Foundation of Shaanxi Province (No. 2023-JC-QN-0819), the Shanghai Natural Science Foundation (No. 16ZR1420300, 18410720400, 19431907400, and 20Y11904900), the Shanghai Municipal Health Commission for advanced and suitable technology promotion projects (No. 2019SY056), the Innovation Project for Medical Integration in XJTU (No. YXJLRH2022080), the Shanghai Jiao Tong University School of Medicine Research Funding Projects (No. TM201708), the Ren Ji Hospital Research Funding Projects (No. RJZZ18-020, PYIII-17–017, and PY2018-IIC-02), the Foundation of Shanghai Hospital Development Center (No. SHDC12015125), and the Clinical Research Plan of SHDC (No. SHDC2020CR4035).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

The study was approved by the Institutional Medical Ethics Committee of Xi’an Jiaotong University (Protocol number: 2021133). Clinical trial number: not applicable.

Consent for publication

Not applicable.

Competing interests

None declared.