Transvaginal 3-Dimensional Ultrasonic Image for the Diagnostic Effect of Intrauterine Adhesion

(e aim was to analyze the application values and diagnostic effects of transvaginal 3-dimensional (3D) ultrasonic image based on extreme learning machine denoising algorithm (ELMDA) in the diagnosis of intrauterine adhesions (IUA). (e speckle noise in the 3D ultrasound image was removed with the ELMDA. Its peak signal-to-noise ratio (PSNR) and the mean square error (MSE) were compared with those of the median filter algorithm (MFA) with the anisotropic diffusion algorithm (ADA) and wavelet threshold. (e ELMDA was used in the diagnosis of 3D ultrasound images to compare the accuracy of hysteroscopy with transvaginal 3D ultrasound and two-dimensional (2D) ultrasound in the diagnosis of IUA. (e results showed that the MSE of ELMDA was dramatically smaller than those of ADA and WT-MFA and its PSNR was higher than those of the other two algorithms (P< 0.05) when the noise variance was constant. (e diagnostic accuracy of mild and moderate adhesions by 2D ultrasound was statistically different (P< 0.05) compared with hysteroscopy. But the diagnosis results of severe adhesions were consistent, and the diagnosed cases were both 6 (11.11%) with no statistical difference (P> 0.05). In addition, there was no statistically great difference in the diagnostic accuracy of IUA by transvaginal 3D ultrasound and hysteroscopy (P> 0.05), and the diagnosis results of moderate and severe adhesions were consistent (both 20 cases (37.04%) and 6 cases (11.11%), respectively) with no statistical difference (P> 0.05). (e diagnostic accuracy of 3D ultrasound was 96.30%, while that of 2D ultrasound was 90.74%, showing a statistical difference (P< 0.05). In conclusion, ELMDA had a good effect of denoising, and there was a high accuracy of the application of 3D transvaginal ultrasound to diagnose IUA, which had reliable clinical application value.


Introduction
IUA is also known as Asherman's syndrome [1]. IUA is one of the common gynecological diseases with a high clinical incidence. is is because multiple factors result in damage of the endometrial basement layer to cause the adhesions of cervical canal or uterine muscle for partial or total occlusion of the uterine cavity, thereby emerging with menstrual abnormalities, infertility, and repeated abortion in patients [2,3]. In recent years, the incidence of IUA has kept rising with the increase of abortion rate, which has become the second major cause of secondary infertility after fallopian tube factors, posing a serious threat to women's health [4,5]. Hysteroscopy is currently an effective operation for the treatment of IUA and the gold standard for diagnosis [6]. e commonly used hysteroscopy and hysterography are both invasive examinations, and transvaginal 2D ultrasound examination also has certain limitations so that the diagnostic accuracy of IUA is low [7,8].
With the continuous development of ultrasound technology and the improvement of the instrument resolution, 3D transvaginal sonograph (3D-TVS) has made up for the defects of 2D ultrasound and has become a new method for clinical diagnosis of IUA [9]. Ahmadi et al. [10] indicated that 3D contrast-enhanced ultrasonography for uterine cavity was a minimally invasive and cost-effective means to explore the adhesions. However, 3D ultrasonic images are directly reconstructed based on the 2D ultrasonic images obtained by scanning, which contain more speckle noises, so the image quality degrades, and the reconstruction accuracy is not high. Extreme learning machine (ELM) is a new and easy-to-use single-hidden-layer feed-forward neural network learning algorithm, which can randomly assign the input weight and offset of hidden layer during algorithm execution to generate the unique optimal solution [11,12]. ELM has fast learning speed and good generalization performance.
erefore, ELMDA can be applied to remove speckle noises in the image to ensure the effectiveness and accuracy of 3D ultrasonic image reconstruction.
In summary, IUA, as a common cause of infertility, amenorrhea, and repeated abortion, seriously endangers women's reproductive health. e transvaginal 3D ultrasonic examination can clarify the actual situation of IUA to improve the clinical diagnosis and treatment. But the reconstructed 3D ultrasonic image contains speckle noises, and ELMDA can remove speckle noises, making the 3D ultrasonic image more accurate. erefore, the transvaginal 3D ultrasound based on ELMDA was employed to promote the diagnostic accuracy of IUA, and the clinical application value of this transvaginal 3D ultrasound was discussed in the diagnosis of IUA.

Research
Objects. 54 patients with IUA who admitted to the hospital from February 2019 to January 2020 were selected as the research objects. eir age ranged from 20 to 45 years with an average age of 36.31 ± 2.25 years. e experiment had been approved by the Medical Ethics Committee of the hospital. e patients and their family members had been informed of the experiment and signed the informed consent.
e criteria for inclusion were defined to include patients who were above 20 years of age, were diagnosed as IUA, had complete preoperative ultrasound examination data, should receive the hysteroscopic treatment, were in good health, and could normally cooperate with the experiment to receive treatment.
e criteria for exclusion were defined to include patients who had uncertain preoperative diagnosis, suffered from uterine polyps and submucosal fibroids, had incomplete clinical data, and dropped out of the experiment due to their own reasons.

Extreme Learning Machine Denoising Algorithm.
e key of ELMDA was to calculate the output weight and determine the learning mathematical model based on the given activation function type and the number of hidden nodes.
Assuming that there were M arbitrary samples (a i , b i ) (a i � [a i1 , a i2 , · · · a im ] B ∈ R m and b i � [b i1 , b i2 , · · · b im ] B ∈ R n ), the single-hidden-layer feed-forward neural network contained T nodes of hidden layer that could be expressed as in the following equation: (1) In equation (1), f(x) represented the activation function, which could generally be chosen as sine function, hyperbolic function, or radial basis function. z i � [z i1 , z i2 , . . . , z im ] B stood for the weight vector that connected the input data with i th nodes of hidden layer, β i � [β i1 , β i2 , . . . , β in ] B expressed the weight vector connected the output data and the i th node of hidden layer, y i stood for the offset of the i th hidden layer node, q j referred to the feed-forward neural network (FNN) of single hidden layer, and z i x i represented the inner product of vector. e objective of the singlehidden-layer feed-forward neural network was to obtain the minimum output error, as shown in the following equation: erefore, β i , z i , and x i existed and could be calculated in the following equation: e above equation could be expressed by a matrix as follows: In equation (4), H, β, and B represented the output matrix of the hidden layer, the output weight of the hidden layer, and the expected output matrix, respectively. In addition, they could be expressed as in the three following equations: If the input weight zi and offset yi of the hidden layer were set in ELMDA, the output matrix H of the hidden layer could be uniquely determined. e training process of the single-hidden-layer feed-forward neural network could be transformed into solving a linear system Hβ � B, and then the output weight β could be determined.
In equation (8), H † expressed the generalized inverse of matrix H, and it could be proved that the norm obtained was the smallest and unique.
ELMDA was used for image denoising, which should be divided into two steps (training stage and testing stage). In this study, sigmoid function was selected as the activation function of ELMDA. e number of nodes in the hidden layer was set to 120, so PSNR of the image was the largest and the denoising effect was the best. e image denoising process of ELMDA is shown in Figure 1.

Evaluation Indexes of Speckle Noise Removal Effect of
ree Algorithms. ADA and WT-MFA were common ultrasound image denoising methods. In order to detect the denoising effect of different algorithms, ADA [13], WT-MFA [14], and ELMDA were employed to detect the image containing speckle noises, and the parameters of the three algorithms were set at the values that achieved the best effect. e denoising results were quantitatively analyzed by comparing the PSNR and MSE of the denoised images.

Examination Method of Intrauterine Adhesions.
e method of transvaginal 2D ultrasound examination was as follows. First, the patients were told to empty the bladder before examination, and the cystolithotomy position was taken on the examination bed. Second, the iodophor cotton balls were applied to disinfect the vaginal scanning probe, the condom was placed on the probe, and the coupling agent was applied.
ird, the operator put the probe into the vaginal area of the patient slowly with disposable gloves. Fourth, there was a comprehensive scanning of the uterus by tilting and rotating to observe its size, shape, and uterine cavity. Fifth, each uterine meridian and endometrial thickness were recorded, and the endometrial morphology, echo, and presence of endometrial blood flow were observed and recorded in detail. e transvaginal 3D ultrasound examination method contained the following steps. e 3D ultrasound examination was adopted after the 2D ultrasound examination. e volume and size of the sampling frame should be adjusted based on the specific needs. e patient should be told that she held her breath, and then 3D data acquisition system was started to collect data of uterus and endometrium. e three axes were adjusted to choose the best observation direction so as to get a clear uterine coronal plane 3D image. en, the changes of uterine cavity morphology, endometrial thickness, and echo were observed and recorded.
Hysteroscopy examination contained the following steps: All patients underwent hysteroscopy. e cystolithotomy position was taken on the examination bed, the cervical tube was disinfected, and cotton swabs were adopted to dip 2% lidocaine for anesthesia of the cervical canal. 5% glucose was added as a distending medium and was put gently in the hysteroscopy so as to observe the condition of cervical tube. e condition, scope, and degree of IUA were recorded, the adhesion sites were separated for the patients with IUA, and antibiotics were given to prevent infection after the surgery.

Classification Standard of Intrauterine Adhesions.
According to relevant studies, ultrasonic diagnosis results of IUA were mild, moderate, and severe, respectively. Mild adhesions meant that the endometrial thickness was greater than 5 mm, its range was less than 1/4 of the length of uterine cavity, and the opening of fallopian tube was normal or the lesion was mild. e moderate adhesions of 1/4 to 3/4 were involved, the opening of fallopian tube was abnormal, and some of the uterine cavity was blocked. e severe adhesions were that endometrial thickness was less than 2 mm (extremely narrow), uterine walls were adhered or the adhesion zone was hypertrophic, and the range was larger than 3/4 of the length diameter of uterine cavity. Besides, some patients had the uterine cavity completely locked.
2.6. Statistical Methods. SPSS 19.0 statistical software was used for data processing in this study, and the measurement data could be expressed as mean ± standard deviation (‾x ± s). In addition, count data was represented by percentage (%). e analysis of variance was used for pairwise comparison. P < 0.05 meant that the results were statistically significant.

e Effects of ree Algorithms to Remove Speckle Noise.
For quantitative analysis of denoising results, MSE and PSNR of denoised images were considered as quantitative indexes. Speckle noises with variances of 0.1, 0.5, and 1.0 were added to the original image in sequence, and the test results were recorded in time. e results showed that the denoising MSE of ADA were 0.0096, 0.0121, and 0.0139 when the noise variances were 0.1, 0.5, and 1.0, respectively. e denoising MSE of different noise variances in WT-MFA were 0.0032, 0.0087, and 0.0135, respectively. e denoising MSE of ELMDA used in this study were 0.0022, 0.0039, and 0.0071 when the noise variances were 0.1, 0.5, and 1.0, respectively. As shown in Figure 2, the MSE of ELMDA was greatly smaller than those of ADA and WT-MFA when the noise variance was constant, showing a statistical difference (P < 0.05). When the noise variance increased, the MSE of the three algorithms all grew but the MSE of ELMDA was still the minimum.
When the noise variances were 0.1, 0.5, and 1.0, the PSNR of ADA was 67.53, 66.49, and 65.98 in sequence; the PSNR of WT-MFA was 71.26, 67.54, and 65.87, respectively; and the PSNR of ELMDA was 74.08, 71.12, and 69.57, respectively. Figure 3 indicates that the PSNR of ELMDA was higher than those of ADA and WT-MFA when the noise variance was constant, and the difference was statistically obvious (P < 0.05). With the increase of noise variances, the PSNR of the three algorithms decreased, while the PSNR of ELMDA was still the largest.

Hysteroscopy Results of Patients with Intrauterine
Adhesions. In this study, 54 patients with IUA were examined by hysteroscopy and all of them were diagnosed with IUA: 28 patients with mild adhesions (51.85%), 20 patients with moderate adhesions (37.04%), and 6 patients with severe adhesions (11.11%) (Figure 4).

Diagnostic Results of Transvaginal 2-Dimensional Ultrasound Examination.
ere were 31 patients with mild adhesions (57.41%) detected by transvaginal 2D ultrasound examination. 12 patients suffered from moderate adhesions, accounting for 22.22%. Besides, 6 patients were diagnosed with severe adhesions, accounting for 11.11%. ere were 5 patients with no adhesions (9.26%). As shown in Figure 5, the diagnostic rate of 2D ultrasound examination for mild adhesions was higher than that of hysteroscopy by comparison with the diagnosis results of 2D ultrasound examination and hysteroscopy, with statistically huge difference (P < 0.05). e diagnostic rate of 2D ultrasound examination   for moderate adhesions was lower than that of hysteroscopy, and the difference was statistically remarkable (P < 0.05). However, there was no statistical difference in the diagnosis of severe adhesions (P > 0.05). ultrasound examination for moderate and severe adhesions were consistent with the results of hysteroscopy, and there was no obvious difference between two diagnoses of mild adhesions (P > 0.05), so the diagnostic rates of transvaginal 3D ultrasound examination and hysteroscopy were not statistically different (P > 0.05). Figure 7 shows the 2D and 3D ultrasound images of one female patient with mild IUA (aged 31 years).

Comparison of Diagnosis Results between Transvaginal 3-Dimensional and 2-Dimensional Ultrasound Examinations.
ere was a comparison of diagnostic results of transvaginal 3D and 2D ultrasound examinations. Based on the above results, the diagnostic rate of 2D ultrasound examination was 90.74%, while that of 3D ultrasound examination was 96.30%. As shown in Figure 8, the overall diagnostic rate of 3D ultrasound examination in diagnosing IUA was markedly higher than that of 2D ultrasound examination, with statistical substantial difference (P < 0.05).
e results of transvaginal 3D ultrasound examination in the diagnosis of mild and moderate adhesions were better than those of 2D ultrasound examination (P < 0.05), which were closer to the diagnosis results of hysteroscopy. e results of 2D and 3D ultrasonic diagnosis of severe adhesions were consistent with no statistical difference (P > 0.05), which were also in accordance with those of hysteroscopy.

Discussion
As a common gynecological disease, IUA seriously endangers women's health and is also a crucial cause of female infertility. Some patients also suffer from repeated abortion, ectopic pregnancy, and stillbirth [15,16]. At present, the main methods to diagnose IUA are transvaginal 2D ultrasound examination, X-ray hysterosalpingography, and hysteroscopy. Among them, hysteroscopy and X-ray hysterosalpingography are both invasive examinations. e traditional transvaginal 2D ultrasound examination is difficult to determine the accurate location of endometrial lesions, and the diagnosis accuracy of IUA is low. In recent years, the emergence of transvaginal 3D ultrasound examination makes up for the deficiency of 2D ultrasound examination and becomes a new method for the diagnosis of IUA [17,18]. e transvaginal 3D ultrasound examination is reconstructed on the basis of 2D ultrasound examination and contains speckle noises, which makes the image precision not enough.
Aiming at the speckle noises in ultrasonic images, ELMDA was adopted in this study and there was a comparison of denoising performance of three different algorithms. Besides, ELMDA was applied to 3D ultrasound examination to diagnose IUA and its results were compared with those of hysteroscopy and transvaginal 2D ultrasound examination. e final results revealed that MSE of ELMDA was dramatically smaller than those of ADA and WT-MFA, and the difference was statistically obvious (P < 0.05). * *

Scientific Programming
Moreover, ELMDA had a higher PSNR than those of ADA and WT-MFA when the noise variance was constant, showing a statistically marked difference (P < 0.05). ELMDA had a better denoising effect, which was different from the research results of Usha et al. [19] due to the fact that the sample size was too small and the included objects were different. e results of hysteroscopy were regarded as the criteria for the diagnosis of IUA, and the results of transvaginal 2D and 3D ultrasound examinations were compared. e results indicated that the diagnostic rate of transvaginal 2D ultrasound examination for mild and moderate adhesions was statistically different compared to that of hysteroscopy (P < 0.05). However, the diagnosis results of severe adhesions were consistent, with 6 (11.11%) cases diagnosed (P > 0.05). What is more, there was no statistically great difference in the diagnosis rate of IUA between transvaginal 3D ultrasound examination and hysteroscopy (P > 0.05), and the diagnosis results of moderate and severe adhesions were the same (both 20 cases (37.04%) and 6 cases (11.11%)), with no statistically significant difference (P > 0.05). e diagnostic accuracy of 3D ultrasound examination was 96.30%, while that of 2D ultrasound examination was 90.74%. us, the diagnostic accuracy of transvaginal 3D ultrasound examination for IUA was remarkably higher than that of 2D ultrasound examination, with statistical difference (P < 0.05). is was similar to the research results of Kim et al. [20], both of which suggested that 3D ultrasound examination could be employed to diagnose IUA with high diagnostic accuracy.

Conclusion
ELMDA was applied to denoise the speckle noises in images based on 3D ultrasound examination and was compared with ADA and WT-MFA. Moreover, ELMDA was used for 3D ultrasound image diagnosis so as to explore the diagnostic accuracy of 3D ultrasound examination based on ELMDA. e results showed that ELMDA had the lowest MSE and the highest PSNR with better denoising effect. In addition, transvaginal 3D ultrasound examination after denoising had a high accuracy in diagnosing IUA, which was superior to transvaginal 2D ultrasound examination, and had certain clinical application value. However, the small sample size of the patients selected in this study might cause a reliable deviation in the results. In the future, the selection range of patients should be increased to further explore the effective rate of transvaginal 3D ultrasound examination in the diagnosis of IUA. To sum up, the results of this study provided a theoretical basis for the investigation of ELMDA and the application of transvaginal 3D ultrasound examination in the diagnosis of IUA.

Data Availability
e data used to support the findings of this study are available from the corresponding author upon request.

Conflicts of Interest
e authors declare no conflicts of interest. ultrasound examinations. * expresses that the difference in diagnostic rate was statistically considerable compared with 2D ultrasound examination (P < 0.05). 6 Scientific Programming