Applying Deep learning in Recognizing the Properties of Vitreous Opacity on Ophthalmic Ultrasound Images

doi:10.21203/rs.3.rs-2446017/v1

Download PDF

Article

Applying Deep learning in Recognizing the Properties of Vitreous Opacity on Ophthalmic Ultrasound Images

https://doi.org/10.21203/rs.3.rs-2446017/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 18 Aug, 2023

Read the published version in Eye →

You are reading this latest preprint version

BACKGROUND: To explore the feasibility of artificial intelligence technology based on deep learning to automatically recognize the properties of vitreous opacities in ophthalmic ultrasound images.

METHODS: The normal and three typical vitreous opacities confirmed as physiological vitreous opacity (VO), asteroid hyalosis (AH) and vitreous hemorrhage (VH)，were selected and marked from 2000 gray scale Color Doppler ultrasound images for each lesion. Five residual networks (ResNet) and two GoogLeNet models were trained to recognize the vitreous lesions. 75% images were randomly selected as the training set, the remaining 25% as a test set. The accuracy and parameters were recorded and compared among these seven different deep learning (DL) models. The precision, recall, FI score and the area under the receiver operating characteristic curves (AUC) values of recognizing the vitreous lesions were calculated with the most accurate DL model.

RESULTS: There were significant statistical differences in the accuracy and parameters among these seven DL models. GoogleNet inception V1 achieved the highest accuracy (95.5%) and the least parameters (10315580) in recognizing the vitreous lesions. GoogleNet inception V1 achieved 0.94, 0.94, 0.96, and 0.96 precision；0.94, 0.93, 0.97and 0.98 recall ；0.94, 0.93, 0.96 and 0.97 F1Score in recognizing normal, VO, AH, and VH. The AUC values of these four vitreous lesions were 0.99, 1.0, 0.99 and 0.99, respectively.

CONCLUSIONS: GoogLeNet inception V1 has shown promising results in recognizing the ophthalmic ultrasound image. With more and more ultrasound image data, a wide variety of hidden information in the eye diseases can be clearly detected automatically by the artificial intelligence technology based on deep learning.

Health sciences/Health care/Diagnosis/Physical examination

Health sciences/Diseases/Eye diseases/Vitreous detachment

Health sciences/Signs and symptoms/Eye manifestations

The vitreous body is a transparent, extracellular gel, which can reduce the transparency in some physiological and pathological states [1]. While, on the ultrasound (US) image, the change of vitreous acoustic transmissibility can be manifested as the echo with different strength and shape, which is defined as generalized vitreous opacity in ultrasonography [2]. In clinical, physiological vitreous opacity (VO) or degenerative opacity, such as asteroid hyalosis (AH) hardly affect vision, so it usually does not require treatment. Pathological vitreous opacity, such as vitreous hemorrhage (VH) can cause a significant loss of vision, which may further lead to traction retinal detachment (TRD) and even blindness without timely medical intervention [3]. Ultrasound (US) is most commonly used imaging modality to determine the properties of vitreous opacity in ophthalmic clinical practice [4]. However, US detecting is an operator-dependent imaging modality, image quality and diagnoses are different with unequal skills [5]. Therefore, it is essential to develop an intelligent method that assesses US image quality and provide feedback to operators accurately to assist in US diagnosis and/or to make such assessment more objective and less operator dependent.

Artificial intelligence (AI) was originally described as a technological ability to think independently and replicate human behavior after a series of training. Since it was introduced from 1959, AI has developed rapidly and brought great benefits to research, industry, medicine and other fields [6, 7]. Especially, AI holds great promise in computational medicine. Several AI algorithms have been developed for medical purposes, such as screening, risk stratification and assisted decision-making [8]. Deep learning (DL), as a subset of AI, uses a programmable neural network that can extract and transform the features and enable the machine to make its own decisions automatically without human aid [9, 10]. Since the first paper using deep learning to screen for diabetic retinopathy (DR) was published in 2016 [11], AI research, particularly with DL, is proliferating at a rapid pace within the field of ophthalmology.

In ophthalmology, significant progress of AI and DL systems has been demonstrated in image-centric specialties. From digital fundus photographs, many studies have shown DL systems are accurate and effective in DR, age-related macular degeneration (AMD), glaucoma, refractive error, retinopathy of prematurity (ROP), as well as in identifying cardiovascular risk factors such as blood pressure [12–17]. Some new studies have also focused on the applications of AI and DL systems combined with optical coherence tomography (OCT) in identifying disease features, progression and treatment response for several retinal conditions such as choroidal neovascular (CNV), diabetic macular edema (DME) [18, 19]. In the research of De Fauw et al. [20], a DL framework was trained to not only detect a single disease, but more than 50 common retinal diseases using three-dimensional optical coherence tomography images.

Deep learning technology not only shows great potential in intelligent diagnosis of fundus images, but also is being widely used in the field of medical ultrasound. It has been recently applied in classifying, detecting, and segmenting the US image of thyroid, breast and liver diseases [21–23]. However, there are still few research on automatically classifying and recognizing the ultrasonic images of ophthalmic diseases using DL. To the best of our knowledge, this is the first paper that uses DL technology to automatically classify and recognize the properties of vitreous opacity.

US Images Selection / Datasets

Gray scale freeze-frame US images in the same gain, scanning depth and width obtained by the same skilled sonographers were exported in DICOM format with the resolution of 800x555 pixels from Color Doppler ultrasound diagnosis device (MyLabClass C, Esaote S. p. A. Genova, Italy) during January 2014 to July 2021 at Ophthalmology Department, the Fourth Affiliated Hospital of China Medical University. The normal and three typical US images containing vitreous opacity were selected: physiological vitreous opacity (VO), asteroid hyalosis (AH) and vitreous hematogenesis (VH) (Fig. 1). Three experienced ophthalmic ultrasound doctors discussed and agreed on the diagnosis and annotated the images, eliminated the low-quality images caused by unclear diagnosis, blurred image and artifacts, and no data augmentation techniques such as flipping, changing brightness, or changing scale were implemented. Finally, in each group, 2000 qualified ultrasound images were retained. Then, the vitreous round area image parts were cropped and resized for DL model training and testing (Fig. 2).

Deep Learning Model

In each group, 1500 processed US images were randomly selected as the training set and the rest (500 images) as the test set. Five kinds of ResNet: ResNet 50, ResNet 101, ResNet 152, Improved ResNet 50, Improved ResNet 101, and two kinds of GoogLeNet: Inception V1 and Inception V2 were trained to automatically classify the vitreous opacities. In the classification model, the algorithm was trained to classify the images into four categories: normal, VO, AH and VH. The classification accuracy and parameter quantity of these seven models were recorded and compared. The model with the highest accuracy and the least parameters was used to recognize the vitreous lesions images of test set, calculated the precision, recall, and F1Score (Fig. 3).

Statistical Analysis

SPSS software (Version 23.0, IBM, Chicago, IL) was used to analyze the results. A Chi square test was performed to analyze the difference in classification accuracy among the seven models. p < 0.05 was considered statistically significant. MedCalc software (version 19.8 for Windows, MedCalc Software, Mariakerke, Belgium) was used for plotting the receiver operating characteristic (ROC) curves, as well as the area under the curve (AUC) values, to evaluate the performance of DL model four kinds of vitreous US image recognition.

The classification accuracy and parameter quantity of the seven DL models were as follows: ResNet 50: 91.8% and 23516228, ResNet 101: 92.2% and 42508356, ResNet 152: 92.3% and 58152004, Improved ResNet50: 94% and 25504628, Improved ResNet101: 93% and 4582136, GoogLeNet Inception V1: 95.5% and 10315580, GoogLeNet Inception V2: 95.5% and 10601260. There were significant statistical differences in the accuracy and parameter quantity among these seven DL models (Fig. 4). GoogleNet inception V1 achieved 0.96 accuracy; 0.94, 0.94, 0.96, and 0.96 precision;0.94, 0.93, 0.97and 0.98 recall; 0.94,0.93,0.96 and 0.97 F1Score in recognizing normal, VO, AH, and VH. The area under the receiver operating characteristic (ROC) (Fig. 5) curve (AUC) values were 0.99, 1.00, 0.99 and 0.99 (Table 1), respectively.

The Confusion Matrix Diagram was used to calculate the accuracy, precision, recall and F1Score. In the confusion matrix diagram, the model true label is represented horizontally and the prediction label is represented vertically. According to the true and predicted labels, the positive metrics of true was labeled as P (Positive) and the negative metrics of true was labeled as N (False Negative). True Positive (TP) was the metrics that was predicted by the model to be positive and whose true labels were also positive. True Negative (TN) was the metrics that was predicted by the model to be negative and whose true labels were also negative. False Positive (FP) was the positive metrics predicted by the model, but the true labels were negative metrics. False Negative (FN) was the negative metrics predicted by the model, while the true labels were the number of positive samples. Then the confusion matrix diagram can be drawn from the above values, as shown in Fig. 6.

Accuracy is the most commonly used metric to describe the accuracy of the model; the proportion of samples correctly identified by the model to the total number of samples, generally the higher the better. In the confusion matrix diagram of this study, the diagonal part of the blue background was correctly predicted by GoogleNet inception V1. The formula is as follows:

$$\text{A}\text{c}\text{c}\text{u}\text{r}\text{a}\text{c}\text{y}=\frac{\text{T}\text{P}+\text{T}\text{N}}{\text{P}+\text{N}}$$

Precision refers to the proportion of samples with a positive and correct prediction to those with a positive prediction. In this study, the precision refers to the accuracy of the model in recognizing each type of vitreous lesions, the formula is as follows:

$$\text{P}\text{r}\text{i}\text{c}\text{i}\text{s}\text{i}\text{o}\text{n}=\frac{\text{T}\text{P}+\text{T}\text{N}}{\text{T}\text{P}+\text{F}\text{P}}$$

Recall represents the number of all samples with a positive true label that are correctly predicted. In this study, the recall refers to the proportion of US images that can be accurately predicted in each type of vitreous lesions by the model, the formula is as follows:

$$\text{R}\text{e}\text{c}\text{a}\text{l}\text{l}=\frac{\text{T}\text{P}}{\text{T}\text{P}+\text{F}\text{N}}$$

Precision and recall reflect two aspects of DL model performance, with a high precision representing a model’s ability to find the right sample and a high recall representing a model’s ability to find the full sample. However, only a single index cannot comprehensively evaluate the model’s functionality. In order to balance the influence of these two metrics and evaluate a DL model comprehensively, F1-Score is introduced as a comprehensive index, the formula is

$$\text{F}1\text{S}\text{c}\text{o}\text{r}\text{e}=2\times \frac{\text{P}\text{r}\text{e}\text{c}\text{i}\text{s}\text{i}\text{o}\text{n}\times \text{R}\text{e}\text{c}\text{a}\text{l}\text{l}}{\text{P}\text{r}\text{e}\text{c}\text{s}\text{i}\text{o}\text{n}+\text{R}\text{e}\text{c}\text{a}\text{l}\text{l}}$$

The vitreous body presents a high acoustic permeability in terms of acoustic property, and appears as anechoic status on US image of normal human eyes [24]. Once there are some lesions, such as VO, VH and retinal detachment (RD) in the vitreous cavity, the echoes with different intensity and shape will be shown on US images. However, limited by the principle of ultrasonic imaging, the ophthalmic US images can also produce speckles and noises like those of other tissues and organs in human body, leading to a linear decline in ultrasonic image quality, which is a key factor for lesion analysis [25, 26]. At the same time, training a qualified and skilled ophthalmic sonographer also requires a certain amount of time and efforts. In addition, due to the subjectivity of ultrasound operation and image analysis, the results vary from person to person [27]. Therefore, objective and accurate acquisition and interpretation of ophthalmic ultrasonic images is increasingly becoming the demand of ophthalmic clinicians and ophthalmic imaging.

In recent years, the application of AI technology based on DL in ophthalmic diseases has been favored by more and more ophthalmic clinicians, and has become a hotspot of ophthalmic imaging research around the world. Many research findings revealed that DL assisted ophthalmic imaging examination improves the objectivity and accuracy of eye disease diagnosis [28, 29]. Therefore, the present study introduced two DL convolutional neural networks (CNN), residual network (Resnet) and GoogLeNet, to train the computer to automatically identify the US images of vitreous opacities with different properties and compare their differences. The results showed that GoogLeNet inceptionV1 had the highest classification and recognition accuracy while using the fewest parameters among the seven DL models. GoogLeNet is a kind of CNN proposed by Szegedy et al. [30], which is used for machine learning classification tasks through Inception network structure. Its core idea is to increase the depth and width of the network without increasing the computational cost [31, 32]. ResNet proposed by He et al. after GoogLeNet also achieved very good results in image recognition and detection tasks, speeding up CNN training speed, but its disadvantage is that the training precision is not high [33]. In addition, if too many parameters are used in network training, it is prone to overfitting [34] and affecting the efficiency of the model. GoogLeNet inceptionV1 is the most suitable DL model in this study to assist the automatic classification and recognition of vitreous opacity US images, and its classification accuracy is as high as 96%. Especially for VH, its precision, recall and F1score can reach 97%, which is more conducive to the formulation and implementation of clinical diagnosis and treatment strategies.

More and more ophthalmology clinicians and scholars are concerning DL assisted automatic recognition of ophthalmic US images. Di Chen et al [35] used the deep convolutional neural network 1 (DNN1), DNN2 and other neural networks to automatically classify and segment ophthalmic B-type ultrasound images of normal, RD, VD, VH and other lesions with an accuracy rate of over 90%. Some research use DL models such as Densenet and CNN to assist ultrasound biomicroscopy (UBM) to automatically determinate phakic status in pediatric and adult [36], classify the anterior angle chamber [37], automatically localize the scleral process [38], and intelligently measure the anterior segment parameters [39], which provide a more accurate and objective imaging verification and measurement means for the evaluation of anterior segment diseases such as lens disease and glaucoma. Our results are consistent with the above research results, which further confirm the application value of DL technology in ophthalmic gray-scale ultrasound image analysis. However, the US images used in their studies are all fan-shaped scanning images with the small probes, which suffer from inconsistent field width [40]. Different from their studies, the US images in this study are scanned by linear array probe of color Doppler flow imaging (CDFI), which is characterized by the same field width in both the far field and the near field [41]. Compared with B-ultrasound for ophthalmology, the display images are clearer and more conducive to the recognition of lesion features by DL.

Our study also has limitations. Although the DL model has achieved ideal classification and recognition results in this study, our study is limited to the automatic recognition of vitreous opacity. For more complex intraocular diseases, such as RD and choroidal detachment, because the data set is still small, the generalization ability of the model and the accuracy of classification and recognition need to be improved. Moreover, considering that CDFI can fully display the blood flow features of eye diseases, we will further study the CDFI images of intraocular lesions. In conclusion, results in this study indicate that an automated classification and recognition DL model can achieve a high accuracy in identifying the vitreous opacity based on CDFI gray scale images and may be a good tool in future clinical practice. The combination of DL technology and ophthalmic ultrasound images not only helps to reduce the misdiagnosis of eye diseases, but also greatly improves the work efficiency of ophthalmic sonographers. At present, there are still many problems in the application of AI in the field of Ophthalmology, but it is believed that with the progress of computer technology and the in-depth research of researchers, AI technology represented by DL will play a more significant role in the field of Ophthalmology research and clinical diagnosis and treatment.

What was known before:

Ultrasound (US) is most commonly used imaging modality to determine the properties of vitreous opacity in ophthalmic clinical practice.
US detecting is an operator-dependent imaging modality, image quality and diagnoses are different with unequal skills.

What this study adds:

GoogLeNet inception V1 has shown promising results in recognizing the ophthalmic ultrasound image.
With more and more ultrasound image data, a wide variety of hidden information in the eye diseases can be clearly detected automatically by the artificial intelligence technology based on deep learning.

Duke-Elder S. Diseases of the vitreous body. Edited by S D-E Henry Kimpton: London. 1969; pp 315 – 75.
Baum G, Greenwood I. The application of ultrasonic locating techniques to ophthalmology; theoretic considerations and acoustic properties of ocular media. I. Reflective properties. Am J Ophthalmol. 1958; 46:319–29.
Tandias R, Lemire CA, Palvadi K, Arroyo JG. Posterior vitreous detachment status a predictive factor for outcomes of vitrectomy for diabetic vitreous hemorrhage. Retina. 2022; 42:1103–10.
Lizzi FL, Coleman DJ. History of ophthalmic ultrasound. J Ultrasound Med. 2004; 23:1255–66.
Liu SF, Wang Y, Yang X, Lei BY, Liu L, Li SX, et al. Deep learning in medical ultrasound analysis: A review. Engineering. 2019; 5:261–75.
McCarthy J, Minsky ML, Rochester N, Shannon CE. A proposal for the dartmouth summer research project on Artificial intelligence. AI MAG. 2006; 27:12–4.
Samuel AL. Some studies in machine learning using the game of checkers. IBM J Res Develop. 2000, 4: 207–26.
Thrall JH, Li X, Li Q, Cruz C, Do S, Dreyer K, et al. Artificial intelligence and machine learning in radiology: Opportunities, challenges, pitfalls, and criteria for success. J Am Coll Radiol. 2018; 15:504–8.
Shen D, Wu G, Suk HI. Deep learning in medical image analysis. Annu Rev Biomed Eng. 2017; 19:221–48.
Litjens G, Kooi T, Bejnordi BE, Setio AAA, Ciompi F, Ghafoorian M, et al. A survey on deep learning in medical image analysis. Med Image Anal. 2017; 42:60–88.
Gulshan V, Peng L, Coram M, Stumpe MC, Wu D, Narayanaswamy A, et al. Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus Photographs. JAMA. 2016; 316:2402–10.
van der Heijden AA, Abramoff MD, Verbraak F, van Hecke MV, Liem A, Nijpels G. Validation of automated screening for referable diabetic retinopathy with the IDx-DR device in the Hoorn Diabetes Care System. Acta Ophthalmol. 2018; 96:63–8.
Burlina PM, Joshi N, Pekala M, Pacheco KD, Freund DE, Bressler NM. Automated grading of age-related macular degeneration from color fundus images using deep convolutional neural networks. JAMA Ophthalmol. 2017; 135:1170–6.
Li Z, He Y, Keel S, Meng W, Chang RT, He M. Efficacy of a deep learning system for detecting glaucomatous optic neuropathy based on color fundus photographs. Ophthalmology. 2018; 125:1199–206.
Varadarajan AV, Poplin R, Blumer K, Angermueller C, Ledsam J, Chopra R, et al. Deep learning for predicting refractive error from retinal fundus images. Invest Ophthalmol Vis Sci. 2018; 59:2861–8.
Campbell JP, Ataer-Cansizoglu E, Bolon-Canedo V, Bozkurt A, Erdogmus D, Kalpathy-Cramer J, et al. Expert diagnosis of plus disease in retinopathy of prematurity from computer-based image analysis. JAMA Ophthalmol. 2016; 134:651–7.
Poplin R, Varadarajan AV, Blumer K, Liu Y, McConnell MV, Corrado GS, et al. Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning. Nat Biomed Eng. 2018; 2:158–64.
Moraru AD, Costin D, Moraru RL, Branisteanu DC. Artificial intelligence and deep learning in ophthalmology - present and future (Review). Exp Ther Med. 2020; 20:3469–73.
Kermany DS, Goldbaum M, Cai W, Valentim CCS, Liang H, Baxter SL, et al. Identifying medical diagnoses and treatable diseases by image-based deep learning. Cell. 2018; 172:1122–31.
De Fauw J, Ledsam JR, Romera-Paredes B, Nikolov S, Tomasev N, Blackwell S, et al. Clinically applicable deep learning for diagnosis and referral in retinal disease. Nat Med. 2018; 24:1342–50.
Akkus Z, Cai J, Boonrod A, Zeinoddini A, Weston AD, Philbrick KA, et al. A Survey of deep-learning applications in ultrasound: Artificial intelligence-powered ultrasound for improving clinical workflow. J Am Coll Radiol. 2019; 16:1318–28.
Wei W, Haishan X, Alpers J, Rak M, Hansen C. A deep learning approach for 2D ultrasound and 3D CT/MR image registration in liver tumor ablation. Comput Methods Programs Biomed. 2021; 2016:106117.
Qian X, Pei J, Zheng H, Xie X, Yan L, Zhang H, et al. Prospective assessment of breast cancer risk from multimodal multiview ultrasound images via clinically applicable deep learning. Nat Biomed Engineering. 2021; 5:522–32.
Ossoinig K. Clinical echo- ophthalmology, in Blodi FC (ed): Current Concepts of Ophthalmology. St Louis, CV Mosby Co. 1972; vol 3: 101 – 30.
Tanter M, Fink M. Ultrafast imaging in biomedical ultrasound. IEEE Trans Ultrason Ferroelectr Freq Control. 2014; 61:102–19.
Chen H, Zheng YF, Park J.H, Heng P.A, Zhou S.K. Iterative multi-domain regularized deep learning for anatomical structure detection and segmentation from ultrasound images. Medical Image Computing and Computer-Assisted Intervention-MICCAI. 2016; 19th International Conference:487 – 95.
Reddy UM, Filly RA, Copel JA. Prenatal imaging: Ultrasonography and magnetic resonance imaging. Obstet Gynecol. 2008; 112:145–57.
Rahimy E. Deep learning applications in ophthalmology. Curr Opin Ophthalmol. 2018; 29:254–60
Ting DSW, Peng L, Varadarajan AV, Keane PA, Burlina PM, Chiang MF, et al. Deep learning in ophthalmology: The technical and clinical considerations. Prog Retin Eye Res. 2019; 72:100759.
Szegedy C, Liu W, Jia YQ, Sermanet P, Reed S, Anguelov D, et al. Going deeper with convolutions. Computer Science. 2014; arXiv:1409.4842v1
Laina I, Rupprecht C, Belagiannis V, Tombari F, Navab N. Deeper depth prediction with fully convolutional Residual networks. IEEE International Conference on 3D Vision. 2016; 1–13.
Ioffe S, Szegedy C. Batch normalization: Accelerating deep network training by reducing internal covariate shift. Machine Learning. 2015; arXiv:1502.03167.
He KM, Zhang XY, Ren SQ, Sun J. Deep residual learning for image recognition. Computer Vision and Pattern Recognition. 2015; arXiv:1512.03385.
Ma M, Gao Z, Wu J, Chen YL, Zheng X. A smile detection method based on improved LeNet-5 and support vector machine. SmartWorld/SCALCOM/UIC/ATC/CBDCom/IOP/SCI. 2018; 446 – 51.
Chen D, Yu Y, Zhou Y, Peng B, Wang Y, Hu S, et al. A deep learning model for screening multiple abnormal findings in ophthalmic ultrasonography (With Video). Transl Vis Sci Technol. 2021; 10:22.
Le C, Baroni M, Vinnett A, Levin MR, Martinez C, Jaafar M, et al. Deep learning model for accurate automatic determination of phakic status in pediatric and adult ultrasound biomicroscopy images. Transl Vis Sci Technol. 2020; 9:63.
Shi G, Jiang Z, Deng G, Liu G, Zong Y, Jiang C, et al. Automatic classification of anterior chamber angle using ultrasound biomicroscopy and deep learning. Transl Vis Sci Technol. 2019; 8:25.
Wang W, Wang L, Wang T, Wang X, Zhou S, Yang J, et al. Automatic localization of the scleral spur using deep learning and ultrasound biomicroscopy. Transl Vis Sci Technol. 2021; 10: 28.
Li W, Chen Q, Jiang Z, Deng G, Zong Y, Shi G, et al. Automatic anterior chamber angle measurement for ultrasound biomicroscopy using deep learning. J Glaucoma. 2020; 29:81–5.
Thijssen, JM. The history of ultrasound techniques in ophthalmology. Ultrasound Med Biol. 1993; 19:599–618.
Lieb WE. Color Doppler imaging of the eye and orbit. Radiol Clin North Am. 1998; 36:1059–71.

Table 1

Performance of the GoogLeNet inceptionV1 in recognizing vitreous US images.
	Prisicion	Recall	F1Score	AUC
Normal	0.94	0.94	0.94	0.99
VO	0.94	0.93	0.93	1.00
AH	0.96	0.97	0.96	0.99
VH	0.96	0.98	0.97	0.99

There is no conflict of interest

Download PDF

Journal Publication

published 18 Aug, 2023

Read the published version in Eye →

Editorial decision: revise
24 Apr, 2023
Reviewer #2 agreed at journal
09 Mar, 2023
Review #1 received at journal
04 Feb, 2023
Reviewer #1 agreed at journal
30 Jan, 2023
Reviewers invited by journal
30 Jan, 2023
Editor assigned by journal
26 Jan, 2023
Submission checks completed at journal
06 Jan, 2023
First submitted to journal
05 Jan, 2023

You are reading this latest preprint version

Applying Deep learning in Recognizing the Properties of Vitreous Opacity on Ophthalmic Ultrasound Images

Status:

Journal Publication

Version 1

Abstract

Figures

Introduction

Materials & Methods

US Images Selection / Datasets

Deep Learning Model

Statistical Analysis

Results

Discussion

Summary

References

Tables

Additional Declarations

Status:

Journal Publication

Version 1