Pub Date : 2022-06-13DOI: 10.4015/s1016237222500375
S. Sabi, J. Jacob, V. Gopi
Age-related Macular Degeneration (AMD) is the prime reason for vision impairment observed in major countries worldwide. Hence an accurate early detection of the disease is vital for more research in this area. Also, having a thorough eye diagnosis to detect AMD is a complex job. This paper introduces a Directed Acyclic Graph (DAG) structure-based Convolutional Neural network (CNN) architecture to better classify Dry or Wet AMD. The DAG architecture can combine features from multiple layers to provide better results. The DAG model also has the capacity to learn multi-level visual properties to increase classification accuracy. Fine tuning of DAG-based CNN model helps in improving the performance of the network. The training and testing of the proposed model are carried out with the Mendeley data set and achieved an accuracy of 99.2% with an AUC value of 0.9999. The proposed model also obtains better results for other parameters such as precision, recall and F1-score. Performance of the proposed network is also compared to that of the related works performed on the same data set. This shows ability of the proposed method to grade AMD images to help early detection of the disease. The model also performs computationally efficient for real-time applications as it does the classification process with few learnable parameters and fewer Floating-Point Operations (FLOPs).
{"title":"CLASSIFICATION OF AGE-RELATED MACULAR DEGENERATION USING DAG-CNN ARCHITECTURE","authors":"S. Sabi, J. Jacob, V. Gopi","doi":"10.4015/s1016237222500375","DOIUrl":"https://doi.org/10.4015/s1016237222500375","url":null,"abstract":"Age-related Macular Degeneration (AMD) is the prime reason for vision impairment observed in major countries worldwide. Hence an accurate early detection of the disease is vital for more research in this area. Also, having a thorough eye diagnosis to detect AMD is a complex job. This paper introduces a Directed Acyclic Graph (DAG) structure-based Convolutional Neural network (CNN) architecture to better classify Dry or Wet AMD. The DAG architecture can combine features from multiple layers to provide better results. The DAG model also has the capacity to learn multi-level visual properties to increase classification accuracy. Fine tuning of DAG-based CNN model helps in improving the performance of the network. The training and testing of the proposed model are carried out with the Mendeley data set and achieved an accuracy of 99.2% with an AUC value of 0.9999. The proposed model also obtains better results for other parameters such as precision, recall and F1-score. Performance of the proposed network is also compared to that of the related works performed on the same data set. This shows ability of the proposed method to grade AMD images to help early detection of the disease. The model also performs computationally efficient for real-time applications as it does the classification process with few learnable parameters and fewer Floating-Point Operations (FLOPs).","PeriodicalId":8862,"journal":{"name":"Biomedical Engineering: Applications, Basis and Communications","volume":"58 1","pages":""},"PeriodicalIF":0.9,"publicationDate":"2022-06-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84129807","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-06-01DOI: 10.4015/s1016237222500272
Sheetal Garg, S. R. Bhagyashree
Magnetic Resonance Imaging (MRI) techniques are a fundamental and imperative part of the medical image processing field. The images acquired from the MRI machines are affected by the noise. This noise degrades the quality of the images. Acquisition of MRI with noise may give erroneous results. Hence, to enhance the image quality, it is necessary to reduce or remove this noise. To enhance the image quality of MRI, a plethora of filtering algorithms are available along with the morphological operations. In this paper, we have implemented numerous filters like Adaptive Median filter, Median filter, Mean filter, bilateral filter, NLM filter, Gaussian filter, Weiner filter, and morphological operations to eliminate the noise in the MRI of the spinal cord. The scenarios considered are 1. Application of filters, 2. Application of filters followed by morphological operations, and 3. Morphological operations followed by the application of filters. Statistical parameters like Peak Signal-to-Noise Ratio (PSNR) and Mean Square Error (MSE) are found for all three approaches and are used to analyze the performance of these techniques. NLM filters are found to give the best performance when compared to other filters. Morphological operations affect the performance of the filters. Application of morphological operations before filtering degrades the filter performance while applying them after improves the performance. The dataset comprises of 250 spinal cord MRIs with noise. The author inferred that the performance of the filters is improved by applying the filtering techniques after the morphological operation.
{"title":"COMPARATIVE ANALYSIS OF VARIOUS FILTERS FOR DENOISING OF THE SPINAL CORD MRIs","authors":"Sheetal Garg, S. R. Bhagyashree","doi":"10.4015/s1016237222500272","DOIUrl":"https://doi.org/10.4015/s1016237222500272","url":null,"abstract":"Magnetic Resonance Imaging (MRI) techniques are a fundamental and imperative part of the medical image processing field. The images acquired from the MRI machines are affected by the noise. This noise degrades the quality of the images. Acquisition of MRI with noise may give erroneous results. Hence, to enhance the image quality, it is necessary to reduce or remove this noise. To enhance the image quality of MRI, a plethora of filtering algorithms are available along with the morphological operations. In this paper, we have implemented numerous filters like Adaptive Median filter, Median filter, Mean filter, bilateral filter, NLM filter, Gaussian filter, Weiner filter, and morphological operations to eliminate the noise in the MRI of the spinal cord. The scenarios considered are 1. Application of filters, 2. Application of filters followed by morphological operations, and 3. Morphological operations followed by the application of filters. Statistical parameters like Peak Signal-to-Noise Ratio (PSNR) and Mean Square Error (MSE) are found for all three approaches and are used to analyze the performance of these techniques. NLM filters are found to give the best performance when compared to other filters. Morphological operations affect the performance of the filters. Application of morphological operations before filtering degrades the filter performance while applying them after improves the performance. The dataset comprises of 250 spinal cord MRIs with noise. The author inferred that the performance of the filters is improved by applying the filtering techniques after the morphological operation.","PeriodicalId":8862,"journal":{"name":"Biomedical Engineering: Applications, Basis and Communications","volume":"19 1","pages":""},"PeriodicalIF":0.9,"publicationDate":"2022-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81625817","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-05-04DOI: 10.4015/s1016237222300012
Suman Dey, Ruby Mishra, M. Mohapatra, S. Sabut
Micro catheters are thin-walled devices gaining pivotal importance in the field of micro invasive surgeries. The need for an efficient design of a micro catheter with the enhancement of its prime characteristics like-kink resistance, lower bending response, increased perumbular capacities, etc., has been the key parameters for research among biomedical engineers. The article highlights the nuances in the technology in the fabrication of micro active catheters and the procedure and necessity of embolization in the process of catheterization. Efficacies of different designs of micro active catheters were studied based on a variety of clinical data trials by several researchers and doctors. Superior materials capable of enhancing the torque efficacy of the device like auxetic materials and their effect on bending angles were studied. Clinical trials were undertaken based on various designs and approaches for the device and the critical characteristics were studied. The micro active catheter with guide-wire shows maximum bending angle and considerable torque making it ideal for micro invasive procedures in constricted as well as divergent blood vessels.
{"title":"MICRO ACTIVE CATHETERS AND EMBOLIZATION TECHNIQUES: A BRIEF REVIEW BASED ON DESIGN AND WORKING EFFICACY","authors":"Suman Dey, Ruby Mishra, M. Mohapatra, S. Sabut","doi":"10.4015/s1016237222300012","DOIUrl":"https://doi.org/10.4015/s1016237222300012","url":null,"abstract":"Micro catheters are thin-walled devices gaining pivotal importance in the field of micro invasive surgeries. The need for an efficient design of a micro catheter with the enhancement of its prime characteristics like-kink resistance, lower bending response, increased perumbular capacities, etc., has been the key parameters for research among biomedical engineers. The article highlights the nuances in the technology in the fabrication of micro active catheters and the procedure and necessity of embolization in the process of catheterization. Efficacies of different designs of micro active catheters were studied based on a variety of clinical data trials by several researchers and doctors. Superior materials capable of enhancing the torque efficacy of the device like auxetic materials and their effect on bending angles were studied. Clinical trials were undertaken based on various designs and approaches for the device and the critical characteristics were studied. The micro active catheter with guide-wire shows maximum bending angle and considerable torque making it ideal for micro invasive procedures in constricted as well as divergent blood vessels.","PeriodicalId":8862,"journal":{"name":"Biomedical Engineering: Applications, Basis and Communications","volume":"13 1","pages":""},"PeriodicalIF":0.9,"publicationDate":"2022-05-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89799678","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-04-28DOI: 10.4015/s1016237222500296
S. Reddy, Shridevi Soma
The major reason for blindness is diabetic macular edema (DME) and hence detection of DME at early stage using optical coherence tomography (OCT) is commonly employed for diagnosing retinal diseases. An accurate disease identification and classification poses a challenging task due to the difficulty in differentiating the abnormal and healthy regions. To overcome these issues and to accurately classify the DME, an effective DME classification approach named antlion spider monkey optimization-based generative adversarial network (ALSMO-based GAN) is proposed in this research for segmenting the retinal layers and to classify the DME more accurately. With the generator and the discriminator components of GAN, the DME is effectively classified so that the devised ALSMO algorithm can be used to train the process of GAN. The inspiration of the foraging and the hunting behavior enable the optimization to increase the rate of convergence and to achieve global optimal solution by reducing the local optima. With the segmented retinal layer, the classification process is progressed through the extraction of relevant features from the retinal layers. The performance of the developed method is verified using measures like accuracy, sensitivity, and specificity which attained values of 92.5%, 98%, and 92.3%, respectively.
{"title":"DIABETIC MACULAR EDEMA CLASSIFICATION WITH OCT IMAGES USING GENERATIVE ADVERSARIAL NETWORK AND ACTIVE CONTOUR MODEL","authors":"S. Reddy, Shridevi Soma","doi":"10.4015/s1016237222500296","DOIUrl":"https://doi.org/10.4015/s1016237222500296","url":null,"abstract":"The major reason for blindness is diabetic macular edema (DME) and hence detection of DME at early stage using optical coherence tomography (OCT) is commonly employed for diagnosing retinal diseases. An accurate disease identification and classification poses a challenging task due to the difficulty in differentiating the abnormal and healthy regions. To overcome these issues and to accurately classify the DME, an effective DME classification approach named antlion spider monkey optimization-based generative adversarial network (ALSMO-based GAN) is proposed in this research for segmenting the retinal layers and to classify the DME more accurately. With the generator and the discriminator components of GAN, the DME is effectively classified so that the devised ALSMO algorithm can be used to train the process of GAN. The inspiration of the foraging and the hunting behavior enable the optimization to increase the rate of convergence and to achieve global optimal solution by reducing the local optima. With the segmented retinal layer, the classification process is progressed through the extraction of relevant features from the retinal layers. The performance of the developed method is verified using measures like accuracy, sensitivity, and specificity which attained values of 92.5%, 98%, and 92.3%, respectively.","PeriodicalId":8862,"journal":{"name":"Biomedical Engineering: Applications, Basis and Communications","volume":"02 1","pages":""},"PeriodicalIF":0.9,"publicationDate":"2022-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"73525398","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-04-27DOI: 10.4015/s1016237222500314
P. Lakshmi Prabha, A. Jayanthy
{"title":"Risk Analysis and Classification of Myocardial Infarction from Carotid Intima Media Thickness of B-Mode Ultrasound Image Using Various Machine Learning and Deep Learning Techniques","authors":"P. Lakshmi Prabha, A. Jayanthy","doi":"10.4015/s1016237222500314","DOIUrl":"https://doi.org/10.4015/s1016237222500314","url":null,"abstract":"","PeriodicalId":8862,"journal":{"name":"Biomedical Engineering: Applications, Basis and Communications","volume":"116 1","pages":""},"PeriodicalIF":0.9,"publicationDate":"2022-04-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90218989","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-04-07DOI: 10.4015/s1016237222500247
S. A. Bhyratae, Neha Mangla
Gene Regulatory Networks (GRNs) reconstruction aims to infer relationships of potential regulation among the genes. With the rapid growth of the biotechnology, such as Ribonucleic acid (RNA)-sequencing and gene chip microarray, the generated high-throughput data provide gene–gene interaction relationships with more opportunities based on gene expression data. Several approaches are introduced to reconstruct the GRNs, but low accuracy is a major drawback. Hence, this paper introduces the hybrid distance measure and the Pearson’s correlation coefficient for reconstructing GRN. The hybrid distance, such as Tversky index, Tanimoto similarity, and Minkowski distance, is employed to connect the edges. The asymmetric partial correlation network is introduced for determining two influence functions for every pair, and edge direction is determined among them. However, the direction of edges is unknown usually and seems difficult to be identified based on gene expression data. Thus, it extends the data processing inequality applying in the directed network for removing the transitive interactions. The influence value of every node is calculated for identifying the significant regulator. The performance of the proposed Hybrid Distance_Entropy based GRN Reconstruction method is analyzed in terms of correlation, reconstruction error, precision, and recall, which provides superior results with values 0.9450, 0.00052, 0.9095, and 0.8913 based on dataset-1.
{"title":"DESIGN AND DEVELOPMENT OF ADVANCED SIMILARITY MEASURE FOR RECONSTRUCTING GRN USING mRNA EXPRESSION PROFILES","authors":"S. A. Bhyratae, Neha Mangla","doi":"10.4015/s1016237222500247","DOIUrl":"https://doi.org/10.4015/s1016237222500247","url":null,"abstract":"Gene Regulatory Networks (GRNs) reconstruction aims to infer relationships of potential regulation among the genes. With the rapid growth of the biotechnology, such as Ribonucleic acid (RNA)-sequencing and gene chip microarray, the generated high-throughput data provide gene–gene interaction relationships with more opportunities based on gene expression data. Several approaches are introduced to reconstruct the GRNs, but low accuracy is a major drawback. Hence, this paper introduces the hybrid distance measure and the Pearson’s correlation coefficient for reconstructing GRN. The hybrid distance, such as Tversky index, Tanimoto similarity, and Minkowski distance, is employed to connect the edges. The asymmetric partial correlation network is introduced for determining two influence functions for every pair, and edge direction is determined among them. However, the direction of edges is unknown usually and seems difficult to be identified based on gene expression data. Thus, it extends the data processing inequality applying in the directed network for removing the transitive interactions. The influence value of every node is calculated for identifying the significant regulator. The performance of the proposed Hybrid Distance_Entropy based GRN Reconstruction method is analyzed in terms of correlation, reconstruction error, precision, and recall, which provides superior results with values 0.9450, 0.00052, 0.9095, and 0.8913 based on dataset-1.","PeriodicalId":8862,"journal":{"name":"Biomedical Engineering: Applications, Basis and Communications","volume":"31 1","pages":""},"PeriodicalIF":0.9,"publicationDate":"2022-04-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"73168684","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-04-07DOI: 10.4015/s1016237222500284
Ankit Soni, K. Rawal
Heart rate variability (HRV) is a fundamental physiological marker for assessing the autonomic nervous system’s (ANS) reaction. The response of the ANS is influenced by a variety of physical activities, i.e. body postural change. This paper aims to analyze the effect of physical activities such as postural change on HRV. To study this, a dataset of 56 subjects electrocardiogram (ECG) was self-recorded in two body postures (i) Supine and (ii) Standing. For the acquisition of ECG, the BIOPAC[Formula: see text]MP36 system has been used at a 500[Formula: see text]Hz sampling frequency. Further, HRV signals of each subject from recorded ECG have been extracted and selected linear, and nonlinear techniques have been used to determine the effect of postural shift on it. Further, the spearman rank correlation coefficient has been evaluated between the calculated parameters to determine the correlation between linear and nonlinear parameters. The obtained results indicate that the HRV is at a higher scale in the supine posture, while it is at a lower scale when the posture has been changed from supine to standing. The change that occurred in the response of HRV with the postural change indicates that the sympathetic activation of ANS is increased in the standing body posture.
{"title":"ANALYSIS OF HRV FOR POSTURAL CHANGE OF YOUNG ADULTS USING SIGNAL PROCESSING METHODS","authors":"Ankit Soni, K. Rawal","doi":"10.4015/s1016237222500284","DOIUrl":"https://doi.org/10.4015/s1016237222500284","url":null,"abstract":"Heart rate variability (HRV) is a fundamental physiological marker for assessing the autonomic nervous system’s (ANS) reaction. The response of the ANS is influenced by a variety of physical activities, i.e. body postural change. This paper aims to analyze the effect of physical activities such as postural change on HRV. To study this, a dataset of 56 subjects electrocardiogram (ECG) was self-recorded in two body postures (i) Supine and (ii) Standing. For the acquisition of ECG, the BIOPAC[Formula: see text]MP36 system has been used at a 500[Formula: see text]Hz sampling frequency. Further, HRV signals of each subject from recorded ECG have been extracted and selected linear, and nonlinear techniques have been used to determine the effect of postural shift on it. Further, the spearman rank correlation coefficient has been evaluated between the calculated parameters to determine the correlation between linear and nonlinear parameters. The obtained results indicate that the HRV is at a higher scale in the supine posture, while it is at a lower scale when the posture has been changed from supine to standing. The change that occurred in the response of HRV with the postural change indicates that the sympathetic activation of ANS is increased in the standing body posture.","PeriodicalId":8862,"journal":{"name":"Biomedical Engineering: Applications, Basis and Communications","volume":"22 8-9 1","pages":""},"PeriodicalIF":0.9,"publicationDate":"2022-04-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"80266755","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-03-30DOI: 10.4015/s1016237222500260
J. Nisha, V. Gopi, P. Palanisamy
Colorectal cancer (CRC) is the common cancer-related cause of death globally. It is now the third leading cause of cancer-related mortality worldwide. As the number of instances of colorectal polyps rises, it is more important than ever to identify and diagnose them early. Object detection models have recently become popular for extracting highly representative features. Colonoscopy is shown to be a useful diagnostic procedure for examining anomalies in the digestive system’s bottom half. This research presents a novel image-enhancing approach followed by a Scaled YOLOv4 Network for the early diagnosis of polyps, lowering the high risk of CRC therapy. The proposed network is trained using the CVC ClinicDB and the CVC ColonDB and the Etis Larib database are used for testing. On the CVC ColonDB database, the performance metrics are precision (95.13%), recall (74.92%), F1-score (83.19%), and F2-score (89.89%). On the ETIS Larib database, the performance metrics are precision (94.30%), recall (77.30%), F1-score (84.90%), and F2-score (80.20%). On both the databases, the suggested methodology outperforms the present one in terms of F1-score, F2-score, and precision compared to the futuristic method. The proposed Yolo object identification model provides an accurate polyp detection strategy in a real-time application.
{"title":"COLORECTAL POLYP DETECTION USING IMAGE ENHANCEMENT AND SCALED YOLOv4 ALGORITHM","authors":"J. Nisha, V. Gopi, P. Palanisamy","doi":"10.4015/s1016237222500260","DOIUrl":"https://doi.org/10.4015/s1016237222500260","url":null,"abstract":"Colorectal cancer (CRC) is the common cancer-related cause of death globally. It is now the third leading cause of cancer-related mortality worldwide. As the number of instances of colorectal polyps rises, it is more important than ever to identify and diagnose them early. Object detection models have recently become popular for extracting highly representative features. Colonoscopy is shown to be a useful diagnostic procedure for examining anomalies in the digestive system’s bottom half. This research presents a novel image-enhancing approach followed by a Scaled YOLOv4 Network for the early diagnosis of polyps, lowering the high risk of CRC therapy. The proposed network is trained using the CVC ClinicDB and the CVC ColonDB and the Etis Larib database are used for testing. On the CVC ColonDB database, the performance metrics are precision (95.13%), recall (74.92%), F1-score (83.19%), and F2-score (89.89%). On the ETIS Larib database, the performance metrics are precision (94.30%), recall (77.30%), F1-score (84.90%), and F2-score (80.20%). On both the databases, the suggested methodology outperforms the present one in terms of F1-score, F2-score, and precision compared to the futuristic method. The proposed Yolo object identification model provides an accurate polyp detection strategy in a real-time application.","PeriodicalId":8862,"journal":{"name":"Biomedical Engineering: Applications, Basis and Communications","volume":"35 1","pages":""},"PeriodicalIF":0.9,"publicationDate":"2022-03-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81458793","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-03-30DOI: 10.4015/s1016237222500259
A. P. Sunija, Adithya K. Krishna, V. Gopi, P. Palanisamy
Diabetic Retinopathy (DR) is the principal cause of vision loss that interrupts the regular interaction of vascular, neural, and retinal constituents leading to impaired neuronal function and retinal abnormalities. Diagnosis of DR from Optical Coherence Tomography (OCT) image is difficult and time-consuming because several small features must be identified and graded, which results in a strenuous diagnosis when integrated with the complexity of the grading system. This study focuses on classifying DR from normal Spectral Domain-OCT (SD-OCT) images using the Directed Acyclic Graph (DAG) network without any pre-processing techniques. The proposed DAG-CNN model comprises 16 convolutional blocks, which learns multi-scale features automatically from multiple layers in the convolutional network and combines them effectively for the DR and normal prediction. The proposed model is tested on the public OCTID_DR and private LFH_DR SD-OCT databases containing DR and healthy OCT images. The model achieved an accuracy, precision, recall, F1-score, and AUC on OCTID_DR database of 0.9841, 0.9727, 0.9818, 0.9772, and 0.9836, respectively; and on LFH_DR database the respective values are 0.9988, 1, 0.9976, 0.9988, and 0.9988 with only 0.1569 Million of learnable parameters. This method significantly reduces the number of learnable parameters and the model’s computational complexity in terms of memory required and FLoating point OPerations (FLOPs). Guided Gradient-weighted Class Activation Mapping (Grad-CAM) is performed to highlight the regions of SD-OCT images that contribute to the decision of the classifier. Our model significantly surpasses the accuracy of the existing models with lower resource consumption and higher real-time performance.
糖尿病视网膜病变(DR)是导致视力丧失的主要原因，它中断了血管、神经和视网膜成分的正常相互作用，导致神经元功能受损和视网膜异常。从光学相干断层扫描(OCT)图像中诊断DR是困难和耗时的，因为必须识别和分级几个小特征，这导致在与分级系统的复杂性相结合时的艰苦诊断。本研究的重点是在没有任何预处理技术的情况下，使用有向无环图(DAG)网络从正常光谱域- oct (SD-OCT)图像中分类DR。提出的DAG-CNN模型由16个卷积块组成，该模型自动从卷积网络的多个层中学习多尺度特征，并将它们有效地组合在一起进行DR和normal预测。在包含DR和健康OCT图像的公共OCTID_DR和私有LFH_DR SD-OCT数据库上对该模型进行了测试。模型在OCTID_DR数据库上的准确率、精密度、召回率、F1-score和AUC分别为0.9841、0.9727、0.9818、0.9772和0.9836;在LFH_DR数据库上，分别为0.9988、1、0.9976、0.9988、0.9988，可学习参数只有0.15.69万个。该方法显著减少了可学习参数的数量，降低了模型在内存和浮点运算方面的计算复杂度。使用梯度加权分类激活映射(Grad-CAM)来突出SD-OCT图像中有助于分类器决策的区域。我们的模型以更低的资源消耗和更高的实时性显著超越了现有模型的准确性。
{"title":"MULTI-SCALE DIRECTED ACYCLIC GRAPH-CNN FOR AUTOMATED CLASSIFICATION OF DIABETIC RETINOPATHY FROM OCT IMAGES","authors":"A. P. Sunija, Adithya K. Krishna, V. Gopi, P. Palanisamy","doi":"10.4015/s1016237222500259","DOIUrl":"https://doi.org/10.4015/s1016237222500259","url":null,"abstract":"Diabetic Retinopathy (DR) is the principal cause of vision loss that interrupts the regular interaction of vascular, neural, and retinal constituents leading to impaired neuronal function and retinal abnormalities. Diagnosis of DR from Optical Coherence Tomography (OCT) image is difficult and time-consuming because several small features must be identified and graded, which results in a strenuous diagnosis when integrated with the complexity of the grading system. This study focuses on classifying DR from normal Spectral Domain-OCT (SD-OCT) images using the Directed Acyclic Graph (DAG) network without any pre-processing techniques. The proposed DAG-CNN model comprises 16 convolutional blocks, which learns multi-scale features automatically from multiple layers in the convolutional network and combines them effectively for the DR and normal prediction. The proposed model is tested on the public OCTID_DR and private LFH_DR SD-OCT databases containing DR and healthy OCT images. The model achieved an accuracy, precision, recall, F1-score, and AUC on OCTID_DR database of 0.9841, 0.9727, 0.9818, 0.9772, and 0.9836, respectively; and on LFH_DR database the respective values are 0.9988, 1, 0.9976, 0.9988, and 0.9988 with only 0.1569 Million of learnable parameters. This method significantly reduces the number of learnable parameters and the model’s computational complexity in terms of memory required and FLoating point OPerations (FLOPs). Guided Gradient-weighted Class Activation Mapping (Grad-CAM) is performed to highlight the regions of SD-OCT images that contribute to the decision of the classifier. Our model significantly surpasses the accuracy of the existing models with lower resource consumption and higher real-time performance.","PeriodicalId":8862,"journal":{"name":"Biomedical Engineering: Applications, Basis and Communications","volume":"21 1","pages":""},"PeriodicalIF":0.9,"publicationDate":"2022-03-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"86002640","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：