Automatic echocardiographic anomalies interpretation using a stacked residual-dense network model

BMC Bioinformatics, Journal Year: 2023, Volume and Issue: 24(1)

Published: Sept. 27, 2023

Echocardiographic interpretation during the prenatal or postnatal period is important for diagnosing cardiac septal abnormalities. However, manual can be time consuming and subject to human error. Automatic segmentation of echocardiogram support cardiologists in making an initial interpretation. such a process does not always provide straightforward information make complete The only identifies region abnormality, whereas should determine based on position defect. In this study, we proposed stacked residual-dense network model segment entire classifying their defect positions generate automatic echocardiographic We generalization with incorporated two modalities: echocardiography. To further evaluate effectiveness our model, its performance was verified by five cardiologists. develop pipeline using 1345 echocardiograms training data 181 unseen from prospective patients acquired standard clinical practice at Muhammad Hoesin General Hospital Indonesia. As result, produced 58.17% intersection over union (IoU), 75.75% dice similarity coefficient (DSC), 76.36% mean average precision (mAP) validation data. Using data, achieved 42.39% IoU, 55.72% DSC, 51.04% mAP. Further, classification had approximately 92.27% accuracy, 94.33% specificity, 92.05% sensitivity. Finally, validated expert varying Kappa value. On average, these results hold promise increasing suitability as supporting diagnostic tool establishing diagnosis.

Language: Английский

---

FetalNet: Low-light fetal echocardiography enhancement and dense convolutional network classifier for improving heart defect prediction

Informatics in Medicine Unlocked, Journal Year: 2022, Volume and Issue: 35, P. 101136 - 101136

Published: Jan. 1, 2022

Fetal heart defect (FHD) examination by ultrasound (US) is challenging because it involves low light, contrast, and brightness. Inadequate US images of fetal echocardiography play an important role in the failure to detect FHDs manually. The automatic interpretation was proposed a previous study. However, quality reduces prediction rate computer-assisted diagnosis results. To increase FHD rate, we propose low-light enhancement stacking with dense convolutional network classifier named "FetalNet." Our FetalNet model developed using 460 produce image model. results showed that all raw could be improved satisfactory performance terms increasing peak signal-to-noise ratio 30.85 dB, structural similarity index 0.96, mean squared error 18.16. Furthermore, reconstructed were used as inputs neural generate best for predicting FHD. increased approximately 25% accuracy, sensitivity, specificity produced 100% predictive negative unseen data. deep learning has potential identify accurately shows practical use identifying congenital diseases future.

Language: Английский

---

SKGC: A General Semantic-level Knowledge Guided Classification Framework for Fetal Congenital Heart Disease

IEEE Journal of Biomedical and Health Informatics, Journal Year: 2024, Volume and Issue: 28(10), P. 6105 - 6116

Published: July 10, 2024

Congenital heart disease (CHD) is the most common congenital disability affecting healthy development and growth, even resulting in pregnancy termination or fetal death. Recently, deep learning techniques have made remarkable progress to assist diagnosing CHD. One very popular method directly classifying ultrasound images, recognized as abnormal normal, which tends focus more on global features neglects semantic knowledge of anatomical structures. The other approach segmentation-based diagnosis, requires a large number pixel-level annotation masks for training. However, detailed segmentation costly unavailable. Based above analysis, we propose SKGC, universal framework identify normal four-chamber (4CH) guided by few masks, while improving accuracy remarkably. SKGC consists semantic-level extraction module (SKEM), multi-knowledge fusion (MFM), classification (CM). SKEM responsible obtaining high-level knowledge, serving an abstract representation structures that obstetricians on. MFM lightweight but efficient fuses with original specific images. CM classifies fused can be replaced any advanced classifier. Moreover, design new loss function enhances constraint between foreground background predictions, quality knowledge. Experimental results collected real-world NA-4CH publicly FEST datasets show achieves impressive performance best 99.68% 95.40%, respectively. Notably, improves from 74.68% 88.14% using only 10 labeled masks.

Language: Английский

---

TVS: a trusted verification scheme for office documents based on blockchain

Complex & Intelligent Systems, Journal Year: 2022, Volume and Issue: 9(3), P. 2865 - 2877

Published: Jan. 5, 2022

Abstract To realize the encryption of document information, authority authentication, and traceability historical records, we propose a trusted verification scheme (TVS) for office documents to ensure security. Specifically, is realized by timestamps, smart contracts (or chaincode), other blockchain technologies. It based on features blockchain, such as security, credibility, immutability, network behavior. And TVS stores users information through blockchain; it can monitor state changes in real time setting trigger conditions contracts. The experiment indicates that have real-time monitoring data records. Moreover, achieved purpose ensuring authenticity objectivity data, avoiding illegal tampering malicious documents.

Language: Английский

---

Comparative Analysis of Machine Learning Methods for Prediction of Heart Diseases

Journal of Machinery Manufacture and Reliability, Journal Year: 2022, Volume and Issue: 51(8), P. 789 - 799

Published: Dec. 1, 2022

Language: Английский

---

CRANet: a comprehensive residual attention network for intracranial aneurysm image classification

BMC Bioinformatics, Journal Year: 2022, Volume and Issue: 23(1)

Published: Aug. 5, 2022

Rupture of intracranial aneurysm is the first cause subarachnoid hemorrhage, second only to cerebral thrombosis and hypertensive mortality rate very high. MRI technology plays an irreplaceable role in early detection diagnosis aneurysms supports evaluating size structure aneurysms. The increase many images, may be a massive workload for doctors, which likely produce wrong diagnosis. Therefore, we proposed simple effective comprehensive residual attention network (CRANet) improve accuracy detection, using extract features aneurysm. Many experiments have shown that CRANet model could detect effectively. In addition, on test set, recall rates reached 97.81% 94%, significantly improved

Language: Английский

---

Tunicate swarm-based grey wolf algorithm for fetal heart chamber segmentation and classification: a heuristic-based optimal feature selection concept

Journal of Intelligent & Fuzzy Systems, Journal Year: 2022, Volume and Issue: 44(1), P. 1029 - 1041

Published: Sept. 30, 2022

Ultrasound image quality management and assessment are an important stage in clinical diagnosis. This operation is often carried out manually, which has several issues, including reliance on the operator’s experience, lengthy labor, considerable intra-observer variance. As a result, automatic evaluation of images particularly desirable medical applications. research work plans to perform fetal heart chamber segmentation classification using novel intelligent technology named as hybrid optimization algorithm Tunicate Swarm-based Grey Wolf Algorithm (TS-GWA). Initially, US data collected undergoes preprocessing total variation technique. From preprocessed images, optimal features extracted TF-IDF approach. Then, Segmentation processed optimally selected Spatially Regularized Discriminative Correlation Filters (SRDCF) method. In final step, done Modified Long Short-Term Memory (MLSTM) Network. The fitness function behind feature selection well hidden neuron MLSTM maximization PSNR minimization MSE. value improved from 3.1 9.8 proposed method accuracy 1.9 12.13 compared other existing techniques. generalization ability adaptability TS-GWA described by conducting various performance analysis. Extensive result shows that techniques performs better than methods.

Language: Английский

---

A deep learning-based intelligent analysis platform for fetal ultrasound four-chamber views

Published: July 22, 2022

The four-chamber view is the primary ultrasound images that clinicians diagnose whether a fetus has congenital heart disease (CHD) in process of prenatal diagnosis and screening, which can provide with clear developmental morphology fetal four chambers (i.e., left atrium, ventricle, right ventricle). early screening for CHD depend on clinicians' experience to large extent. Deep learning technology achieved great success medical image analysis. Hence, applying deep analysis help improve diagnostic accuracy make it more objective. we design learning-based intelligent platform (DLIAP) views, includes an input module, visualization output information query module. DLIAP assist objectively analyzing views further CHD.

Language: Английский

---

Big Data Classification of Ultrasound Doppler Scan Images Using a Decision Tree Classifier Based on Maximally Stable Region Feature Points

International Journal on Recent and Innovation Trends in Computing and Communication, Journal Year: 2022, Volume and Issue: 10(8), P. 76 - 87

Published: Aug. 31, 2022

The classification of ultrasound scan images is important in monitoring the development prenatal and maternal structures. This paper proposes a big data system for Doppler that combines residual maximally stable extreme regions speeded up robust features (SURF) with decision tree classifier. algorithm first preprocesses before detecting extremal (MSER). A few essential are chosen from MSER regions, along region provides best Region Interest (ROI). SURF points represent detected using gradient estimated cumulative interest. To extract feature pixels surround points, Triangular Vertex Transform (TVT) transform used. classifier used to train extracted TVT features. proposed image validated performance parameters such as accuracy, specificity, precision, sensitivity, F1 score. For validation, large dataset 12,400 collected 1792 patients method has an F1score 94.12%, accuracy 93.57%, 97.96%, respectively. evaluation results show classifying better than other algorithms have been past.

Language: Английский

---

Selection of MSER region based Ultrasound Doppler scan Image Big data classification using a faster RCNN network

International Journal of Computer Engineering in Research Trends, Journal Year: 2022, Volume and Issue: 9(10), P. 184 - 192

Published: Oct. 26, 2022

This paper proposes an ultrasound Doppler scan image big data classification approach that uses a selection process to estimate the best regions for extracting feature of faster region-based convolutional neural network (RCNN) network.This scheme initially pre-processes images.From pre-processed image, several maximally stable extremal (MSER) and residual are estimated.The region few selected from used extract features.A correlation-based is select features.The gradient values triangular vertex transform-based features (TVT).The extracted TVT trained using RCNN categorize as femur, brain, abdomen,cervix, thorax, other regions.The evaluation metrics namely precision, recall, F1-score validate algorithm.The proposed provides sensitivity, F1-score, specificity, accuracy 96.13%, 94.74%, 94.26%, 98.82%, 98.27% respectively.

Language: Английский

---