Correcting motion artifacts in MRI scans using a deep neural network with automatic motion timing detection

Abstract We assessed the accuracy of semi-automated tumor volume maps of plexiform neurofibroma (PN) generated by a deep neural network, compared to manual segmentation using diffusion weighted imaging (DWI) data. NF1 Patients were recruited from a phase II clinical trial for the treatment of PN. Multiple b-value DWI was imaged over the largest PN. All DWI datasets were registered and intensity normalized prior to segmentation with a multi-spectral neural network classifier (MSNN). Manual volumes of PN were performed on 3D-T2 images registered to diffusion images and compared to MSNN volumes with the Sørensen-Dice coefficient. Intravoxel incoherent motion (IVIM) parameters were calculated from resulting volumes. 35 MRI scans were included from 14 subjects. Sørensen-Dice coefficient between the semi-automated and manual segmentation was 0.77 ± 0.016. Perfusion fraction (f) was significantly higher for tumor versus normal tissue (0.47 ± 0.42 vs. 0.30 ± 0.22, p = 0.02), similarly, true diffusion (D) was significantly higher for PN tumor versus normal (0.0018 ± 0.0003 vs. 0.0012 ± 0.0002, p < 0.0001). By contrast, the pseudodiffusion coefficient (D*) was significantly lower for PN tumor versus normal (0.024 ± 0.01 vs. 0.031 ± 0.005, p < 0.0001). Volumes generated by a neural network from multiple diffusion data on PNs demonstrated good correlation with manual volumes. IVIM analysis of multiple b-value diffusion data demonstrates significant differences between PN and normal tissue.

Download Full-text

An externally validated deep learning model for the accurate segmentation of the lumbar paravertebral muscles

10.1101/2021.10.25.21265466 ◽

2021 ◽

Author(s):

Frank Niemeyer ◽

Annika Zanker ◽

René Jonas ◽

Youping Tao ◽

Fabio Galbusera ◽

...

Keyword(s):

Neural Network ◽

Deep Learning ◽

Large Scale ◽

Deep Neural Network ◽

External Validation ◽

Erector Spinae ◽

Published Data ◽

Spinal Disorders ◽

Paravertebral Muscles ◽

Mri Scans

Purpose. Imaging studies about the relevance of muscles in spinal disorders, and sarcopenia in general, require the segmentation of the muscles in the images which is very labour-intensive if performed manually and poses a practical limit to the number of investigated subjects. This study aimed at developing a deep learning-based tool able to fully automatically perform an accurate segmentation of the lumbar muscles in axial MRI scans, and at validating the new tool on an external dataset. Methods. A set of 60 axial MRI images of the lumbar spine was retrospectively collected from a clinical database. Psoas major, quadratus lumborum, erector spinae, and multifidus were manually segmented in all available slices. The dataset was used to train and validate a deep neural network able to segment muscles automatically. Subsequently, the network was externally validated on images purposely acquired from 22 healthy volunteers. Results. The Jaccard index for the individual muscles calculated for the 22 subjects of the external validation set ranged between 0.862 and 0.935, demonstrating a generally excellent performance of the network. Cross-sectional area and fat fraction of the muscles were in agreement with published data. Conclusions. The externally validated deep neural network was able to perform the segmentation of the paravertebral muscles in axial MRI scans in an accurate and fully automated manner, and is therefore a suitable tool to perform large-scale studies in the field of spinal disorders and sarcopenia, overcoming the limitations of non-automated methods.

Download Full-text

Normalization and Comparison of Photoplethysmography Between Normal and Patient Groups Using Deep Neural Networks

10.21203/rs.3.rs-556724/v1 ◽

2021 ◽

Author(s):

Ji Woon Kim ◽

Seong-Wook Choi

Keyword(s):

Neural Network ◽

Healthy Subjects ◽

Deep Neural Network ◽

Deep Neural Networks ◽

Pulse Wave ◽

Group Identification ◽

Motion Artifacts ◽

The Other ◽

Patient Groups ◽

Automatic Algorithm

Abstract Photoplethysmography (PPG) is easy to perform and provides a variety of measurements, including details of heart rate and arrhythmia. However, automated PPG methods have not been developed because of their susceptibility to motion artifacts and differences in waveform characteristics among individuals. With increasing use of telemedicine, there is growing interest in application of deep neural network (DNN) technology for efficient analysis of vast amounts of PPG data. This study proposes an automatic algorithm incorporating DNNs for individual and patient-group identification; this is achieved by selecting normally measured waveforms, deleting error regions, and normalizing the pulse wave to obtain 10 “section values” that can be easily compared to other waveforms. The proposed algorithm was able to distinguish between patients aged 60–75 years with diabetes and hypertension and healthy subjects aged 25–35 years (AUC = 0.998). On the other hand, errors were frequently observed in identification of individuals (AUC = 0.819).

Download Full-text

Removal of Motion Artifacts from ECG signals by Combination of Recurrent Neural Networks and Deep Neural Networks

10.36227/techrxiv.14526828 ◽

2021 ◽

Author(s):

Muhammad Zubair

Keyword(s):

Neural Network ◽

Neural Networks ◽

Recurrent Neural Networks ◽

Deep Neural Network ◽

Deep Neural Networks ◽

Motion Artifacts ◽

Machine Learning Algorithms ◽

Unique Combination ◽

Ecg Signals ◽

Electrocardiogram Ecg

<div><div><div><p>Electrocardiogram (ECG) is the graphical portrayal of heart usefulness. The ECG signals holds its significance in the discovery of heart irregularities. These ECG signals are frequently tainted by antiques from various sources. It is basic to diminish these curios and improve the exactness just as dependability to show signs of improvement results identified with heart usefulness. The most commonly disturbed artifact in ECG signals is Motion Artifacts (MA). In this paper, we have proposed a new concept on how machine learning algorithms can be used for de-noising the ECG signals. Towards the goal, a unique combination of Recurrent Neural Network (RNN) and Deep Neural Network (DNN) is used to efficiently remove MA. The proposed algorithm is validated using ECG records obtained from the MIT-BIH Arrhythmia Database. To eliminate MA using the proposed method, we have used Adam optimization algorithm to train and fit the contaminated ECG data in RNN and DNN models. Performance evaluation results in terms of SNR and RRMSE show that the proposed algorithm outperforms other existing MA removal methods without significantly distorting the morphologies of ECG signals.</p></div></div></div>

Download Full-text

Removal of Motion Artifacts from ECG signals by Combination of Recurrent Neural Networks and Deep Neural Networks

10.36227/techrxiv.14526828.v1 ◽

2021 ◽

Author(s):

Muhammad Zubair

Keyword(s):

Neural Network ◽

Neural Networks ◽

Recurrent Neural Networks ◽

Deep Neural Network ◽

Deep Neural Networks ◽

Motion Artifacts ◽

Machine Learning Algorithms ◽

Unique Combination ◽

Ecg Signals ◽

Electrocardiogram Ecg

<div><div><div><p>Electrocardiogram (ECG) is the graphical portrayal of heart usefulness. The ECG signals holds its significance in the discovery of heart irregularities. These ECG signals are frequently tainted by antiques from various sources. It is basic to diminish these curios and improve the exactness just as dependability to show signs of improvement results identified with heart usefulness. The most commonly disturbed artifact in ECG signals is Motion Artifacts (MA). In this paper, we have proposed a new concept on how machine learning algorithms can be used for de-noising the ECG signals. Towards the goal, a unique combination of Recurrent Neural Network (RNN) and Deep Neural Network (DNN) is used to efficiently remove MA. The proposed algorithm is validated using ECG records obtained from the MIT-BIH Arrhythmia Database. To eliminate MA using the proposed method, we have used Adam optimization algorithm to train and fit the contaminated ECG data in RNN and DNN models. Performance evaluation results in terms of SNR and RRMSE show that the proposed algorithm outperforms other existing MA removal methods without significantly distorting the morphologies of ECG signals.</p></div></div></div>

Download Full-text

Abstract 238: A Deep Neural Network Approach for Continuous Electrocardiogram-based Shock Advisory System During Cardio Pulmonary Resuscitation

Circulation ◽

10.1161/circ.142.suppl_4.238 ◽

2020 ◽

Vol 142 (Suppl_4) ◽

Author(s):

Shirin Hajeb Mohammadalipour ◽

Alicia Cascella ◽

Matt Valentine ◽

K.H. Chon

Keyword(s):

Neural Network ◽

Deep Learning ◽

Deep Neural Network ◽

Learning Algorithm ◽

Motion Artifacts ◽

Learning Approaches ◽

Neural Network Approach ◽

Advisory System ◽

Deep Learning Algorithm ◽

Cardio Pulmonary Resuscitation

The ability of an automatic external defibrillator (AED) to make a reliable shock decision during cardio pulmonary resuscitation (CPR) would improve the survival rate of patients with out-of-hospital cardiac arrest. Since chest compressions induce motion artifacts in the electrocardiogram (ECG), current AEDs instruct the user to stop CPR while an automated rhythm analysis is performed. It has been shown that minimizing interruptions in CPR increases the chance of survival. While deep learning approaches have been used successfully for arrhythmia classification, their performance has not been evaluated for creating an AED shock advisory system that can coexist with CPR. To this end, the objective of this study was to apply a deep-learning algorithm using convolutional layers and residual networks to classify shockable versus non-shockable rhythms in the presence and absence of CPR artifact using only the ECG data. The feasibility of the deep learning method was validated using 8-sec segments of ECG with and without CPR. Two separate databases were used: 1) 40 subjects’ data without CPR from Physionet with 1131 shockable and 2741 non-shockable classified recordings, and 2) CPR artifacts that were acquired from a commercial AED during asystole delivered by 43 different resuscitators. For each 8-second ECG segment, randomly chosen CPR data from 43 different types were added to it so that 5 non-shockable and 10 shockable CPR-contaminated ECG segments were created. We used 30 subjects’ and the remaining 10 for training and test datasets, respectively, for the database 1). For the database 2), we used 33 and 10 subjects’ data for training and testing, respectively. Using our deep neural network model, the sensitivity and specificity of the shock versus no-shock decision for both datasets using the four-fold cross-validation were found to be 95.21% and 86.03%, respectively. For shockable versus non-shockable classification of ECG without CPR artifact, the sensitivity was 99.04% and the specificity was 95.2%. A sensitivity of 94.21% and a specificity of 86.14% were obtained for ECG with CPR artifact. These results meet the AHA sensitivity requirement (>90%).

Download Full-text