Imperial College London

DrPaulBentley

Faculty of MedicineDepartment of Brain Sciences

Senior Clinical Research Fellow
 
 
 
//

Contact

 

p.bentley

 
 
//

Location

 

10L21Charing Cross HospitalCharing Cross Campus

//

Summary

 

Publications

Publication Type
Year
to

105 results found

D'Anna L, Ellis N, Bentley P, Brown Z, Halse O, Jamil S, Jenkins H, Malik A, Kalladka D, Kwan J, Venter M, Banerjee Set al., 2021, Delivering telemedicine consultations for patients with transient ischaemic attack during the COVID-19 pandemic in a comprehensive tertiary stroke centre in the United Kingdom, European Journal of Neurology, Vol: 28, Pages: 3456-3460, ISSN: 1351-5101

Background and purposeThe global COVID-19 pandemic led many stroke centres worldwide to shift from in-person to telemedicine consultations to assess patients with transient ischaemic attacks (TIAs). We aimed to investigate the impact of telemedicine during the COVID-19 pandemic on the management and outcome of the patients with TIA.MethodsWe retrospectively analysed data from a registry of consecutive TIA patients assessed at the Stroke Department, Imperial College Health Care Trust, London, during the national lockdown period (between March 23 2020 and 30 June 2020). As controls, we evaluated the clinical reports and stroke quality metrics of patients presenting to the TIA clinic in the same period of 2019.ResultsBetween 23 March 2020 and 30 June 2020, 136 patients were assessed using the telemedicine TIA clinic, compared to 180 patients evaluated with face-to-face consultation in the same period in 2019. Patients’ characteristics were similar in both groups. At 3 months after the TIA, there were no significant differences in the proportion of patients admitted to the hospital for recurrent TIA/stroke or any other cardiovascular cause from the 2020 period compared to the same period in 2019.ConclusionsOur analysis showed that during the pandemic, our telemedicine consultations of TIA patients were not associated with an increased 3-month rate of recurrent TIA/stroke or cardiovascular hospital admissions. More robust studies looking at this model of care will be needed to assess its long-term effects on patients and health care systems.

Journal article

Broderick M, Almedom L, Burdet E, Burridge J, Bentley Pet al., 2021, Self-Directed Exergaming for Stroke Upper Limb Impairment Increases Exercise Dose Compared to Standard Care, NEUROREHABILITATION AND NEURAL REPAIR, ISSN: 1545-9683

Journal article

Formstone L, Huo W, Wilson S, McGregor A, Bentley P, Vaidyanathan Ret al., 2021, Quantification of motor function post-stroke using wearable inertial and ,echanomyographic Sensors, IEEE Transactions on Neural Systems and Rehabilitation Engineering, Vol: 29, Pages: 1158-1167, ISSN: 1534-4320

Subjective clinical rating scales represent the goldstandard diagnosis of motor function following stroke, however in practice they suffer from well-recognised limitations including variance between assessors, low inter-rater reliability and low resolution. Automated systems have been proposed for empirical quantification but have significantly impacted clinical practice. We address translational challenges in this arena through: (1) implementation of a novel sensor suite fusing inertial measurement and mechanomyography (MMG) to quantify hand and wrist motor function; and (2) introduction of a new range of signal features extracted from the suite to supplement predicted clinical scores. The wearable sensors, signal features, and sensor fusion algorithms have been combined to produce classified ratings from the Fugl-Meyer clinical assessment rating scale. Furthermore, we have designed the system to augment clinical rating with several sensor-derived supplementary features encompassing critical aspects of motor dysfunction (e.g. joint angle, muscle activity, etc.). Performance is validated through a large-scale study on a poststroke cohort of 64 patients. Fugl-Meyer Assessment tasks were classified with 75% accuracy for gross motor tasks and 62% for hand/wrist motor tasks. Of greater import, supplementary features demonstrated concurrent validity with Fugl-Meyer ratings, evidencing their utility as new measures of motor function suited to automated assessment. Finally, the supplementary features also provide continuous measures of sub-components of motor function, offering the potential to complement low accuracy but well-validated clinical rating scales when high-quality motor outcome measures are required. We believe this work provides a basis for widespread clinical adoption of inertial-MMG sensor use for post-stroke clinical motor assessment.Index Terms—Stroke, Fugl-Meyer assessment, automated upper-limb assessment, wearables, machine learning, mechanomyogra

Journal article

D'Anna L, Filippidis FT, Harvey K, Marinescu M, Bentley P, Korompoki E, Veltkamp Ret al., 2021, Extent of white matter lesion is associated with early hemorrhagic transformation in acute ischemic stroke related to atrial fibrillation, Brain and Behavior, ISSN: 2162-3279

BackgroundHemorrhagic transformation (HT) after stroke, related to atrial fibrillation (AF), is a frequent complication, and it can be associated with a delay in the (re-)initiation of oral anticoagulation therapy. We investigated the effect of the presence and severity of white matter disease (WMD) on early HT after stroke related to AF.MethodsA consecutive series of patients with recent (<4 weeks) ischemic stroke and AF, treated at the Hyper Acute Stroke Unit of the Imperial College London between 2010 and 2017, were enrolled. Patients with brain MRI performed 24–72 h from stroke onset and not yet started on anticoagulant treatment were included. WMD was graded using the Fazekas score.ResultsAmong the 441 patients eligible for the analysis, 91 (20.6%) had any HT. Patients with and without HT showed similar clinical characteristics. Patients with HT had a larger diffusion-weighted imaging (DWI) infarct volume compared to patients without HT (p < .001) and significant difference in the distribution of the Fazekas score (p = .001). On multivariable analysis, HT was independently associated with increasing DWI infarct volume (odd ratio (OR), 1.03; 95% confidence interval (CI), 1.01–1.05; p < .001), higher Fazekas scores (OR, 1.94; 95% CI, 1.47–2.57; p < .001) and history of previous intracranial hemorrhage (OR, 4.80; 95% CI, 1.11–20.80; p = .036).ConclusionsPresence and severity of WMD is associated with increased risk of development of early HT in patients with stroke and AF. Further evidence is needed to provide reliable radiological predictors of the risk of HT in cardioembolic stroke.

Journal article

Ken-Dror G, Wade C, Sharma SS, Irvin-Sellers M, Robin J, Fluck D, Bentley P, Sharma Pet al., 2021, SARS-CoV-2 antibody seroprevalence in NHS healthcare workers in a large double-sited UK hospital, CLINICAL MEDICINE, Vol: 21, Pages: E290-E294, ISSN: 1470-2118

Journal article

Kwan J, Brown M, Bentley P, D'Anna L, Hall C, Halse O, Jamil S, Jenkins H, Kalladka D, Patel M, Rane N, Singh A, Venter M, Lobotesis K, Banerjee Set al., 2021, Impact of COVID-19 pandemic on a regional stroke thrombectomy service in the United Kingdom, Cerebrovascular Diseases, Vol: 50, Pages: 178-184, ISSN: 1015-9770

Introduction: We examined the impact of the COVID-19 pandemic on our regional stroke thrombectomy service in the UK. Methods: This was a single-centre health service evaluation. We began testing for COVID-19 on 3 March and introduced a modified “COVID Stroke Thrombectomy Pathway” on 18 March. We included 61 consecutive stroke thrombectomy patients admitted between 1 January and 30 April, and analyzed data on patient demographics, thrombectomy procedures, thrombectomy pathway time-points, and early neurological outcomes. We compared the data for January and February (“pre-COVID”, n=33) vs. March and April (“during COVID”, n=28). Results: Patient demographics were similar between the two groups (mean age 71±12.8 years, 39% female). During the COVID-19 pandemic, a) total stroke admissions fell by 17% but the stroke thrombectomy rate was maintained at 17%; b) successful recanalization rate was maintained at 81%; c) early neurological outcomes (neurological improvement following thrombectomy and inpatient mortality) were not significantly different; d) use of general anesthesia fell significantly from 85% to 32% as intended; and e) time intervals from onset to arrival, groin puncture, and recanalization were not significantly different, whereas internal delays for external referrals were significantly reduced for door-to-groin puncture [48 (IQR 39-57) vs. 33 (IQR 27-44) minutes, p=.013] and door-to-recanalization [82·5 (IQR 61-110) vs. 60 (IQR 55-70) minutes, p=.018].Conclusion: The COVID-19 pandemic had lowered stroke admission numbers but not stroke thrombectomy rate, successful recanalization rate, or chance of a favorable outcome. Internal delays were actually reduced during the COVID-19 pandemic. Further studies can examine the effects of COVID-19 pandemic on longer term outcome after stroke thrombectomy.

Journal article

D'Anna L, Brown M, Oishi S, Ellis N, Brown Z, Bentley P, Drumm B, Halse O, Jamil S, Jenkins H, Malik A, Kalladka D, Venter M, Kwan J, Banerjee Set al., 2021, Impact of national lockdown on the hyperacute stroke care and rapid transient ischaemic attack outpatient service in a comprehensive tertiary stroke centre during the COVID-19 pandemic, Frontiers in Neurology, Vol: 12, ISSN: 1664-2295

Background: The COVID-19 pandemic is having major implications for stroke services worldwide. We aimed to study the impact of the national lockdown period during the COVID-19 outbreak on stroke and transient ischemic attack (TIA) care in London, UK. Methods: We retrospectively analyzed data from a quality improvement registry of consecutive patients presenting with acute ischemic stroke and TIA to the Stroke Department, Imperial College Health Care Trust London during the national lockdown period (between March 23rd and 30th June 2020). As controls, we evaluated the clinical reports and stroke quality metrics of patients presenting with stroke or TIA in the same period of 2019. Results: Between March 23rd and 30th June 2020, we documented a fall in the number of stroke admissions by 31.33% and of TIA outpatient referrals by 24.44% compared to the same period in 2019. During the lockdown, we observed a significant increase in symptom onset-to-door time in patients presenting with stroke (median = 240 vs. 160 min, p = 0.020) and TIA (median = 3 vs. 0 days, p = 0.002) and a significant reduction in the total number of patients thrombolysed [27 (11.49%) vs. 46 (16.25%, p = 0.030)]. Patients in the 2020 cohort presented with a lower median pre-stroke mRS (p = 0.015), but an increased NIHSS (p = 0.002). We registered a marked decrease in mimic diagnoses compared to the same period of 2019. Statistically significant differences were found between the COVID and pre-COVID cohorts in the time from onset to door (median 99 vs. 88 min, p = 0.026) and from onset to needle (median 148 vs. 126 min, p = 0.036) for thrombolysis whilst we did not observe any significant delay to reperfusion therapies (door-to-needle and door-to-groin puncture time). Conclusions: National lockdown in the UK due to the COVID-19 pandemic was associated with a significant decrease in acute stroke admission and TIA evaluations at our stroke center. Moreover, a lower proportion of acute stroke patients in

Journal article

Clarke AK, Atashzar SF, Vecchio AD, Barsakcioglu D, Muceli S, Bentley P, Urh F, Holobar A, Farina Det al., 2021, Deep learning for robust decomposition of high-density surface EMG signals, IEEE Transactions on Biomedical Engineering, Vol: 68, Pages: 526-534, ISSN: 0018-9294

Blind source separation (BSS) algorithms, such as gradient convolution kernel compensation (gCKC), can efficiently and accurately decompose high-density surface electromyography (HD-sEMG) signals into constituent motor unit (MU) action potential trains. Once the separation matrix is blindly estimated on a signal interval, it is also possible to apply the same matrix to subsequent signal segments. Nonetheless, the trained separation matrices are sub-optimal in noisy conditions and require that incoming data undergo computationally expensive whitening. One unexplored alternative is to instead use the paired HD-sEMG signal and BSS output to train a model to predict MU activations within a supervised learning framework. A gated recurrent unit (GRU) network was trained to decompose both simulated and experimental unwhitened HD-sEMG signal using the output of the gCKC algorithm. The results on the experimental data were validated by comparison with the decomposition of concurrently recorded intramuscular EMG signals. The GRU network outperformed gCKC at low signal-to-noise ratios, proving superior performance in generalising to new data. Using 12 seconds of experimental data per recording, the GRU performed similarly to gCKC, at rates of agreement of 92.5% (84.5%-97.5%) and 94.9% (88.8%-100.0%) respectively for GRU and gCKC against matched intramuscular sources.

Journal article

Auepanwiriyakul C, Waibel S, Songa J, Bentley P, Faisal AAet al., 2020, Accuracy and acceptability of wearable motion tracking for inpatient monitoring using smartwatches, Sensors, Vol: 20, ISSN: 1424-8220

Inertial Measurement Units (IMUs) within an everyday consumer smartwatch offer a convenient and low-cost method to monitor the natural behaviour of hospital patients. However, their accuracy at quantifying limb motion, and clinical acceptability, have not yet been demonstrated. To this end we conducted a two-stage study: First, we compared the inertial accuracy of wrist-worn IMUs, both research-grade (Xsens MTw Awinda, and Axivity AX3) and consumer-grade (Apple Watch Series 3 and 5), and optical motion tracking (OptiTrack). Given the moderate to strong performance of the consumer-grade sensors, we then evaluated this sensor and surveyed the experiences and attitudes of hospital patients (N = 44) and staff (N = 15) following a clinical test in which patients wore smartwatches for 1.5–24 h in the second study. Results indicate that for acceleration, Xsens is more accurate than the Apple Series 5 and 3 smartwatches and Axivity AX3 (RMSE 1.66 ± 0.12 m·s−2; R2 0.78 ± 0.02; RMSE 2.29 ± 0.09 m·s−2; R2 0.56 ± 0.01; RMSE 2.14 ± 0.09 m·s−2; R2 0.49 ± 0.02; RMSE 4.12 ± 0.18 m·s−2; R2 0.34 ± 0.01 respectively). For angular velocity, Series 5 and 3 smartwatches achieved similar performances against Xsens with RMSE 0.22 ± 0.02 rad·s−1; R2 0.99 ± 0.00; and RMSE 0.18 ± 0.01 rad·s−1; R2 1.00± SE 0.00, respectively. Surveys indicated that in-patients and healthcare professionals strongly agreed that wearable motion sensors are easy to use, comfortable, unobtrusive, suitable for long-term use, and do not cause anxiety or limit daily activities. Our results suggest that consumer smartwatches achieved moderate to strong levels of accuracy compared to laboratory gold-standard and are acceptable for pervasive monitoring of motion/behaviour within hospital settings.

Journal article

Auepanwiriyakul C, Waibel S, Songa J, Bentley P, Faisal AAet al., 2020, Accuracy and Acceptability of Wearable Motion Tracking Smartwatches for Inpatient Monitoring, Sensors, ISSN: 1424-8220

<jats:p>: Inertial Measurement Units (IMUs) within an everyday consumer smartwatch offer a convenient and low-cost method to monitor the natural behaviour of hospital patients. However, their accuracy at quantifying limb motion, and clinical acceptability, have not yet been demonstrated. To this end we conducted a two-stage study: First, we compared the inertial accuracy of wrist-worn IMUs, both research-grade (Xsens MTw Awinda, and Axivity AX3) and consumer-grade (Apple Watch Series 3 and 5), relative to gold-standard optical motion tracking (OptiTrack). Given the moderate to the strong performance of the consumer-grade sensors we then evaluated this sensor and surveyed the experiences and attitudes of hospital patients (N=44) and staff (N=15) following a clinical test in which patients wore smartwatches for 1.5-24 hours in the second study. Results indicate that for acceleration, Xsens is more accurate than the Apple smartwatches and Axivity AX3 (RMSE 0.17+/-0.01 g; R2 0.88+/-0.01; RMSE 0.22+/-0.01 g; R2 0.64+/-0.01; RMSE 0.42+/-0.01 g; R2 0.43+/-0.01, respectively). However, for angular velocity, the smartwatches are marginally more accurate than Xsens (RMSE 1.28+/-0.01 rad/s; R2 0.85+/-0.00; RMSE 1.37+/-0.01 rad/s; R2 0.82+/-0.01, respectively). Surveys indicated that in-patients and healthcare professionals strongly agreed that wearable motion sensors are easy to use, comfortable, unobtrusive, suitable for long term use, and do not cause anxiety or limit daily activities. Our results suggest that smartwatches achieved moderate to strong levels of accuracy compared to a gold-standard reference and are likely to be accepted as a pervasive measure of motion/behaviour within hospitals.</jats:p>

Journal article

D'Anna L, Filippidis FT, Antony S, Brown Z, Wyatt H, Malik A, Sivakumaran P, Harvey K, Marinescu M, Bentley P, Korompoki E, Veltkamp Ret al., 2020, Early initiation of direct anticoagulation after stroke in patients with atrial fibrillation., European Journal of Neuroscience, Vol: 27, Pages: 2168-2175, ISSN: 0953-816X

BACKGROUND: The safety of early initiation of anticoagulant therapy in patients with ischaemic stroke related to atrial fibrillation (AF) is unknown. We investigated the safety of early initiation of direct oral anticoagulants (DOACs), vitamin K antagonists (VKAs) or no anticoagulation. METHODS: This observational, retrospective, single-centre study included consecutive patients with recent (< 4 weeks) ischaemic stroke and AF. The primary outcome was the rate of major (intra- and extracranial) bleeding in patients on different treatment schemes: DOACs, VKAs and not anticoagulated. We also investigated the rate of ischaemic cerebrovascular events and mortality. RESULTS: We included 959 consecutive patients with AF and ischaemic stroke followed up for an average time of 16.1 days after the index event. 559 patients of 959 (58.3%) were anticoagulated with either VKAs (259) or DOACs (300). Anticoagulation was started after a mean time of 7± 9.4 in the DOACs group and 11.9± 19.7 in the VKAs group. Early initiation of any anticoagulant was not associated with an increased risk of any major bleeding (OR 0.49; CI, 0.21-1.16) and in particular of intracranial bleeding (OR 0.47; CI, 0.17-1.29; p = 0.143) compared with no anticoagulation. In contrast to VKAs (OR 0.78; CI, 0.28-2.13), treatment with DOACs (OR 0.32; CI, 0.10-0.96) reduced the rate of major bleeding compared to no-anticoagulation. Early recurrences of ischaemic stroke did not differ significantly among the three groups. CONCLUSIONS: Starting DOACs within a mean time of 7 days after stroke appears safe. Randomised controlled studies are needed to establish the added efficacy of starting anticoagulation early after stroke.

Journal article

Bentley P, Sharma P, 2020, Distinguishing early from late seizures after cerebral venous thrombosis Cinderepilepsy, NEUROLOGY, Vol: 95, Pages: 513-514, ISSN: 0028-3878

Journal article

Bentley P, 2020, Magnetic Resonance Imaging for Acute Minor Neurological Symptoms Good for Ruling Stroke in, Not Out, JAMA NEUROLOGY, Vol: 77, Pages: 775-+, ISSN: 2168-6149

Journal article

Li K, Bentley P, Nair A, Halse O, Barker G, Russell C, Soto D, Malhotra PAet al., 2020, Reward sensitivity predicts dopaminergic response in spatial neglect, Cortex, Vol: 122, Pages: 213-224, ISSN: 0010-9452

It has recently been revealed that spatial neglect can be modulated by motivational factors including anticipated monetary reward. A number of dopaminergic agents have been evaluated as treatments for neglect, but the results have been mixed, with no clear anatomical or cognitive predictors of dopaminergic responsiveness. Given that the effects of incentive motivation are mediated by dopaminergic pathways that are variably damaged in stroke, we tested the hypothesis that the modulatory influences of reward and dopaminergic drugs on neglect are themselves related.We employed a single-dose, double-blind, crossover design to compare the effects of Co-careldopa and placebo on a modified visual cancellation task in patients with neglect secondary to right hemisphere stroke. Whilst confirming that reward improved visual search in this group, we showed that dopaminergic stimulation only enhances visual search in the absence of reward. When patients were divided into REWARD-RESPONDERs and REWARD-NON-RESPONDERs, we found an interaction, such that only REWARD-NON-RESPONDERs showed a positive response to reward after receiving Co-careldopa, whereas REWARD-RESPONDERs were not influenced by drug. At a neuroanatomical level, responsiveness to incentive motivation was most associated with intact dorsal striatum.These findings suggest that dopaminergic modulation of neglect follows an ‘inverted U’ function, is dependent on integrity of the reward system, and can be measured as a behavioural response to anticipated reward.

Journal article

Ken-Dror G, Wade C, Sharma S, Law J, Russo C, Sharma A, Joy E, John J, Robin J, John S, Mahana K, Fluck D, Bentley P, Sharma Pet al., 2020, COVID-19 outcomes in UK centre within highest health and wealth band: a prospective cohort study, BMJ OPEN, Vol: 10, ISSN: 2044-6055

Journal article

Gasimova A, Seegoolam G, Chen L, Bentley P, Rueckert Det al., 2020, Spatial Semantic-Preserving Latent Space Learning for Accelerated DWI Diagnostic Report Generation, Pages: 333-342, ISSN: 0302-9743

In light of recent works exploring automated pathological diagnosis, studies have also shown that medical text reports can be generated with varying levels of efficacy. Brain diffusion-weighted MRI (DWI) has been used for the diagnosis of ischaemia in which brain death can follow in immediate hours. It is therefore of the utmost importance to obtain ischaemic brain diagnosis as soon as possible in a clinical setting. Previous studies have shown that MRI acquisition can be accelerated using variable-density Cartesian undersampling methods. In this study, we propose an accelerated DWI acquisition pipeline for the purpose of generating text reports containing diagnostic information. We demonstrate that we can learn a semantic-preserving latent space for minor as well as extremely undersampled MR images capable of achieving promising results on a diagnostic report generation task.

Conference paper

Chen L, Bentley P, Mori K, Misawa K, Fujiwara M, Rueckert Det al., 2019, Self-supervised learning for medical image analysis using image context restoration, Medical Image Analysis, Vol: 58, Pages: 1-12, ISSN: 1361-8415

Machine learning, particularly deep learning has boosted medical image analysis over the past years. Training a good model based on deep learning requires large amount of labelled data. However, it is often difficult to obtain a sufficient number of labelled images for training. In many scenarios the dataset in question consists of more unlabelled images than labelled ones. Therefore, boosting the performance of machine learning models by using unlabelled as well as labelled data is an important but challenging problem. Self-supervised learning presents one possible solution to this problem. However, existing self-supervised learning strategies applicable to medical images cannot result in significant performance improvement. Therefore, they often lead to only marginal improvements. In this paper, we propose a novel self-supervised learning strategy based on context restoration in order to better exploit unlabelled images. The context restoration strategy has three major features: 1) it learns semantic image features; 2) these image features are useful for different types of subsequent image analysis tasks; and 3) its implementation is simple. We validate the context restoration strategy in three common problems in medical imaging: classification, localization, and segmentation. For classification, we apply and test it to scan plane detection in fetal 2D ultrasound images; to localise abdominal organs in CT images; and to segment brain tumours in multi-modal MR images. In all three cases, self-supervised learning based on context restoration learns useful semantic features and lead to improved machine learning models for the above tasks.

Journal article

Math N, Han TS, Lubomirova I, Hill R, Bentley P, Sharma Pet al., 2019, Influences of genetic variants on stroke recovery: a meta-analysis of the 31,895 cases, Neurological Sciences, Vol: 40, Pages: 2437-2445, ISSN: 1590-1874

BackgroundThe influences of genetic variants on functional clinical outcomes following stroke are unclear. In order to reliably quantify these influences, we undertook a comprehensive meta-analysis of outcomes after acute intracerebral haemorrhage (ICH) or ischaemic stroke (AIS) in relation to different genetic variants.MethodsPubMed, PsycInfo, Embase and Medline electronic databases were searched up to January 2019. Outcomes, defined as favourable or poor, were assessed by validated scales (Barthel index, modified Rankin scale, Glasgow outcome scale and National Institutes of Health stroke scale).ResultsNinety-two publications comprising 31,895 cases met our inclusion criteria. Poor outcome was observed in patients with ICH who possessed the APOE4 allele: OR =2.60 (95% CI = 1.25–5.41, p = 0.01) and in AIS patients with the GA or AA variant at the BDNF-196 locus: OR = 2.60 (95% CI = 1.25–5.41, p = 0.01) or a loss of function allele of CYP2C19: OR = 2.36 (95% CI = 1.56–3.55, p < 0.0001). Poor outcome was not associated with APOE4: OR = 1.02 (95% CI = 0.81–1.27, p = 0.90) or IL6-174 G/C: OR = 2.21 (95% CI = 0.55–8.86, p = 0.26) in patients with AIS.ConclusionsWe demonstrate that recovery of AIS was unfavourably associated with variants of BDNF and CYP2C19 genes whilst recovery of ICH was unfavourably associated with APOE4 gene.

Journal article

Chen L, Bentley P, Mori K, Misawa K, Fujiwara M, Rueckert Det al., 2019, Intelligent image synthesis to attack a segmentation CNN using adversarial learning, Simulation and Synthesis in Medical Imaging. SASHIMI 2019, Publisher: Springer Verlag, ISSN: 0302-9743

Deep learning approaches based on convolutional neural networks (CNNs) have been successful in solving a number of problems in medical imaging, including image segmentation. In recent years, it has been shown that CNNs are vulnerable to attacks in which the input image is perturbed by relatively small amounts of noise so that the CNN is no longer able to perform a segmentation of the perturbed image with sufficient accuracy. Therefore, exploring methods on how to attack CNN-based models as well as how to defend models against attacks have become a popular topic as this also provides insights into the performance and generalization abilities of CNNs. However, most of the existing work assumes unrealistic attack models, i.e. the resulting attacks were specified in advance. In this paper, we propose a novel approach for generating adversarial examples to attack CNN-based segmentation models for medical images. Our approach has three key features: (1) The generated adversarial examples exhibit anatomical variations (in form of deformations) as well as appearance perturbations; (2) The adversarial examples attack segmentation models so that the Dice scores decrease by a pre-specified amount; (3) The attack is not required to be specified beforehand. We have evaluated our approach on CNN-based approaches for the multi-organ segmentation problem in 2D CT images. We show that the proposed approach can be used to attack different CNN-based segmentation models.

Conference paper

Bentley P, Lovell B, 2019, Memorizing Medicine: Second Edition, Publisher: CRC Press, ISBN: 978-1138332690

Book

Hill R, Han TS, Lubomirova I, Math N, Bentley P, Sharma Pet al., 2019, Prothrombin Complex Concentrates are Superior to Fresh Frozen Plasma for Emergency Reversal of Vitamin K Antagonists: A Meta-Analysis in 2606 Subjects, Drugs, ISSN: 0012-6667

Journal article

Lotay R, Mace M, Rinne P, Burdet E, Bentley Pet al., 2019, optimizing self-exercise scheduling in motor stroke using Challenge Point Framework theory, 2019 IEEE 16th International Conference on Rehabilitation Robotics (ICORR), Publisher: IEEE, Pages: 435-440

An important challenge for technology-assisted self-led rehabilitation is how to automate appropriate schedules of exercise that are responsive to patients’ needs, and optimal for learning. While random scheduling has been found to be superior for long-term learning relative to fixed scheduling (Contextual Interference), this method is limited by not adequately accounting for task difficulty, or skill acquisition during training. One method that combines contextual interference with adaptation of the challenge to the skill-level of the player is Challenge Point Framework (CPF) theory. In this pilot study we test whether self-led motor training based upon CPF scheduling achieves faster learning than deterministic, fixed scheduling. Training was implemented in a mobile gaming device adapted for arm disability, allowing for grip and wrist exercises. We tested 11 healthy volunteers and 12 hemiplegic stroke patients in a single-blinded no crossover controlled randomized trial. Results suggest that patients training with CPF-based adaption performed better than those training with fixed conditions. This was not seen for healthy volunteers whose performance was close to ceiling. Further data collection is required to determine the significance of the results.

Conference paper

Formstone L, Pucek M, Wilson S, Bentley P, McGregor A, Vaidyanathan Ret al., 2019, Myographic Information Enables Hand Function Classification in Automated Fugl-Meyer Assessment, 9th IEEE/EMBS International Conference on Neural Engineering (NER), Publisher: IEEE, Pages: 239-242, ISSN: 1948-3546

Conference paper

Mousele C, Bentley P, Tai YF, 2018, A rare presentation of orthostatic tremor as abdominal tremor, Tremor and Other Hyperkinetic Movements, Vol: 8, ISSN: 2160-8288

Background: Orthostatic tremor (OT) is a weight-bearing hyperkinetic disorder characterized by unsteadiness while standing that is relieved when sitting or walking.Case report: A 66-year-old male presented with a 5 year-history of tremor in his abdomen, but only when he stood in a stationary position. The tremor disappeared when he stood or walked. On examination, he had palpable tremor in his rectus abdominis and gastrocnemius virtually instantaneously after standing. His electromyography findings confirmed the presence of a 12-Hz tremor in the tibialis anterior while standing, with subharmonics recorded in the external obliques and rectus abdominis.Discussion: Our case illustrates an unusual presentation of OT. The diagnosis is supported by its characteristic frequency and specific appearance only during upright stance.

Journal article

Mousele C, Bentley P, Tai YF, 2018, A Rare Presentation of Orthostatic Tremor as Abdominal Tremor, Tremor and Other Hyperkinetic Movements, Vol: 8, Pages: 603-603

Journal article

Bowles C, Liang C, Bentley P, Guerrero R, Gunn R, Hammers A, Dickie D, Hernandez M, Wardlaw J, Rueckert Det al., 2018, Gan augmentation: augmenting training data using generative adversarial networks

One of the biggest issues facing the use of machine learning in medical imaging is the lack of availability of large, labelled datasets. The annotation of medical images is not only expensive and time-consuming but also highly dependent on the availability of expert observers. The limited amount of training data can inhibit the performance of supervised machine learning algorithms which often need very large quantities of data on which to train to avoid overfitting. So far, much effort has been directed at extracting as much information as possible from what data is available. Generative Adversarial Networks (GANs) offer a novel way to unlock additional information from a dataset by generating synthetic samples with the appearance of real images. This paper demonstrates the feasibility of introducing GAN derived synthetic data to the training datasets in two brain segmentation tasks, leading to improvements in Dice Similarity Coefficient(DSC) of between 1 and 5 percentage points under different conditions, with the strongest effects seen fewer than ten training image stacks are available.

Working paper

Chen L, Carlton Jones AL, Mair G, Patel R, Gontsarova A, Ganesalingam J, Math N, Dawson A, Basaam A, Cohen D, Mehta A, Wardlaw J, Rueckert D, Bentley Pet al., 2018, Rapid automated quantification of cerebral leukoaraiosis on CT: a multicentre validation study, Radiology, Vol: 288, Pages: 573-581, ISSN: 0033-8419

Purpose - To validate a fully-automated, machine-learning method (random forest) for segmenting cerebral white matter lesions (WML) on computerized tomography (CT). Materials and Methods – A retrospective sample of 1082 acute ischemic stroke cases was obtained, comprising unselected patients: 1) treated with thrombolysis; or 2) undergoing contemporaneous MR imaging and CT; and 3) a subset of IST-3 trial participants. Automated (‘Auto’) WML images were validated relative to experts’ manual tracings on CT, and co-registered FLAIR-MRI; and ratings using two conventional ordinal scales. Analyses included correlations between CT and MR imaging volumes, and agreements between Auto and expert ratings.Results - Auto WML volumes correlated strongly with expert-delineated WML volumes on MR imaging and on CT (r2=0.85, 0.71 respectively; p<0.001). Spatial-similarity of Auto-maps, relative to MRI-WML, was not significantly different to that of expert CT-WML tracings. Individual expert CT-WML volumes correlated well with each other (r2=0.85), but varied widely (range: 91% of mean estimate; median 11 cc; range: 0.2 – 68 cc). Agreements between Auto and consensus-expert ratings were superior or similar to agreements between individual pairs of experts (kappa: 0.60, 0.64 vs. 0.51, 0.67 for two score systems; p<0.01 for first comparison). Accuracy was unaffected by established infarction, acute ischemic changes, or atrophy (p>0.05). Auto preprocessing failure rate was 4%; rating errors occurred in a further 4%. Total Auto processing time averaged 109s (range: 79 - 140 s). Conclusion - An automated method for quantifying CT cerebral white matter lesions achieves a similar accuracy to experts in unselected and multicenter cohorts.

Journal article

Bentley P, Sharma P, 2018, Neurological disorders - epilepsy, Parkinson's disease and multiple sclerosis, Clinical pharmacology: 12th edition, Editors: Brown, Sharma, Mir, Bennett, Publisher: Elsevier, ISBN: 978-0702073281

Book chapter

Chen L, Bentley P, Mori K, Misawa K, Fujiwara M, Rueckert Det al., 2018, DRINet for medical image segmentation, IEEE Transactions on Medical Imaging, Vol: 37, Pages: 2453-2462, ISSN: 0278-0062

Convolutional neural networks (CNNs) have revolutionized medical image analysis over the past few years. The UNet architecture is one of the most well-known CNN architectures for semantic segmentation and has achieved remarkable successes in many different medical image segmentation applications. The U-Net architecture consists of standard convolution layers, pooling layers, and upsampling layers. These convolution layers learn representative features of input images and construct segmentations based on the features. However, the features learned by standard convolution layers are not distinctive when the differences among different categories are subtle in terms of intensity, location, shape, and size. In this paper, we propose a novel CNN architecture, called Dense-Res-Inception Net (DRINet), which addresses this challenging problem. The proposed DRINet consists of three blocks, namely a convolutional block with dense connections, a deconvolutional block with residual Inception modules, and an unpooling block. Our proposed architecture outperforms the U-Net in three different challenging applications, namely multi-class segmentation of cerebrospinal fluid (CSF) on brain CT images, multi-organ segmentation on abdominal CT images, multi-class brain tumour segmentation on MR images.

Journal article

Rinne P, Hassan M, Fernandes C, Han E, Hennessy E, Waldman A, Sharma P, Soto D, Leech R, Malhotra P, Bentley Pet al., 2018, Motor dexterity and strength depend upon integrity of the attention-control system, Proceedings of the National Academy of Sciences, Vol: 115, Pages: E536-E545, ISSN: 0027-8424

Attention control (or executive control) is a higher cognitive function involved in response selection and inhibition, through close interactions with the motor system. Here, we tested whether influences of attention control are also seen on lower level motor functions of dexterity and strength—by examining relationships between attention control and motor performance in healthy-aged and hemiparetic-stroke subjects (n = 93 and 167, respectively). Subjects undertook simple-tracking, precision-hold, and maximum force-generation tasks, with each hand. Performance across all tasks correlated strongly with attention control (measured as distractor resistance), independently of factors such as baseline performance, hand use, lesion size, mood, fatigue, or whether distraction was tested during motor or nonmotor cognitive tasks. Critically, asymmetric dissociations occurred in all tasks, in that severe motor impairment coexisted with normal (or impaired) attention control whereas normal motor performance was never associated with impaired attention control (below a task-dependent threshold). This implies that dexterity and force generation require intact attention control. Subsequently, we examined how motor and attention-control performance mapped to lesion location and cerebral functional connectivity. One component of motor performance (common to both arms), as well as attention control, correlated with the anatomical and functional integrity of a cingulo-opercular “salience” network. Independently of this, motor performance difference between arms correlated negatively with the integrity of the primary sensorimotor network and corticospinal tract. These results suggest that the salience network, and its attention-control function, are necessary for virtually all volitional motor acts while its damage contributes significantly to the cardinal motor deficits of stroke.

Journal article

This data is extracted from the Web of Science and reproduced under a licence from Thomson Reuters. You may not copy or re-distribute this data in whole or in part without the written consent of the Science business of Thomson Reuters.

Request URL: http://wlsprd.imperial.ac.uk:80/respub/WEB-INF/jsp/search-html.jsp Request URI: /respub/WEB-INF/jsp/search-html.jsp Query String: respub-action=search.html&id=00525846&limit=30&person=true