Search

Machine-learning models for medical tasks can match or surpass the performance of clinical experts. However, in settings differing from those of the training dataset, the performance of a model can deteriorate substantially. Here we report a representation-learning strategy for machine-learning models applied to medical-imaging tasks that mitigates such 'out of distribution' performance problem and that improves model robustness and training efficiency. The strategy, which we named REMEDIS (for 'Robust and Efficient Medical Imaging with Self-supervision'), combines large-scale supervised transfer learning on natural images and intermediate contrastive self-supervised learning on medical images and requires minimal task-specific customization. We show the utility of REMEDIS in a range of diagnostic-imaging tasks covering six imaging domains and 15 test datasets, and by simulating three realistic out-of-distribution scenarios. REMEDIS improved in-distribution diagnostic accuracies up to 11.5% with respect to strong supervised baseline models, and in out-of-distribution settings required only 1-33% of the data for retraining to match the performance of supervised models retrained using all available data. REMEDIS may accelerate the development lifecycle of machine-learning models for medical imaging., (© 2023. The Author(s), under exclusive licence to Springer Nature Limited.)

Introduction: The Crescent ® is a recently introduced dual lumen cannula by which veno-venous extracorporeal membrane oxygenation (VV ECMO) is delivered. It has a number of features that enhance its ease of placement, pressure-flow dynamics and may reduce catheter-related complication rates., Methods: We present the first case series of its kind analysing this device by means of a retrospective observational study of prospectively collected data from the first year of its use in a high volume severe acute respiratory failure centre (Glenfield, UK). We compare complication rates of the Crescent ® , with data from the international ELSO database and our own historic centre data and discuss subjective clinician experience of introducing this device., Results: Over the first 12 months of its use (23/09/2019-23/09/2020), 54 patients were cannulated using a Crescent ® catheter. There were no serious/life-threatening adverse events and a low number of minor cannula-related complications. Subjectively users found it has a number of advantages over other devices and configurations, not captured within current data collection frameworks., Conclusion: The Crescent ® is a safe and effective device by which to deliver VV ECMO support to patients with severe acute respiratory failure.

Study Objective: To evaluate the quality of postoperative pain relief during the first 3 days after surgery and to evaluate with the incidence of persistent pain at 6 months after surgery., Design: Retrospective single-center audit., Setting: University hospital., Patients: Five hundred four patients who underwent thoracotomy., Interventions: Review of patient records, questionnaire, and telephone review., Results: Of the 364 survivors, 306 were contacted. Five or more episodes of severe pain (numerical rating scale >6/10 at rest or movement) during the first 72 hours after surgery occurred in 133 patients. Persistent postsurgical pain at 6 months was present in 82% (109/133) of these patients. Patient satisfaction with acute postoperative pain management was excellent (36%), good (43%), and fair or poor (21%).The incidence of postthoracotomy pain was 56% (mild 32%, moderate 18%, and severe 6%)., Conclusions: Poorly controlled acute postoperative pain correlated with persistent postsurgical pain at 6 months. In view of such a high incidence in thoracotomy patients, preventative strategies assume great significance., (Copyright © 2016 Elsevier Inc. All rights reserved.)

We propose an efficient optimization algorithm to select a subset of training data as the inducing set for sparse Gaussian process regression. Previous methods either use different objective functions for inducing set and hyperparameter selection, or else optimize the inducing set by gradient-based continuous optimization. The former approaches are harder to interpret and suboptimal, whereas the latter cannot be applied to discrete input domains or to kernel functions that are not differentiable with respect to the input. The algorithm proposed in this work estimates an inducing set and the hyperparameters using a single objective. It can be used to optimize either the marginal likelihood or a variational free energy. Space and time complexity are linear in training set size, and the algorithm can be applied to large regression problems on discrete or continuous domains. Empirical evaluation shows state-of-art performance in discrete cases, competitive prediction results as well as a favorable trade-off between training and test time in continuous cases.

Study Objective: Transversus abdominis plane (TAP) blocks have been reported to be an effective method of providing analgesia after abdominal surgery. To perform a prospective audit on the effectiveness of a novel technique of providing continuous transversus abdominis plane (TAP) analgesia in patients undergoing emergency and elective abdominal surgery., Design: Prospective single center audit over a 3-year period., Setting: University hospital., Patients: One hundred twenty-four American Society of Anesthesiologists I to IV adult patients presenting for elective as well as emergency abdominal surgery in whom epidural analgesia was contraindicated or refused., Interventions: Four quadrant TAP blocks and continuous TAP analgesia., Measurements: Numerical rating scale pain scores at rest and on coughing, nausea scores, satisfaction scores, complications, frequency of analgesia failure, therapeutic failure with continuous TAP analgesia and opioid consumption., Results: One hundred twenty-four patients who received continuous TAP analgesia were audited. This included 34 patients for elective open surgery, 36 patients for emergency laparotomy, and 54 patients who underwent elective laparoscopic colorectal surgery. Surgical incision was within the dermatomal limit of the block in 70% of the patients (88/124). Therapeutic failure with the technique was 10%. Frequency of analgesic failure over the 48-hour period was none in 39% and below 5 episodes in 57%., Conclusion: Four quadrant transversus abdominis plane blocks and continuous TAP analgesia is an effective technique for providing postoperative analgesia after abdominal surgery. It has the potential to be used as a sole analgesic technique when the surgical incision is within its dermatomal limit., (Copyright © 2015 Elsevier Inc. All rights reserved.)

An important factor in the universal failure in phase III trials in mild to moderate Alzheimer's disease in the past decade is the lack of phase II clinical data prior to entering phase III, with common reliance on biomarker results alone. Conduction of two learn-confirm cycles according to the Sheiner model would allow go/no-go decision making to include reliable clinical efficacy data prior to conducting phase III and would likely bring the rate of late stage failure more into line with that of other neurological indications. In studies in earlier disease stages, combined phase IIB/III adaptive approaches merit consideration in view of the long timelines of each study, though advantages and disadvantages of this approach versus the classical development pathway still need careful assessment.

Spinal cord segmentation is a developing area of research intended to aid the processing and interpretation of advanced magnetic resonance imaging (MRI). For example, high resolution three-dimensional volumes can be segmented to provide a measurement of spinal cord atrophy. Spinal cord segmentation is difficult due to the variety of MRI contrasts and the variation in human anatomy. In this study we propose a new method of spinal cord segmentation based on one-dimensional template matching and provide several metrics that can be used to compare with other segmentation methods. A set of ground-truth data from 10 subjects was manually-segmented by two different raters. These ground truth data formed the basis of the segmentation algorithm. A user was required to manually initialize the spinal cord center-line on new images, taking less than one minute. Template matching was used to segment the new cord and a refined center line was calculated based on multiple centroids within the segmentation. Arc distances down the spinal cord and cross-sectional areas were calculated. Inter-rater validation was performed by comparing two manual raters (n = 10). Semi-automatic validation was performed by comparing the two manual raters to the semi-automatic method (n = 10). Comparing the semi-automatic method to one of the raters yielded a Dice coefficient of 0.91 +/- 0.02 for ten subjects, a mean distance between spinal cord center lines of 0.32 +/- 0.08 mm, and a Hausdorff distance of 1.82 +/- 0.33 mm. The absolute variation in cross-sectional area was comparable for the semi-automatic method versus manual segmentation when compared to inter-rater manual segmentation. The results demonstrate that this novel segmentation method performs as well as a manual rater for most segmentation metrics. It offers a new approach to study spinal cord disease and to quantitatively track changes within the spinal cord in an individual case and across cohorts of subjects.

In the North Sea, the amount of litter present in the marine environment represents a severe environmental problem. In order to assess the magnitude of the problem and measure changes in abundance, the results of two beach litter monitoring programmes were compared and analysed for long-term trends applying multivariate techniques. Total beach litter pollution was persistently high. Spatial differences in litter abundance made it difficult to identify long-term trends: Partly more than 8000 litter items year(-1) were recorded on a 100 m long survey site on the island of Scharhörn, while the survey site on the beach on the island of Amrum revealed abundances lower by two orders of magnitude. Beach litter was dominated by plastic with mean proportions of 52%-91% of total beach litter. Non-parametric time series analyses detected many significant trends, which, however, did not show any systematic spatial patterns. Cluster analyses partly led to groupings of beaches according to their expositions to sources of litter, wind and currents. Surveys in short intervals of one to two weeks were found to give higher annual sums of beach litter than the quarterly surveys of the OSPAR method. Surveys at regular intervals of four weeks to five months would make monitoring results more reliable., (Copyright © 2015 Elsevier Ltd. All rights reserved.)

Background and Purpose: Advanced MR imaging techniques are critical to understanding the pathophysiology of conditions involving the spinal cord. We provide a novel, quantitative solution to map vertebral and spinal cord levels accounting for anatomic variability within the human spinal cord. For the first time, we report a population distribution of the segmental anatomy of the cervical spinal cord that has direct implications for the interpretation of advanced imaging studies most often conducted across groups of subjects., Materials and Methods: Twenty healthy volunteers underwent a T2-weighted, 3T MRI of the cervical spinal cord. Two experts marked the C3-C8 cervical nerve rootlets, C3-C7 vertebral bodies, and pontomedullary junction. A semiautomated algorithm was used to locate the centerline of the spinal cord and measure rostral-caudal distances from a fixed point in the brain stem, the pontomedullary junction, to each of the spinal rootlets and vertebral bodies. Distances to each location were compared across subjects. Six volunteers had 2 additional scans in neck flexion and extension to measure the effects of patient positioning in the scanner., Results: We demonstrated that substantial variation exists in the rostral-caudal position of spinal cord segments among individuals and that prior methods of predicting spinal segments are imprecise. We also show that neck flexion or extension has little effect on the relative location of vertebral-versus-spinal levels., Conclusions: Accounting for spinal level variation is lacking in existing imaging studies. Future studies should account for this variation for accurate interpretation of the neuroanatomic origin of acquired MR signals., (© 2015 by American Journal of Neuroradiology.)

Objective: Cognitive impairments contribute significantly to inadequate functional recovery following illness episodes in bipolar disorder, yet data on treatment interventions are sparse. We assessed the cognitive effects of a standardized extract of the medicinal herb Withania somnifera (WSE) in bipolar disorder., Method: Sixty euthymic subjects with DSM-IV bipolar disorder were enrolled in an 8-week, double-blind, placebo-controlled, randomized study of WSE (500 mg/d) as a procognitive agent added adjunctively to the medications being used as maintenance treatment for bipolar disorder. Study enrollment and data analyses were completed between December 2008 and September 2012. Cognitive testing at baseline and 8 weeks assessed primary efficacy outcomes. Psychopathology and adverse events were monitored at scheduled visits., Results: Fifty-three patients completed the study (WSE, n = 24; placebo, n = 29), and the 2 groups were matched in terms of demographic, illness, and treatment characteristics. Compared to placebo, WSE provided significant benefits for 3 cognitive tasks: digit span backward (P = .035), Flanker neutral response time (P = .033), and the social cognition response rating of the Penn Emotional Acuity Test (P = .045). The size of the WSE treatment effect for digit span backward was in the medium range (Cohen d = 0.51; 95% CI, 0.25-0.77). None of the other cognitive tasks showed significant between-group differences. Mood and anxiety scale scores remained stable, and adverse events were minor., Conclusions: Although results are preliminary, WSE appears to improve auditory-verbal working memory (digit span backward), a measure of reaction time, and a measure of social cognition in bipolar disorder. Given the paucity of data for improving cognitive capacity in bipolar disorder, WSE offers promise, appears to have a benign side-effects profile, and merits further study., Trial Registration: ClinicalTrials.gov identifier: NCT00761761., (© Copyright 2013 Physicians Postgraduate Press, Inc.)

Background: Targeting glutamatergic dysfunction provides an exciting opportunity to improve cognitive impairment in schizophrenia. One treatment approach has targeted inadequate antioxidant defenses at glutamatergic synapses. Animal and human data suggest NMDA antagonists worsen executive cognitive controls--e.g. increase perseverative responses and impair set-shifting. We conducted a preliminary study to test the hypothesis that L-carnosine, an antioxidant and anti-glycation agent which is co-localized and released with glutamate would improve executive dysfunction, a cognitive domain associated with glutamate., Methods: Seventy-five symptomatically stable adults with chronic schizophrenia were randomly assigned to L-carnosine as adjunctive treatment (2 g/day) or a matched placebo in a double-blind manner for 3 months. Cognitive domains (executive dysfunction, memory, attention and motor speed) were assessed using a computerized battery at baseline, 4 and 12 weeks, along with psychopathology ratings and safety parameters., Results: The L-carnosine group performed significantly faster on non-reversal condition trials of the set-shifting test compared with placebo but reversal reaction times and errors were not significantly different between treatments. On the strategic target detection test, the L-carnosine group displayed significantly improved strategic efficiency and made fewer perseverative errors compared with placebo. Other cognitive tests showed no significant differences between treatments. Psychopathology scores remained stable. The carnosine group reported more adverse events (30%) compared with the placebo group (14%). Laboratory indices remained within acceptable ranges., Conclusions: These preliminary findings suggest that L-carnosine merits further consideration as adjunctive treatment to improve executive dysfunction in persons with schizophrenia., (Copyright © 2012 Elsevier B.V. All rights reserved.)

Background/aims: The aim of this single-center, open-label study was to assess the absolute bioavailability of an oral (tablet) versus intravenous formulation of almorexant in healthy subjects., Methods: A pilot phase in 3 healthy male subjects, which preceded the main study, consisted of a single 30-min intravenous infusion of 10 mg almorexant. Its objectives were to ensure the tolerability of the intravenous formulation and to select the intravenous dose for the main study. The main study was a randomized, two-way crossover study in 20 healthy subjects (10 males and 10 females). Subjects received a single oral dose of 200 mg almorexant and a single 30-min intravenous infusion of 20 mg almorexant., Results: All 23 subjects completed the study as planned. Almorexant was well tolerated; the main observed adverse events were somnolence and fatigue. A geometric mean total body clearance of 43 l/h (95% CI 39-47) and a volume of distribution of 683 liters (95% CI 552-845) were determined. The absolute oral bioavailability of almorexant was 11.2% (90% CI 9.6-13.1)., Conclusion: Almorexant seems to possess a pronounced first-pass effect and metabolism., (Copyright © 2012 S. Karger AG, Basel.)

Stereoscopic depth perception is based on binocular disparities. Although neurons in primary visual cortex (V1) are selective for binocular disparity, their responses do not explicitly code perceived depth. The stereoscopic pathway must therefore include additional processing beyond V1. We used functional magnetic resonance imaging (fMRI) to examine stereo processing in V1 and other areas of visual cortex. We created stereoscopic stimuli that portrayed two planes of dots in depth, placed symmetrically about the plane of fixation, or else asymmetrically with both planes either nearer or farther than fixation. The interplane disparity was varied parametrically to determine the stereoacuity threshold (the smallest detectable disparity) and the upper depth limit (largest detectable disparity). fMRI was then used to quantify cortical activity across the entire range of detectable interplane disparities. Measured cortical activity covaried with psychophysical measures of stereoscopic depth perception. Activity increased as the interplane disparity increased above the stereoacuity threshold and dropped as interplane disparity approached the upper depth limit. From the fMRI data and an assumption that V1 encodes absolute retinal disparity, we predicted that the mean response of V1 neurons should be a bimodal function of disparity. A post hoc analysis of electrophysiological recordings of single neurons in macaques revealed that, although the average firing rate was a bimodal function of disparity (as predicted), the precise shape of the function cannot fully explain the fMRI data. Although there was widespread activity within the extrastriate cortex (consistent with electrophysiological recordings of single neurons), area V3A showed remarkable sensitivity to stereoscopic stimuli, suggesting that neurons in V3A may play a special role in the stereo pathway.

Many previous studies have examined the ease with which two spatially adjacent textures can be segmented. Our goal is to examine the representational system that determines the appearance of isolated patches of visual texture. To this end, similarity judgments from three subjects were obtained for 20 artificial textures comprising filtered noise. Multidimensional scaling (MDS) revealed that three perceptual dimensions explain most of the variance in subjects' similarity judgments. In addition, the three subjects' similarity judgments and MDS solutions were highly correlated. A computational model utilizing the energy responses in seven bandpass filters explains an average of 80% of the variability in the original similarity scores of individual subjects. In the model, energy responses are mapped to the perceptual space through a linear transformation that can be decomposed into two components. The first component decorrelates initial filter responses and the second component maps the decorrelated filter responses to a perceptual space. These latter transformations show remarkable agreement between the three subjects.

Previous research has suggested that the processing of binocular disparity in complex cells may be described with an energy formalism. The energy formalism allows for a representation of disparity by differences in the position or in the phase of monocular receptive subfields of binocular cells, or by combination of these two types. We studied the coding of disparities with an approach complementary to previous algorithmic investigations. Since realization of these representations is probably not genetically determined but learned during ontogeny, we used backpropagation networks to study which of these three possibilities were realized within neural nets. Three types of networks were trained with noise patterns in analogy to the three types of energy models. The networks learned the task and generalized to untrained correlated noise pattern input. Outputs were broadly tuned to spatial frequency and did not respond to anti-correlated noise patterns. Although the energy model was not explicitly implemented, we could analyze the outputs of the networks using predictions of the energy formalism. After learning was completed, the model neurons preferred position shifts over phase shifts in representing disparity. We discuss the general meaning of these findings and the correspondences and deviations between the energy model, V1 neurons, and our networks.

We report two experiments concerning the site of the principal nonlinearity in second-order stereopsis. The first exploits the asymmetry in perceiving transparency with second-order stimuli found by Langley et al. (1998) (Proceedings of the Royal Society of London B, 265, 1837-1845) i.e. the product of a positive-valued contrast envelope and a mean-zero carrier grating can be seen transparently only when the disparities are consistent with the envelope appearing in front of the carrier. We measured the energy at the envelope frequencies that must be added in order to negate this asymmetry. We report that this amplitude can be predicted from the envelope sidebands and not from the magnitude of compressive pre-cortical nonlinearities measured by other researchers. In the second experiment, contrast threshold elevations were measured for the discrimination of envelope disparities following adaptation to sinusoidal gratings. It is reported that perception of the envelope's depth was affected most when the adapting grating was similar (in orientation and frequency) to the carrier, rather than to the contrast envelope. These results suggest that the principal nonlinearity in second-order stereopsis is cortical, occurring after orientation- and frequency-selective linear filtering.

When the product of a vertical square-wave grating (contrast envelope) and a horizontal sinusoidal grating (carrier) are viewed binocularly with different disparity cues they can be perceived transparently at different depths. We found, however, that the transparency was asymmetric; it only occurred when the envelope was perceived to be the overlaying surface. When the same two signals were added, the percept of transparency was symmetrical; either signal could be seen in front of or behind the other at different depths. Differences between these multiplicative and additive signal combinations were examined in two experiments. In one, we measured disparity thresholds for transparency as a function of the spatial frequency of the envelope. In the other, we measured disparity discrimination thresholds. In both experiments the thresholds for the multiplicative condition, unlike the additive condition, showed distinct minima at low envelope frequencies. The different sensitivity curves found for multiplicative and additive signal combinations suggest that different processes mediated the disparity signal. The data are consistent with a two-channel model of binocular matching, with multiple depth cues represented at single retinal locations.

First- and second-order motions differ in their ability to induce motion aftereffects (MAEs) and the kinetic depth effect (KDE). To test whether second-order stimuli support computations relating to motion-in-depth we examined the vection illusion (illusory self motion induced by image flow) using a vection stimulus (V, expanding concentric rings) that depicted a linear path through a circular tunnel. The set of vection stimuli contained differing amounts of first- and second-order motion energy (ME). Subjects reported the duration of the perceived MAEs and the duration of their vection percept. In Experiment 1 both MAEs and vection durations were longest when the first-order (Fourier) components of V were present in the stimulus. In Experiment 2, V was multiplicatively combined with static noise carriers having different check sizes. The amount of first-order ME associated with V increases with check size. MAEs were found to increase with check size but vection durations were unaffected. In general MAEs depend on the amount of first-order ME present in the signal. Vection, on the other hand, appears to depend on a representation of image flow that combines first- and second-order ME.

Background: Nonlinearities play a significant role in early visual processing. They are central to the perception of spatial contrast variations, multiplicative transparencies and texture boundaries. This article concerns the stage of processing at which nonlinearities first become significant., Results: Subjects were adapted to a high contrast sinusoidal grating followed by a brief presentation of a contrast modulated test (plaid) pattern. Thresholds for the detection of the contrast modulation (the beat) were measured. Results show that threshold elevation is greatest when the orientation and spatial frequency of the adapting grating are close to the principal Fourier frequency (the carrier) of the test pattern. Adaptation to sinewave-gratings near the frequency of the contrast modulation has relatively little effect. The data also show that the processing of contrast is frequency selective, with a peak tuning frequency near 0.4 cycles per degree., Conclusions: The data are consistent with a model in which the contrast beats are processed in a frequency-specific manner, after an initial stage of frequency-specific and orientation-specific linear filtering.

Neurophysiological data support two models for the disparity selectivity of binocular simple and complex cells in primary visual cortex. These involve binocular combinations of monocular receptive fields that are shifted in retinal position (the position-shift model) or in phase (the phase-shift model) between the two eyes. This article presents a formal description and analysis of a binocular energy model with these forms of disparity selectivity. We propose how one might measure the relative contributions of phase and position shifts in simple and complex cells. The analysis also reveals ambiguities in disparity encoding that are inherent in these model neurons, suggesting a need for a second stage of processing. We propose that linear pooling of the binocular responses across orientations and scales (spatial frequency) is capable of producing an unambiguous representation of disparity.

Non-Fourier motion is now commonplace in research on visual motion perception, yet lacks a computational framework. This paper examines this issue based on the observation that many non-Fourier motion stimuli have a simple characterization in the frequency domain, in terms of oriented power distributions that lie along lines (or planes) that do not pass through the origin. This provides a unifying theoretical framework for a very diverse class of non-Fourier phenomena. It also allows us to examine some central issues concerning the computational nature of non-Fourier models, and naturally occurring sources of non-Fourier motion. For example, it is shown that the orientation of power in frequency domain corresponds to the velocity of a multiplicative envelope, and may arise as a restricted form of lighting effects, translucency or occlusion. We also show that both the location and orientation of spectral power may be extracted from the phase and amplitude output of band-pass filters, consonant with existing non-Fourier models.

We examine the implications of significant inseparable behaviour in centre-surround retinal cell types. From the form of a spatiotemporal centre-surround (CS) model which agrees qualitatively with physiological observations, we find that the sustained/transient dichotomy is a poor distinction for X-type/Y-type retinal ganglion cells since both exhibit inseparability. Static centre-surround models and spatiotemporal separable models are not valid for time-varying stimuli. Our results contradict the models for X- and Y-type ganglion cells proposed by Marr and Hildreth (1980) and Marr and Ullman (1981), and raise doubts about the physiological validity of Marr's zero-crossing theory. The CS filter is an attractive precursor to the extraction of 2-d motion information.