Anomaly Detection (AD) in images is a fundamental computer vision problem and refers to identifying images and/or image substructures that deviate significantly from the norm. Popular AD algorithms ...commonly try to learn a model of normality from scratch using task specific datasets, but are limited to semi-supervised approaches employing mostly normal data due to the inaccessibility of anomalies on a large scale combined with the ambiguous nature of anomaly appearance. We follow an alternative approach and demonstrate that deep feature representations learned by discriminative models on large natural image datasets are well suited to describe normality and detect even subtle anomalies in a transfer learning setting. Our model of normality is established by fitting a multivariate Gaussian (MVG) to deep feature representations of classification networks trained on ImageNet using normal data only. By subsequently applying the Mahalanobis distance as the anomaly score we outperform the current state of the art on the public MVTec AD dataset, achieving an Area Under the Receiver Operating Characteristic curve of 95.8 ± 1.2% (mean ± SEM) over all 15 classes. We further investigate why the learned representations are discriminative to the AD task using Principal Component Analysis. We find that the principal components containing little variance in normal data are the ones crucial for discriminating between normal and anomalous instances. This gives a possible explanation to the often subpar performance of AD approaches trained from scratch using normal data only. By selectively fitting a MVG to these most relevant components only, we are able to further reduce model complexity while retaining AD performance. We also investigate setting the working point by selecting acceptable False Positive Rate thresholds based on the MVG assumption. Code is publicly available at https://github.com/ORippler/gaussian-ad-mvtec.
Anomaly detection (AD) in images is a fundamental computer vision problem and refers to identifying images that deviate significantly from normality. State-of-the-art AD algorithms commonly learn a ...model of normality from scratch using task-specific datasets in either semisupervised or self-supervised manner. We follow an alternative approach and model the distribution of normal data in deep feature representations learned from ImageNet via a multivariate Gaussian (MVG). This lightweight approach achieves a new state of the art in AD on the public MVTec AD dataset. In addition to the empirical benefits, we give a clear motivation for the seemingly simplistic approach via the ties between deep generative and discriminative modeling revealed recently. We further elucidate why ImageNet representations are discriminative in the transfer learning AD setting using the principal component analysis. Here, we find that the principal components containing little variance in normal data are the ones crucial for discriminating between normal and anomalous instances, giving an explanation for the unreasonable effectiveness of our approach. We also investigate setting the working point of our approach by selecting acceptable false-positive rate thresholds based on the MVG assumption and the resistance of our approach to unlabeled anomalies in the dataset. Finally, we investigate whether our approach is prone to exploiting spurious correlations using explainable AI techniques. Code is publicly available at https://github.com/ORippler/gaussian-ad-mvtec .
To investigate brain mechanisms whereby electrical stimulation of the motor cortex (MCS) may induce pain relief in patients with neuropathic pain, cerebral blood flow (CBF) changes were studied using ...H
2O PET in 19 consecutive patients treated with MCS for refractory neuropathic pain. Patients were studied in three conditions, (a) before MCS (Baseline, stimulator stopped 4 weeks before), (b) during a 35-min period of MCS and (c) during a 75-min period after MCS had been discontinued (OFF). Compared to Baseline, turning on the stimulator was associated with CBF increase in the contralateral (anterior) midcingulate cortex (aMCC, BA24 and 32) and in the dorso-lateral prefrontal (BA10) cortices. The most important changes of CBF were observed in the 75 min after discontinuation of MCS (OFF). This post-stimulation period was associated with CBF increases in a large set of cortical and subcortical regions (from posterior MCC (pMCC) to pregenual (pg) ACC, orbitofrontal cortex, putamen, thalami, posterior cingulate and prefrontal areas) and in the brainstem (mesencephalon/periaqueductal grey (PAG) and pons). CBF changes in the post-stimulation period correlated with pain relief. Functional connectivity analysis showed significant correlation between pgACC and PAG, basal ganglia, and lower pons activities, supporting the activation of descending ACC-to-PAG connections. MCS may act in part through descending (top-down) inhibitory controls that involve prefrontal, orbitofrontal and ACC as well as basal ganglia, thalamus and brainstem. These hemodynamic changes are lengthened and might therefore underlie the long-lasting clinical effects that largely outlast the actual stimulation periods.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UL, UM, UPCLJ, UPUK
Brain 11C-diprenorphine positron emission tomography scans help clinicians to select the best candidates for motor cortex stimulation among patients with severe refractory neuropathic pain.
The ...clinical effects of motor cortex stimulation (MCS) for neuropathic pain (NP) is thought to be mediated primarily by the secretion of endogenous opioids in humans and in animal models. Because opioid receptor density is itself decreased in patients with NP, we investigated whether the magnitude and distribution of the remaining opioid receptors in patients with NP could be biological predictors of the pain-relieving effects of MCS. Using 11C-diprenorphine positron emission tomography scans, opioid receptor availability was assessed in 15 patients suffering refractory NP, who subsequently received chronically implanted MCS. All patients underwent 2 preoperative baseline scans at 2-wk intervals and were clinically assessed after 7mo of chronic MCS. The levels of preoperative opioid-binding in the insula, thalamus, periaqueductal gray, anterior cingulate, and orbitofrontal cortex were significantly and positively correlated with postoperative pain relief at 7mo. Patients with receptor density values below the lower limits in age-matched controls in the thalamus, periaqueductal gray and contralateral insula were the least likely to benefit from MCS. Opioid-receptor availability as shown in preoperative positron emission tomography scans appears to be related to the efficacy of MCS in NP and may help clinicians to select the candidates most likely to benefit from this procedure.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, OILJ, SBJE, UL, UPUK
A 63-year-old man with a history of motorcycle accident 42 years ago suffered a left brachial plexus avulsion (BPA). Neuropathic pain in his left upper limb was felt in the C6-C7-C8 dermatomes. The ...rationale for performing “DREZotomy” is to preferentially interrupt the nociceptive inputs in the lateral part of the dorsal root entry zone (DREZ).1-3 For pain with complete deafferentation, as observed in BPA, the aim is to destroy the hyperactive nociceptive neurons deep in the apex of the dorsal horn (DH).4 Surgery is performed under general anesthesia, with the patient in prone position. Once the dura mater is opened, the arachnoid needs extensive dissection to open the dorsolateral and lateral spinal cisterns.5 In cases of BPA, the dorsolateral sulcus may be difficult to identify. Three anatomic elements can facilitate its recognition. Firstly, the remaining intact rootlets (above and below the avulsed segments) allow us to roughly localize the dorsolateral sulcus by joining these cranial and caudal normal rootlets. Secondly, blood vessels running on the spinal cord penetrate into the dorsolateral sulcus and often delineate the sulcus. Thirdly, scarring can be seen along the sulcus with small holes where the rootlets used to penetrate the cord. DREZotomy is performed using a graduated sharp bipolar instrument to allow precise microcoagulations of the DH. Preoperative surgical planning helps the surgeon by giving the angle between the DH and median plane.6 In the immediate postoperative period, the patient described the complete disappearance of neuropathic pain in his left upper limb, persistent at last follow-up (1 year) (Video 1).
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UILJ, UL, UM, UPCLJ, UPUK, ZAGLJ, ZRSKP
Purpose
The anatomy of both the brain and the skull is particularly difficult to learn and to teach. Since their anatomical structures are numerous and gathered in a complex tridimensional (3D) ...architecture, classic schematical drawing or photography in two dimensions (2D) has difficulties in providing a clear, simple, and accurate message. Advances in photography and computer sciences have led to develop stereoscopic 3D visualization, firstly for entertainment then for education. In the present study, we report our experience of stereoscopic 3D lecture for neuroanatomy teaching to early medical school students.
Methods
High-resolution specific pictures were taken on various specimen dissections in the Anatomy Laboratory of the University of Lyon, France. Selected stereoscopic 3D views were displayed on a large dedicated screen using a doubled video projector. A 2-h stereoscopic neuroanatomy lecture was given by two neuroanatomists to third-year medicine students who wore passive 3D glasses. Setting up lasted 30 min and involved four people. The feedback from students was collected and analyzed.
Results
Among the 483 students who have attended the stereoscopic 3D lecture, 195 gave feedback, and all (100%) were satisfied. Among these, 190 (97.5%) reported a better knowledge transfer of brain anatomy and its 3D architecture. Furthermore, 167 (86.1%) students felt it could change their further clinical practice, 179 (91.8%) thought it could enhance their results in forthcoming anatomy examinations, and 150 (76.9%) believed such a 3D lecture might allow them to become better physicians. This 3D anatomy lecture was graded 8.9/10 a mean against 5.9/10 for previous classical 2D lectures.
Discussion–conclusion
The stereoscopic 3D teaching of neuroanatomy made medical students enthusiastic involving digital technologies. It could improve their anatomical knowledge and test scores, as well as their clinical competences. Depending on university means and the commitment of teachers, this new tool should be extended to other anatomical fields. However, its setting up requires resources from faculties and its impact on clinical competencies needs to be objectively assessed.
The aim of this paper was to identify and characterize all the segmental radiculomedullary arteries (RMAs) that supply the thoracic and lumbar spinal cord.
All RMAs from T4 to L5 were studied ...systematically in 25 cadaveric specimens. The RMA with the greatest diameter in each specimen was termed the artery of Adamkiewicz (AKA). Other supporting RMAs were also identified and characterized.
A total of 27 AKAs were found in 25 specimens. Twenty-two AKAs (81%) originated from a left thoracic or a left lumbar radicular branch, and 5 (19%) arose from the right. Two specimens (8%) had two AKAs each: one specimen with two AKAs on the left side and the other specimen with one AKA on each side. Eight cadaveric specimens (32%) had 10 additional RMAs; among those, a single additional RMA was found in 6 specimens (75%), and 2 additional RMAs were found in each of the remaining 2 specimens (25%). Of those specimens with a single additional RMA, the supporting RMA was ipsilateral to the AKA in 5 specimens (83%) and contralateral in only 1 specimen (17%). The specimens containing 2 additional RMAs were all (100%) ipsilateral to their respective AKAs.
The segmental RMAs supplying the thoracic and lumbar spinal cord can be unilateral, bilateral, or multiple. Multiple AKAs or additional RMAs supplying a single anterior spinal artery are common and should be considered when dealing with the spinal cord at the thoracolumbar level.
OBJECTIVE:To confirm the efficacy and safety of deep brain stimulation (DBS) of the internal part of the globus pallidus in improving severe tardive dyskinesia (TD).
METHODS:Nineteen patients with ...severe pharmacoresistant TD were included. All were assessed at baseline and at 3, 6 (main outcome measure), and 12 months, and in the long term (6–11 years) for 14 patients, after bilateral pallidal DBS, using motor scales (Extrapyramidal Symptoms Rating Scale ESRS, Abnormal Involuntary Movement Scale AIMS), cognitive scales, and a psychiatric assessment. At 6 months, a double-blind ESRS evaluation was performed in the stimulation “on” and stimulation “off” conditions.
RESULTS:At 6 months, all patients had a decrease of more than 40% on the ESRS. The efficacy of the procedure was confirmed by a double-blind evaluation. This improvement was maintained at 12 months (ESRSdecrease of 58% 21%–81%; AIMSdecrease of 50% 7%–77%) and in the long term (ESRSdecrease of 60% 22%–90%; AIMSdecrease of 63% 14%–94%, n = 14). All the subscores of the ESRS (parkinsonism, dystonia, and chorea) and of the AIMS (facial, oral, extremities, and trunk movements) improved. Despite psychiatric comorbidities at baseline, cognitive and psychiatric tolerability of the procedure was excellent. No cognitive decline was observed and mood was improved in most of the patients.
CONCLUSIONS:Pallidal DBS procedure should be considered as a therapeutic option in disabling TD refractory to medical treatment.
CLASSIFICATION OF EVIDENCE:This study provides Class II evidence that in patients with severe pharmacoresistant TD with implanted pallidal leads, the stimulation “on” condition significantly improved ESRS scores compared to the stimulation “off” condition.
Summary Background Cerebral palsy (CP) with dystonia-choreoathetosis is a common cause of disability in children and in adults, and responds poorly to medical treatment. Bilateral pallidal deep brain ...stimulation (BP-DBS) of the globus pallidus internus (GPi) is an effective treatment for primary dystonia, but the effect of this reversible surgical procedure on dystonia-choreoathetosis CP, which is a subtype of secondary dystonia, is unknown. Our aim was to test the effectiveness of BP-DBS in adults with dystonia-choreoathetosis CP. Methods We did a multicentre prospective pilot study of BP-DBS in 13 adults with dystonia-choreoathetosis CP who had no cognitive impairment, little spasticity, and only slight abnormalities of the basal ganglia on MRI. The primary endpoint was change in the severity of dystonia-choreoathetosis after 1 year of neurostimulation, as assessed with the Burke–Fahn–Marsden dystonia rating scale. The accuracy of surgical targeting to the GPi was assessed masked to the results of neurostimulation. Analysis was by intention to treat. Findings The mean Burke–Fahn–Marsden dystonia rating scale movement score improved from 44·2 (SD 21·1) before surgery to 34·7 (21·9) at 1 year post-operatively (p=0·009; mean improvement 24·4 21·1%, 95% CI 11·6–37·1). Functional disability, pain, and mental health-related quality of life were significantly improved. There was no worsening of cognition or mood. Adverse events were related to stimulation (arrest of the stimulator in one patient, and an adjustment to the current intensity in four patients). The optimum therapeutic target was the posterolateroventral region of the GPi. Little improvement was seen when the neurostimulation diffused to adjacent structures (mainly to the globus pallidus externus GPe). Interpretation Bilateral pallidal neurostimulation could be an effective treatment option for patients with dystonia-choreoathetosis CP. However, given the heterogeneity of motor outcomes and the small sample size, results should be interpreted with caution. The optimum placement of the leads seemed to be a crucial, but not exclusive, factor that could affect a good outcome. Funding National PHRC; Cerebral Palsy Foundation: Fondation Motrice/APETREIMC; French INSERM Dystonia National Network; Medtronic.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UL, UM, UPCLJ, UPUK
Thirty-one patients with medically refractory neuropathic pain were included in a prospective evaluation of motor cortex stimulation. The long-term outcome was evaluated using five variables: (a) ...rate (percentage) of pain relief, (b) pain scores as assessed on VAS, (c) postoperative decrease in VAS scores, (d) reduction in analgesic drug intake, (e) a dichotomic (yes/no) response to the question whether the patient would accept, under similar circumstances, to be operated on again. Pain relief was rated as excellent (>70 % pain relief) in 10 % of cases, good (40-69 %) in 42 %, poor (10-39 %) in 35 % and negligible (0-9 %) in 13 %. Intake of analgesic drugs was decreased in 52 % of patients and unchanged in 45 % (unavailable data in 3 %), with complete withdrawal of analgesic drugs in 36 % of patients. Twenty-one patients (70 %) declared themselves favourable to re-intervention if the same beneficial outcome could be guaranteed. Neither preoperative motor status, pain characteristics, type or localisation of lesions, quantitative sensory testing, Somatosensory Evoked Potentials, nor the interval between pain and surgery were found to predict the efficacy of MCS. The level of pain relief, as evaluated in the first month following implantation was a strong predictor of long-term relief (regression analysis, R=0.744; p<0.0001). These results confirm that MCS can be a satisfactory and durable alternative to medical treatments in patients with refractory pain, and suggest that the efficacy of MCS may be predicted in the first month of therapy.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, OILJ, SBJE, UL, UPUK
PDF
