Latent Transformer Models for out-of-distribution detection.

Any clinically-deployed image-processing pipeline must be robust to the full range of inputs it may be presented with. One popular approach to this challenge is to develop predictive models that can provide a measure of their uncertainty. Another approach is to use generative modelling to quantify the likelihood of inputs. Inputs with a low enough likelihood are deemed to be out-of-distribution and are not presented to the downstream predictive model. In this work, we evaluate several approaches to segmentation with uncertainty for the task of segmenting bleeds in 3D CT of the head. We show that these models can fail catastrophically when operating in the far out-of-distribution domain, often providing predictions that are both highly confident and wrong. We propose to instead perform out-of-distribution detection using the Latent Transformer Model: a VQ-GAN is used to provide a highly compressed latent representation of the input volume, and a transformer is then used to estimate the likelihood of this compressed representation of the input. We demonstrate this approach can identify images that are both far- and near- out-of-distribution, as well as provide spatial maps that highlight the regions considered to be out-of-distribution. Furthermore, we find a strong relationship between an image's likelihood and the quality of a model's segmentation on it, demonstrating that this approach is viable for filtering out unsuitable images.

A novel case of MSTO1 gene related congenital muscular dystrophy with progressive neurological involvement.

Recessive mutations in the MSTO1 gene, encoding for a mitochondrial distribution and morphology regulator, have been recently described in a very limited number of patients with multisystem involvement, mostly characterized by myopathy or dystrophy, cerebellar ataxia, pigmentary retinopathy and raised creatine kinase levels. Here we report an additional patient with recessive MSTO1-related muscular dystrophy (MSTO1-RD), and clinical and radiological evidence of progressive cerebellar involvement. Whole-exome sequencing identified two novel MSTO1 missense variants, c.766C > T (p. (Arg256Trp) and c.1435C > T (p. (Pro479Ser), predicted as damaging by in silico tools. We also report a distinct pattern of selective involvement on muscle MRI in MSTO1-RD. This case confirms a consistent MSTO1-related neuromuscular phenotype and in addition suggests a progressive neurological component at least in some patients, in keeping with the mitochondrial role of the defective protein.

Giant thoracic discs: treatment, outcome, and follow-up of 33 patients in a single centre.

PURPOSE: To help guide treatment strategies and create insight into functional outcomes in patients with Giant herniated thoracic discs (GHTD), which are defined as occupying more than 40% of spinal canal. METHODS: Authors did a retrospective analysis of prospective cohort of 33 cases of GHTD, using clinical letters, notes, and telephonic questionnaires to determine their pre and postoperative functional status, surgical details, and complication rates. 16 males and 17 females operated between 2006 and 2014 were included in the study. A total of 23 patients underwent thoracotomy, 9 costotransversectomy, and 2 transpedicular approaches for excision of thoracic discs. Neuromonitoring was used in seven patients. RESULTS: Frankel grade improved by 1 in 13 patients and by 2 in 1 patient. One patient of T11/12 GHTD with neurogenic claudication recovered completely, taking the overall improvement rate to 45.5%. It remained static in 15 patients (45.5%) and deteriorated by 1 in 3 patients (9%). By mJOA scoring too, the outcome was favorable in majority (84.4%) of patients. There were three intraoperative complications (9%), which included two incidental durotomies and one massive blood loss. Late postoperative complications were seen in 12(39%) patients. They included intercostal neuralgia, mechanical pain around costotransversectomy, progressive thoracic paraplegia due to spinal cord herniation and residual disc fragments, reactive pleural effusion, CSF fistula induced pleural effusion, and infective discitis. CONCLUSIONS: Surgery for giant herniated thoracic disc has favorable outcome in majority (91%) of patients. However, significant approach-related complications are to be anticipated in patients undergoing thoracotomies, most of them being manageable. Author recommends costotransversectomy, only in paracentral discs with smaller percentage canal stenosis.

Delayed diagnosis of Pendred syndrome.

We describe a case of a delayed diagnosis of Pendred syndrome. The patient had a history of hearing loss from childhood and presented to the endocrine clinic when aged 32 with an enlarging goitre. The characteristic Mondini deformity was noted on auditory canal MRI scan. Genetic tests confirmed the diagnosis. Pendred syndrome is an autosomal recessive disorder, characterised by congenital sensorineural hearing loss, goitre and impaired iodide organification and can present at any age. We highlight to the general physician the classical features of this syndrome that would aid early diagnosis.

Globus Pallidus Internus Deep Brain Stimulation for Traumatic Hemidystonia Following Penetrating Head Injury.

BACKGROUND: Deep brain stimulation (DBS) has been a major advance in the treatment of dystonias. Outcomes are, however, less predictable for secondary dystonias, predominantly due to progression of disease or specific brain lesions. There are few cases reported of globus pallidus internus (GPi)-DBS for posttraumatic dystonia. We describe the successful use of unilateral GPi-DBS in a patient with hemidystonia following penetrating head injury. To our knowledge, this is the first description of the use of DBS following penetrating head injury. CASE DESCRIPTION: We present the case of a 47-year-old man with phasic hemidystonia. At the age of 3 years he suffered a penetrating head injury from a welding needle. The patient developed dystonic and phasic right-sided movements. Preoperative Burke-Fahn-Marsden score was 26. Magnetic resonance imaging showed a linear encephalomalacic track extending from the cortex in the left parieto-occipital region, traversing just superolateral to the left trigone into the left thalamus and ending in the region of left cerebral peduncle and subthalamic nucleus. There was no left GPi lesion. A left GPi-DBS electrode was inserted. At 6 months' follow-up, the patient's arm was more relaxed and his spasms lessened in their severity and frequency. Although the Burke-Fahn-Marsden score of 21 had improved modestly by 20%, pain and comfort levels had significantly improved with 50% improvement in visual analog scale score, translating in better quality of life. There were no complications. The clinical benefit persists at 5 years post surgery. CONCLUSION: Selected patients with posttraumatic hemidystonia, including following penetrating head injury, represent one group of secondary dystonias that might benefit from DBS surgery.

EPG5-related Vici syndrome: a paradigm of neurodevelopmental disorders with defective autophagy.

Vici syndrome is a progressive neurodevelopmental multisystem disorder due to recessive mutations in the key autophagy gene EPG5. We report genetic, clinical, neuroradiological, and neuropathological features of 50 children from 30 families, as well as the neuronal phenotype of EPG5 knock-down in Drosophila melanogaster. We identified 39 different EPG5 mutations, most of them truncating and predicted to result in reduced EPG5 protein. Most mutations were private, but three recurrent mutations (p.Met2242Cysfs*5, p.Arg417*, and p.Gln336Arg) indicated possible founder effects. Presentation was mainly neonatal, with marked hypotonia and feeding difficulties. In addition to the five principal features (callosal agenesis, cataracts, hypopigmentation, cardiomyopathy, and immune dysfunction), we identified three equally consistent features (profound developmental delay, progressive microcephaly, and failure to thrive). The manifestation of all eight of these features has a specificity of 97%, and a sensitivity of 89% for the presence of an EPG5 mutation and will allow informed decisions about genetic testing. Clinical progression was relentless and many children died in infancy. Survival analysis demonstrated a median survival time of 24 months (95% confidence interval 0-49 months), with only a 10th of patients surviving to 5 years of age. Survival outcomes were significantly better in patients with compound heterozygous mutations (P = 0.046), as well as in patients with the recurrent p.Gln336Arg mutation. Acquired microcephaly and regression of skills in long-term survivors suggests a neurodegenerative component superimposed on the principal neurodevelopmental defect. Two-thirds of patients had a severe seizure disorder, placing EPG5 within the rapidly expanding group of genes associated with early-onset epileptic encephalopathies. Consistent neuroradiological features comprised structural abnormalities, in particular callosal agenesis and pontine hypoplasia, delayed myelination and, less frequently, thalamic signal intensity changes evolving over time. Typical muscle biopsy features included fibre size variability, central/internal nuclei, abnormal glycogen storage, presence of autophagic vacuoles and secondary mitochondrial abnormalities. Nerve biopsy performed in one case revealed subtotal absence of myelinated axons. Post-mortem examinations in three patients confirmed neurodevelopmental and neurodegenerative features and multisystem involvement. Finally, downregulation of epg5 (CG14299) in Drosophila resulted in autophagic abnormalities and progressive neurodegeneration. We conclude that EPG5-related Vici syndrome defines a novel group of neurodevelopmental disorders that should be considered in patients with suggestive features in whom mitochondrial, glycogen, or lysosomal storage disorders have been excluded. Neurological progression over time indicates an intriguing link between neurodevelopment and neurodegeneration, also supported by neurodegenerative features in epg5-deficient Drosophila, and recent implication of other autophagy regulators in late-onset neurodegenerative disease.

A study on clinical and radiological features and outcome in patients with posterior reversible encephalopathy syndrome (PRES).

UNLABELLED: Posterior reversible encephalopathy syndrome (PRES) is characterized clinically by headaches, seizures, vomiting, nausea, visual abnormalities, and altered mental function and is often (but not invariably) accompanied by parieto-occipital imaging features. The aim of this study is to describe the clinical and radiological features and outcome following PRES in a paediatric cohort. From a retrospectively identified cohort, case records were studied to confirm a diagnosis of PRES. Neuroimaging was reviewed again to assign to recently described radiological subtypes parieto-occipital pattern, holohemispheric watershed pattern, dominant superior frontal sulcus pattern, and asymmetrical or partial expression of the three primary patterns (A/P). Patient outcome was measured by the modified Rankin scale (mRS) scores. Nine boys and three girls with mean age of 12 were identified. Hypertensive episodes (n = 11), tacrolimus toxicity (n = 4), and autoimmunity (n = 1) were identified as potential risk factors/etiologies. Their median mRS at the peak of illness was 2 (range 2-5); three children required intensive care support. After mean follow-up of 35 months (median 37 months; range 3-60 months), all patients improved significantly with mean mRS of 1 (median 1; range 0-1). CONCLUSION: PRES is easily recognizable by the clinical and radiological features. Although severe at presentation, the outcome from this condition is favorable.

Posterior fossa ependymomas: new radiological classification with surgical correlation.

PURPOSE: The key determinant of long-term outcome in infratentorial ependymomas remains the extent of surgical resection. We describe a new radiological classification system which is validated against surgical findings and correlated with risk of post-operative residual tumour. METHODS: Twenty-five consecutive patients (12 females, mean age 4.9 years, range 0.5-17 years) with infratentorial ependymomas were studied. Lesions were classified on pre-operative MRI according to the pattern of extension, brainstem displacement and involvement of the obex, as lateral-type or midfloor-type tumours. Twenty-one operative records were reviewed with respect to the microanatomical tumour origin by a paediatric neurosurgeon, blinded to MRI findings. Follow-up imaging studies were evaluated for residual tumour. RESULTS: There were 15 cases of midfloor-type tumour (anterior displacement of brainstem, infiltration of obex) and 10 cases of lateral-type tumour (lateral displacement of brainstem, obex free of tumour). Extension into prepontine or cerebellopontine cisterns was more common in lateral-type tumours. Agreement between the radiological classification and tumour origin, as defined by operative records, was seen in 18 out of 20 cases. Risk of residual tumour in lateral-type tumours was more than twice that of midfloor-type tumours (80% vs. 33%, p=0.04). Risk of tumour residual was also significantly higher when vessel encasement or prepontine extension was observed. CONCLUSIONS: Infratentorial ependymomas can be pre-operatively classified as lateral-type or midfloor-type tumours. This correlates well with operative findings. Lateral-type tumours have significantly increased risk of residual tumour compared to midfloor- type tumours and this may influence intensity of imaging surveillance.

Characterization of carotid plaque hemorrhage: a CT angiography and MR intraplaque hemorrhage study.

BACKGROUND AND PURPOSE: The main objective of this study was to evaluate CT angiographic (CTA) features that are able to predict the presence of intraplaque hemorrhage (IPH) as defined by MR-IPH. METHODS: One hundred sixty-seven consecutive patients (mean age 69 years, SD 12.8; 58 females) underwent both MR-IPH and CTA within 3 weeks. MR-IPH, the gold standard, was performed at 1.5 T using a neurovascular phased-array coil as a coronal T1-weighted 3-dimensional fat-suppressed acquisition. CTA was performed using a 4-slice or a 64-slice CT machine and evaluated, blinded to MR-IPH findings, for carotid stenosis, plaque density, and plaque ulceration. Plaque density was defined as the mean attenuation of plaque at the site of maximum stenosis and 2 sections above and below. Plaque ulceration was defined as outpouching of contrast into the plaque at least 2 mm deep on any single plane. RESULTS: Prevalence of IPH increased at higher degrees of carotid stenosis. Mean CT plaque density was higher for plaques with MRI-defined IPH (47 Hounsfield units) compared with without IPH (43 Hounsfield units; P=0.02). However, significant overlap between distributions of plaque densities limited the value of mean plaque density for prediction of IPH. CTA plaque ulceration had high sensitivity (80.0% to 91.4%), specificity (93.0% to 92.3%), positive predictive value (72.0% to 71.8%), and negative predictive value (95.0% to 97.9%) for prediction of IPH. Interobserver agreement for presence/absence of CTA plaque ulceration was excellent (kappa=0.80). CONCLUSIONS: CTA plaque ulceration, but not mean CTA plaque density, was useful for prediction of IPH as defined by the MR-IPH technique.

Carotid-artery imaging in the diagnosis and management of patients at risk of stroke.

Carotid atherosclerotic disease is one of the major preventable causes of ischaemic strokes. In clinical practice, decision making with regard to carotid endarterectomy or stenting is still primarily based on the extent of luminal stenosis. In most centres worldwide, luminal stenosis is now mainly assessed with non-invasive techniques, such as Doppler ultrasound, magnetic resonance angiography, and CT angiography, either alone or in combination. Although intra-arterial digital subtraction angiography remains the historical gold standard, it has now mostly been replaced by these non-invasive techniques. Moreover, in addition to luminal stenosis, certain morphological features of carotid plaques, such as large lipid cores, intraplaque haemorrhage, or thin or ruptured fibrous caps, are increasingly believed to be associated with heightened risk of stroke. In this Review, we discuss current state-of-the-art non-invasive diagnostic imaging strategies for luminal stenosis and describe the most promising novel imaging techniques, such as high-resolution MRI and CT combined with PET imaging, which can be used to characterise vulnerable carotid-plaque features in vivo.

The ATHEROMA (Atorvastatin Therapy: Effects on Reduction of Macrophage Activity) Study. Evaluation using ultrasmall superparamagnetic iron oxide-enhanced magnetic resonance imaging in carotid disease.

OBJECTIVES: The aim of this study was to evaluate the effects of low-dose (10 mg) and high-dose (80 mg) atorvastatin on carotid plaque inflammation as determined by ultrasmall superparamagnetic iron oxide (USPIO)-enhanced carotid magnetic resonance imaging (MRI). The hypothesis was that treatment with 80 mg atorvastatin would demonstrate quantifiable changes in USPIO-enhanced MRI-defined inflammation within the first 3 months of therapy. BACKGROUND: Preliminary studies indicate that USPIO-enhanced MRI can identify macrophage infiltration in human carotid atheroma in vivo and hence may be a surrogate marker of plaque inflammation. METHODS: Forty-seven patients with carotid stenosis >40% on duplex ultrasonography and who demonstrated intraplaque accumulation of USPIO on MRI at baseline were randomly assigned in a balanced, double-blind manner to either 10 or 80 mg atorvastatin daily for 12 weeks. Baseline statin therapy was equivalent to 10 mg of atorvastatin or less. The primary end point was change from baseline in signal intensity (DeltaSI) on USPIO-enhanced MRI in carotid plaque at 6 and 12 weeks. RESULTS: Twenty patients completed 12 weeks of treatment in each group. A significant reduction from baseline in USPIO-defined inflammation was observed in the 80-mg group at both 6 weeks (DeltaSI 0.13; p = 0.0003) and at 12 weeks (DeltaSI 0.20; p < 0.0001). No difference was observed with the low-dose regimen. The 80-mg atorvastatin dose significantly reduced total cholesterol by 15% (p = 0.0003) and low-density lipoprotein cholesterol by 29% (p = 0.0001) at 12 weeks. CONCLUSIONS: Aggressive lipid-lowering therapy over a 3-month period is associated with significant reduction in USPIO-defined inflammation. USPIO-enhanced MRI methodology may be a useful imaging biomarker for the screening and assessment of therapeutic response to "anti-inflammatory" interventions in patients with atherosclerotic lesions. (Effects of Atorvastatin on Macrophage Activity and Plaque Inflammation Using Magnetic Resonance Imaging [ATHEROMA]; NCT00368589).

Quality of CT pulmonary angiography for suspected pulmonary embolus in pregnancy.

The main objective of this study was to assess the quality of CT pulmonary angiography (CTPA) for suspected pulmonary embolus (PE) in the pregnant population. We retrospectively identified 40 consecutive pregnant patients who underwent CTPA from January 2005 to December 2006. Forty consecutive age-matched non-pregnant women were used as a control group. Studies were subjectively graded according to overall image quality by two readers in consensus, in randomised and blinded manner. Moreover, contrast enhancement of pulmonary arteries was subjectively and objectively evaluated. The proportion of sub-optimal studies was more than three times higher in the pregnant group (27.5%, n = 11) compared with the non-pregnant group (7.5%, n = 3; p = 0.015). Mean contrast enhancement was consistently higher in the non-pregnant group compared with pregnant group, both subjectively and objectively. The percentage of inadequately opacified vascular segments was more than two times higher in the pregnant group (28.7%, n = 264) than in the non-pregnant group (13.3%, n = 122; p = 0.0001). The incidence of sub-optimal CTPA studies is higher in pregnancy when compared with an age-matched non-pregnant control group. In addition to radiation issues, this should also be considered when implementing diagnostic strategies for suspected PE in pregnancy.

Correlation of carotid atheromatous plaque inflammation using USPIO-enhanced MR imaging with degree of luminal stenosis.

BACKGROUND AND PURPOSE: Inflammation is a recognized risk factor for the vulnerable atherosclerotic plaque. The study explores the relationship between the degree of Magnetic Resonance (MR)-defined inflammation using Ultra Small Super-Paramagnetic Iron Oxide (USPIO) particles and the severity of luminal stenosis in asymptomatic carotid plaques. METHODS: Seventy-one patients with an asymptomatic carotid stenosis of > or = 40% underwent multi-sequence USPIO-enhanced MR imaging. Stenosis severity was measured according to the NASCET and ECST methods. RESULTS: No demonstrable relationship between inflammation as measured by USPIO-enhanced signal change and the degree of luminal stenosis was found. CONCLUSIONS: Inflammation and stenosis are likely to be independent risk factors, although this needs to be further validated.

Comparison of the inflammatory burden of truly asymptomatic carotid atheroma with atherosclerotic plaques contralateral to symptomatic carotid stenosis: an ultra small superparamagnetic iron oxide enhanced magnetic resonance study.

BACKGROUND: Inflammation is a recognised risk factor for the vulnerable atherosclerotic plaque. The aim of this study was to explore whether there is a difference in the degree of magnetic resonance (MR) defined inflammation using ultra small superparamagnetic iron oxide (USPIO) particles within carotid atheroma in completely asymptomatic individuals and the asymptomatic carotid stenosis contralateral to the symptomatic side. METHODS: 20 symptomatic patients with contralateral disease and 20 completely asymptomatic patients underwent multi-sequence MR imaging before and 36 h after USPIO infusion. Images were manually segmented into quadrants and signal change in each quadrant was calculated following USPIO administration. Mean signal change was compared across all quadrants in the two groups. RESULTS: The mean percentage of quadrants showing signal loss was 53% in the contralateral group compared with 31% in completely asymptomatic individuals (p = 0.025). The mean percentages showing enhancement were 44% and 65%, respectively (p = 0.024). The mean signal difference between the two groups was 8.6% (95% CI 1.6% to 15.6%; p = 0.017). CONCLUSIONS: Truly asymptomatic plaques seem to demonstrate inflammation but not to the extent of the contralateral asymptomatic stenosis to the symptomatic side. Inflammatory activity may be a significant risk factor in asymptomatic disease.

Internal carotid artery stenosis: accuracy of subjective visual impression for evaluation with digital subtraction angiography and contrast-enhanced MR angiography.

PURPOSE: To prospectively determine, for both digital subtraction angiography (DSA) and contrast material-enhanced magnetic resonance (MR) angiography, the accuracy of subjective visual impression (SVI) in the evaluation of internal carotid artery (ICA) stenosis, with objective caliper measurements serving as the reference standard. MATERIALS AND METHODS: Local ethics committee approval and written informed patient consent were obtained. A total of 142 symptomatic patients (41 women, 101 men; mean age, 70 years; age range, 44-89 years) suspected of having ICA stenosis on the basis of Doppler ultrasonographic findings underwent both DSA and contrast-enhanced MR angiography. With each modality, three independent neuroradiologists who were blinded to other test results first visually estimated and subsequently objectively measured stenoses. Diagnostic accuracy and percentage misclassification for correct categorization of 70%-99% stenosis were calculated for SVI, with objective measurements serving as the reference standard. Interobserver variability was determined with kappa statistics. RESULTS: After exclusion of arteries that were unsuitable for measurement, 180 vessels remained for analysis with DSA and 159 vessels remained for analysis with contrast-enhanced MR angiography. With respect to 70%-99% stenosis, SVI was associated with average misclassification of 8.9% for DSA (8.9%, 7.8%, and 10.0% for readers A, B, and C, respectively) and of 11.7% for contrast-enhanced MR angiography (11.3%, 8.8%, and 15.1% for readers A, B, and C, respectively). Negative predictive values were excellent (92.3%-100%). Interobserver variability was higher for SVI (DSA, kappa = 0.62-0.71; contrast-enhanced MR angiography, kappa = 0.57-0.69) than for objective measurements (DSA, kappa = 0.75-0.80; contrast-enhanced MR angiography, kappa = 0.66-0.72). CONCLUSION: SVI alone is not recommended for evaluation of ICA stenosis with both DSA and contrast-enhanced MR angiography. SVI may be acceptable as an initial screening tool to exclude the presence of 70%-99% stenosis, but caliper measurements are warranted to confirm the presence of such stenosis.

Non-invasive MR imaging of inflammation in a patient with both asymptomatic carotid atheroma and an abdominal aortic aneurysm: a case report.

Inflammation is a recognized risk factor for the vulnerable atherosclerotic plaque. USPIO-enhanced MRI imaging is a promising non-invasive method to identify high-risk atheromatous plaque inflammation in vivo in humans, in which areas of focal signal loss on MR images have been shown to correspond to the location of activated macrophages, typically at the shoulder regions of the plaque. This is the first report in humans describing simultaneous USPIO uptake within atheroma in two different arterial territories and again emphasises that atherosclerosis is a truly systemic disease. With further work, USPIO-enhanced MR imaging may be useful in identifying inflamed vulnerable atheromatous plaques in vivo, so refining patient selection for intervention and allowing appropriate early aggressive pharmacotherapy to prevent plaque rupture.

Imaging the cellular biology of the carotid plaque.

Carotid atherosclerotic disease is a significant preventable cause of stroke. Clinical decision-making in current practice is based primarily on detection of the severity of luminal stenosis, as determined by ultrasound or conventional angiographic imaging modalities. New insights in the biology of atherosclerosis now suggests that the morphological characteristics of the carotid plaque as well as the molecular and cellular processes occurring within it may be more important markers of plaque vulnerability and stroke risk. This review summarizes emerging applications in the molecular imaging of atherosclerosis and detection of the vulnerable carotid plaque. We discuss how advances in imaging platforms and biochemical technology (e.g. targeted contrast agents) have driven some exciting and promising novel diagnostic imaging approaches from bench to bedside.

Identifying inflamed carotid plaques using in vivo USPIO-enhanced MR imaging to label plaque macrophages.

BACKGROUND: Inflammation within atherosclerotic lesions contributes to plaque instability and vulnerability to rupture. We set out to evaluate the use of a macrophage labeling agent to identify carotid plaque inflammation by in vivo magnetic resonance imaging (MRI). METHODS AND RESULTS: Thirty patients with symptomatic severe carotid stenosis scheduled for carotid endarterectomy underwent multi-sequence MRI of the carotid bifurcation before and after injection of ultrasmall superparamagnetic particles of iron oxide (USPIOs). USPIO particles accumulated in macrophages in 24 of 30 plaques (80%). Areas of signal intensity reduction, corresponding to USPIO/macrophage-positive histological sections, were visualized in 24 of 27 (89%) patients, with an average reduction in signal intensity induced by the USPIO particles of 24% (range, 3.1% to 60.8%). CONCLUSIONS: USPIO-enhanced MRI can identify plaque inflammation in vivo by accumulation of USPIO within macrophages in carotid plaques.