Perfect Density Models Cannot Guarantee Anomaly Detection.

Thanks to the tractability of their likelihood, several deep generative models show promise for seemingly straightforward but important applications like anomaly detection, uncertainty estimation, and active learning. However, the likelihood values empirically attributed to anomalies conflict with the expectations these proposed applications suggest. In this paper, we take a closer look at the behavior of distribution densities through the lens of reparametrization and show that these quantities carry less meaningful information than previously thought, beyond estimation issues or the curse of dimensionality. We conclude that the use of these likelihoods for anomaly detection relies on strong and implicit hypotheses, and highlight the necessity of explicitly formulating these assumptions for reliable anomaly detection.

Estimating GFR using serum beta trace protein: accuracy and validation in kidney transplant and pediatric populations.

The limitations of estimates of glomerular filtration rate (GFR) based only on serum creatinine measurements have spurred an interest in more sensitive markers of GFR. Beta-trace protein (BTP), a low-molecular-weight glycoprotein freely filtered through the glomerular basement membrane and with minimal non-renal elimination, may be such a marker. We have recently derived two GFR estimation equations based on BTP. To validate these equations, we measured BTP and the plasma clearance of (99)mTc-DTPA in 92 adult kidney transplant recipients and 54 pediatric patients with impaired kidney function. GFR was estimated using the serum creatinine-based Modification of Diet in Renal Disease (MDRD) Study equation for adults, the Schwartz and updated Schwartz equations in children, and 4 novel BTP-derived equations (our 2 equations and 2 proposed by Poge). In adults, our BTP-based equations had low median bias and high accuracy such that 89-90% of estimates were within 30% of measured GFR. In children, the median bias of our 2 equations was low and accuracy was high such that 78-83% of estimates were within 30% of measured GFR. These results were an improvement compared to the MDRD and Schwartz equations, both of which had high median bias and reduced accuracy. The updated Schwartz equation also performed well.

Effect of clinical variables and immunosuppression on serum cystatin C and beta-trace protein in kidney transplant recipients.

BACKGROUND: Cystatin C and beta-trace protein (BTP) are low-molecular-weight proteins that have generated interest as alternative endogenous markers of glomerular filtration rate (GFR). Studies examining the effect of demographic, biometric, clinical, and biochemical variables on cystatin C levels have yielded conflicting results, perhaps because of the reliance on inferior methods of GFR determination. The aim of this study is to examine the independent effect of various clinical parameters on serum concentrations of creatinine, cystatin C, and BTP in kidney transplant recipients. STUDY DESIGN: Cross-sectional study. SETTING & PARTICIPANTS: 207 kidney transplant recipients with stable kidney function. PREDICTORS: GFR, age, race, sex, body mass index, albumin level, proteinuria, smoking status, prednisone, and calcineurin inhibitor and mycophenolate mofetil use. OUTCOMES & MEASUREMENTS: Multiple linear regression analysis was used to examine the relationship between predictor variables and cystatin C, BTP, and creatinine levels. GFR was measured by using technetium 99m-radiolabeled diethylenetriaminepentaacetic acid clearance. RESULTS: After adjusting for GFR, cystatin C and BTP levels were significantly lower in women compared with men. Greater albumin concentration was associated with significantly lower cystatin C and BTP concentrations. There was a statistically significant, but clinically small, association between body mass index and cystatin C level, but no association between the other demographic variables or medications analyzed. LIMITATIONS: Predominantly white population; results may not be applicable to other racial groups. CONCLUSION: Important nonrenal factors can influence BTP and cystatin C concentrations and need to be considered when interpreting BTP and cystatin C values in kidney transplant patients.

The ability of ultrasonography, magnetic resonance imaging and bone mineral densitometry to predict the strength of human Achilles' tendons.

OBJECTIVE: To assess the value of ultrasonography (US), magnetic resonance imaging (MRI), and bone mineral densitometry (BMD) in evaluating human Achilles' tendon strength. DESIGN: Cross-sectional observational study. SETTING: Tertiary care hospital. PARTICIPANTS: Ninety-eight Achilles' tendons from 49 consecutive cadavers (26 men and 23 women with a mean age of 66.6 years) undergoing hospital autopsy were assessed. INTERVENTIONS: Not applicable. MAIN OUTCOME MEASURES: Tendon dimensions on US and MRI, and T1-weighted optical density were measured. Areas of hypodensity, hyperdensity, calcification, and heterogeneity were identified on US. The BMD of each calcaneus was recorded. The tendons were mechanically tested to determine peak load at failure. RESULTS: Sixteen patients (32.7%, 27 tendons) had abnormalities in 1 or both tendons on US and/or MRI (17 on US, 17 on MRI). Fifty-seven tendons (58%) ruptured in their midsubstance, at an average peak load of 4722+/-990N. Tendons with and without abnormalities on imaging had similar strengths (P>.05). Calcaneal BMD correlated weakly with peak load at failure (r=.21, P<.05). CONCLUSIONS: The prevalence of Achilles' tendons abnormalities on US or MRI was 32.7% in our study group. Abnormalities on clinical imaging (US or MRI) were not predictive of the load at failure. Therefore, tendons with imaging abnormalities are not necessarily weaker, and one cannot predict the likelihood of rupture based on imaging results. Further, higher-powered studies could explore the ability of BMD to detect minimal clinically important differences and to predict Achilles' tendon weakness.

Mechanical alterations of rabbit Achilles' tendon after immobilization correlate with bone mineral density but not with magnetic resonance or ultrasound imaging.

OBJECTIVE: To assess the usefulness of magnetic resonance imaging (MRI), ultrasound (US) imaging, or bone mineral density (BMD) in predicting the mechanical properties of immobilized rabbit Achilles' tendons. DESIGN: Experimental study. SETTING: Basic university laboratory. ANIMALS: Twenty-eight rabbits. INTERVENTIONS: Twelve rabbits had 1 hindlimb casted for 4 weeks and 10 rabbits were casted for 8 weeks. Contralateral legs and 12 normal hindlimbs served as controls. MAIN OUTCOME MEASURES: Achilles' tendon dimensions on MRI and US, T1- and T2-signal intensities on MRI, classification of abnormalities on MRI and US; BMD of the calcaneus with dual-energy x-ray absorptiometry. Biomechanic measures consisted of peak load, stiffness, and stress. Imaging variables were correlated with biomechanic alterations. RESULTS: Immobilized Achilles' tendons were weaker and showed decreased mechanical stress compared with their contralateral legs and controls (all P<.05). MRI and US revealed larger Achilles' tendons after immobilization. However, neither increased MRI nor US signal abnormality was found. BMD was lower in immobilized calcanei and larger in contralateral legs than controls. Only BMD correlated with both the decreased peak load (R2=.42, P<.05) and stress (R2=.54, P<.05) of immobilized Achilles' tendon. CONCLUSIONS: This study established weakened mechanical properties of immobilized Achilles' tendons. BMD of the calcaneus, but not MRI and US, was predictive of the mechanical alterations in immobilized Achilles' tendons. BMD may be a useful biomarker to monitor disease and recovery in Achilles' tendons.

Deep Independence Network Analysis of Structural Brain Imaging: Application to Schizophrenia.

Linear independent component analysis (ICA) is a standard signal processing technique that has been extensively used on neuroimaging data to detect brain networks with coherent brain activity (functional MRI) or covarying structural patterns (structural MRI). However, its formulation assumes that the measured brain signals are generated by a linear mixture of the underlying brain networks and this assumption limits its ability to detect the inherent nonlinear nature of brain interactions. In this paper, we introduce nonlinear independent component estimation (NICE) to structural MRI data to detect abnormal patterns of gray matter concentration in schizophrenia patients. For this biomedical application, we further addressed the issue of model regularization of nonlinear ICA by performing dimensionality reduction prior to NICE, together with an appropriate control of the complexity of the model and the usage of a proper approximation of the probability distribution functions of the estimated components. We show that our results are consistent with previous findings in the literature, but we also demonstrate that the incorporation of nonlinear associations in the data enables the detection of spatial patterns that are not identified by linear ICA. Specifically, we show networks including basal ganglia, cerebellum and thalamus that show significant differences in patients versus controls, some of which show distinct nonlinear patterns.

Thawing of frozen calcaneus bone specimens has no effect on the bone mineral density using dual energy x-ray absorptiometry: a study in rabbits and humans.

The purpose of this paper is to study whether the deep freezing of bone and later thawing affects the bone mineral density (BMD) measurement. We used 56 calcanei from 28 adult female New Zealand white rabbits and 102 human calcanei from 51 donors post-mortem (27 men and 24 women, age 30-89). Dual energy x-ray absorptiometry evaluated BMD of the frozen specimen and of the thawed specimen. A main analysis compared BMD of each specimen in the frozen and thawed states. The mean BMD of 224 areas of frozen rabbits' calcanei was 0.31 +/- 0.08 g cm(-2) (95% confidence interval (CI) from 0.30 to 0.32 g cm(-2)) while the mean BMD of thawed rabbits' calcanei was 0.31 +/- 0.08 g cm(-2) (95% CI from 0.30 to 0.32 g cm(-2); paired t-test p > 0.01). The mean BMD of 306 areas of frozen human calcaneus was 0.73 +/- 0.22 g cm(-2) (95% CI from 0.70 to 0.76 g cm(-2)) while the mean BMD of thawed human calcaneus was 0.73 +/- 0.22 g cm(-2) (95% CI from 0.70 to 0.76 g cm(-2); p > 0.01). For both the rabbit and the human calcanei, a Bland-Altman analysis showed a mean difference between the BMD in the frozen and thawed states of 0.00 (limits of agreement, rabbit: -0.07 to 0.08 g cm(-2), human: -0.16 to 0.21 g cm(-2)). A high correlation was observed between calcaneus BMD in the frozen and thawed states (r = 0.94, 0.97, 0.92 and 0.99 respectively in all rabbit calcanei, all human calcanei, immobilized rabbit calcanei and osteopenic human calcanei, all p < 0.01). Bone mineral density is not affected by deep freezing and later thawing of the specimen. Therefore, the specimens need not be thawed to obtain valid and precise BMD measurement. These results are relevant to general musculoskeletal as well as osteoporosis research where the specimens undergo multiple tests in series.

Restoration of strength despite low stress and abnormal imaging after Achilles injury.

PURPOSE: To determine the usefulness of clinical imaging in predicting the mechanical properties of rabbit Achilles tendons after acute injury. METHODS: We created a 2 x 7-mm full-thickness central tendon defect in one Achilles tendon of healthy rabbits. Rabbits in groups of 10 were killed immediately and 4 and 8 wk after surgery (n = 30). We then performed magnetic resonance (MR) imaging, ultrasonography (US), bone mineral densitometry (BMD), and mechanical testing to failure using a dual-cryofixation assembly on experimental and contralateral tendons. The main outcome measures included tendon dimensions, optical density (OD) of T1-weighted, proton density (PD), and T2-weighted MR sequences, US focal abnormalities, BMD of the calcaneus, and stress and peak load to failure. RESULTS: On MR imaging and US, all dimensions of the injured tendons after 2 wk and more were greater than those of the contralateral tendons (P < 0.05). The mean T1-weighted OD was greater at 4 wk (256 +/- 53) and 8 wk (184 +/- 24) than immediately after surgery (149 +/- 15). Mechanical stress was markedly lower in the experimental than in the contralateral tendons at both 4 wk (39 +/- 9 vs 77 +/- 16 N x mm(-2)) and 8 wk (58 +/- 6 vs 94 +/- 26 N x mm(-2); P < 0.05). Mean peak load to failure was significantly lower immediately after surgery (332 +/- 128 N) than at 4 and 8 wk (712 +/- 106 and 836 +/- 90 N, respectively). Both high T1-weighted OD (r = -0.73) and PD OD (r = -0.69) correlated with lower mechanical stress (P < 0.05). In the experimental tendons, higher T1-weighted OD correlated with lower peak load (r = -0.46; P < 0.05). CONCLUSIONS: Normal peak loads 4 wk after injury were withstood by an enlarged tendon of lower stress. These findings support progressive physical loading 4 wk after an Achilles tendon injury. T1-weighted OD constituted a marker of tendon mechanical recovery.

Chronic kidney disease stage in renal transplantation classification using cystatin C and creatinine-based equations.

BACKGROUND: Current clinical guidelines recommend that renal transplant recipients (RTRs) be classified into chronic kidney disease (CKD) stage using a creatinine-based estimate of glomerular filtration rate (GFR). However, creatinine-based equations are inaccurate in RTRs leading to frequent CKD stage misclassification. It is not known whether the classification of CKD stage would be improved using a cystatin C-based estimate of GFR. METHODS: We measured (99m)Tc-DTPA GFR, cystatin C and creatinine in 198 stable RTRs. GFR was estimated using cystatin C-based equations (Filler, Le Bricon and Rule) and four creatinine-based equations. We determined the proportion, overall and by CKD stage, that were classified correctly by each equation as compared to the (99m)Tc-DTPA GFR. RESULTS: The Filler equation correctly classified 76% of patients compared to only 65% with the abbreviated modification of diet in renal disease (MDRD) equation and 69% with the Cockcroft-Gault equation. In CKD stages two and four, the Filler equation correctly classified 77% and 60% of patients whereas the abbreviated MDRD equation correctly classified 46% and 93% of patients. The area under the curve by receiver operating curve analysis for overall stage classification was uniformly poor for all equations (0.52-0.56). CONCLUSIONS: The cystatin C-based Filler and Le Bricon GFR estimates classified slightly more patients into the correct CKD stage than the standard creatinine-based equations in stable RTRs although the overall diagnostic accuracies were similar. The differences are modest and prospective studies will be needed to determine if the adoption of these equations for classification would lead to improved recognition of CKD complications or patient care.

A novel equation to estimate glomerular filtration rate using beta-trace protein.

BACKGROUND: Beta-trace protein (BTP) is a low molecular weight glycoprotein that is a more sensitive marker of glomerular filtration rate (GFR) than serum creatinine. The utility of BTP has been limited by the lack of an equation to translate BTP into an estimate of GFR. The objectives of this study were to develop a BTP-based GFR estimation equation. METHODS: We measured BTP and GFR by (99m)technetium-diethylenetriaminepentaacetic acid in 163 stable adult renal transplant recipients. Stepwise multiple regression models were created to predict GFR corrected for body surface area. The following variables were considered for entry into the model: BTP, urea, sex, albumin, creatinine, age, and race. RESULTS: BTP alone accounted for 75.6% of variability in GFR. The model that included all the predictor variables had the largest coefficient of determination (R(2)) at 0.821. The model with only BTP, urea, and sex had only a slightly lower R(2) of 0.81 and yielded the following equation: GFR mL . min(-1) . (1.73 m(2))(-1) = 112.1 x BTP(-0.662) x Urea(-0.280) x (0.88 if female). A 2nd equation (R(2) = 0.79) using creatinine instead of urea was also developed: GFR mL . min(-1) . (1.73 m(2))(-1) = 1.678 x BTP(-0.758) x creatinine(-0.204) x (0.871 if female). CONCLUSIONS: We have shown that BTP can be used in a simple equation to estimate GFR. Further study is needed in other populations to determine accuracy and clinical utility of this equation.

Estimating glomerular filtration rate in kidney transplantation: a comparison between serum creatinine and cystatin C-based methods.

Accurate measurement of GFR is critical for the evaluation of new therapies and the care of renal transplant recipients. Although not accurate in renal transplantation, GFR is often estimated using creatinine-based equations. Cystatin C is a marker of GFR that seems to be more accurate than creatinine. Equations to predict GFR based on the serum cystatin C concentration have been developed, but their accuracy in transplantation is unknown. GFR was estimated using four equations (Filler, Le Bricon, Larsson, and Hoek) that are based on serum cystatin C and seven equations that are based on serum creatinine in 117 adult renal transplant recipients. GFR was measured using radiolabeled diethylenetriaminepentaacetic acid (99mTc-DTPA), and the bias, precision, and accuracy of each equation were determined. The mean (99m)Tc-DTPA GFR was 58 +/- 23 ml/min per 1.73 m(2). The cystatin C-based equations of Filler and Le Bricon had the lowest bias (-1.7 and -3.8 ml/min per 1.73 m2), greatest precision (11.4 and 11.8 ml/min per 1.73 m2), and highest accuracy (87 and 89% within 30% of measured GFR, respectively). The cystatin C equations remained accurate even when the measured GFR was >60 ml/min per 1.73 m2. The creatinine-based equations were not as accurate, with only 53 to 80% of estimates within 30% of measured GFR. Cystatin C-based equations are more accurate at predicting GFR in renal transplant recipients than traditional creatinine-based equations. Further prospective studies with repetitive measurement of cystatin C are needed to determine whether cystatin C-based estimates of GFR will be sufficiently accurate to monitor long-term allograft function.