Supervising the Decoder of Variational Autoencoders to Improve Scientific Utility.

Probabilistic generative models are attractive for scientific modeling because their inferred parameters can be used to generate hypotheses and design experiments. This requires that the learned model provides an accurate representation of the input data and yields a latent space that effectively predicts outcomes relevant to the scientific question. Supervised Variational Autoencoders (SVAEs) have previously been used for this purpose, as a carefully designed decoder can be used as an interpretable generative model of the data, while the supervised objective ensures a predictive latent representation. Unfortunately, the supervised objective forces the encoder to learn a biased approximation to the generative posterior distribution, which renders the generative parameters unreliable when used in scientific models. This issue has remained undetected as reconstruction losses commonly used to evaluate model performance do not detect bias in the encoder. We address this previously-unreported issue by developing a second-order supervision framework (SOS-VAE) that updates the decoder parameters, rather than the encoder, to induce a predictive latent representation. This ensures that the encoder maintains a reliable posterior approximation and the decoder parameters can be effectively interpreted. We extend this technique to allow the user to trade-off the bias in the generative parameters for improved predictive performance, acting as an intermediate option between SVAEs and our new SOS-VAE. We also use this methodology to address missing data issues that often arise when combining recordings from multiple scientific experiments. We demonstrate the effectiveness of these developments using synthetic data and electrophysiological recordings with an emphasis on how our learned representations can be used to design scientific experiments.

Premature Fusion of the Sagittal Suture as an Incidental Radiographic Finding in Young Children.

BACKGROUND: Craniosynostosis typically develops prenatally and creates characteristic changes in craniofacial form. Nevertheless, postnatal forms of craniosynostosis have been described. The purpose of this study was to determine the prevalence of incidentally identified, but temporally premature, cranial suture fusion in normocephalic children. METHODS: Computed tomographic scans obtained from children aged 1 to 5 years evaluated in the authors' emergency department between 2005 and 2016 were reviewed for evidence of craniosynostosis. Patients with prior ventriculoperitoneal shunt, brain or cranial abnormality, or known syndromes were excluded. The presence of craniosynostosis and cranial index was assessed by a panel of three craniofacial surgeons and one pediatric neurosurgeon. Demographic information, fusion type, reason for the computed tomographic scan, and medical history were recorded as covariates. Cranial shape and intracranial volume were calculated using a previously validated automated system. RESULTS: Three hundred thirty-one patients met the inclusion criteria. The mean age was 2.4 ± 1.3 years. Eleven patients (3.3 percent) were found to have a complete (n = 9) or partial (n = 2) fusion of the sagittal suture. All patients had a normal cranial index (0.80; range, 0.72 to 0.87) and a grossly normal head shape. Only two fusions (18.2 percent) were documented by the radiologist. Cranial shape analysis performed in five of the 11 patients showed subtle phenotypic changes along the scaphocephaly spectrum in four patients, with a normal shape in the remaining case. CONCLUSIONS: Sagittal fusion is present in 3.3 percent of otherwise phenotypically normal children aged 1 to 5 years. The clinical significance of this result is unclear, but routine screening of affected patients is paramount. CLINICAL QUESTION/LEVEL OF EVIDENCE: Risk, IV.

Automated Measurement of Intracranial Volume Using Three-Dimensional Photography.

BACKGROUND: Current methods to analyze three-dimensional photography do not quantify intracranial volume, an important metric of development. This study presents the first noninvasive, radiation-free, accurate, and reproducible method to quantify intracranial volume from three-dimensional photography. METHODS: In this retrospective study, cranial bones and head skin were automatically segmented from computed tomographic images of 575 subjects without cranial abnormality (average age, 5 ± 5 years; range, 0 to 16 years). The intracranial volume and the head volume were measured at the cranial vault region, and their relation was modeled by polynomial regression, also accounting for age and sex. Then, the regression model was used to estimate the intracranial volume of 30 independent pediatric patients from their head volume measured using three-dimensional photography. Evaluation was performed by comparing the estimated intracranial volume with the true intracranial volume of these patients computed from paired computed tomographic images; two growth models were used to compensate for the time gap between computed tomographic and three-dimensional photography. RESULTS: The regression model estimated the intracranial volume of the normative population from the head volume calculated from computed tomographic images with an average error of 3.81 ± 3.15 percent (p = 0.93) and a correlation (R) of 0.96. The authors obtained an average error of 4.07 ± 3.01 percent (p = 0.57) in estimating the intracranial volume of the patients from three-dimensional photography using the regression model. CONCLUSION: Three-dimensional photography with image analysis provides measurement of intracranial volume with clinically acceptable accuracy, thus offering a noninvasive, precise, and reproducible method to evaluate normal and abnormal brain development in young children. CLINICAL QUESTION/LEVEL OF EVIDENCE: Diagnostic, V.

Occult Scaphocephaly: A Forme Fruste Phenotype of Sagittal Craniosynostosis.

INTRODUCTION: Latent cranial suture fusions may present with mild or absent phenotypic changes that make the clinical diagnosis challenging. Recent reports describe patients with sagittal synostosis and a normal cranial index (CI), a condition termed normocephalic sagittal craniosynostosis (NSC). The goal of this study is to evaluate the shape and intracranial volume (ICV) in a cohort of NSC patients using quantitative cranial shape analysis (CSA). METHODS: We identified 19 patients (7.5 ±â€Š2.28 years) between 2011 and 2016, who presented to our hospital with NSC. Cranial index and CSA were measured from the computed tomography image. Cranial shape analysis calculates the distances between the patient's cranial shape and its closest normal shape. Intracranial volume was measured and compared to an established age-matched normative database. RESULTS: Cranial index revealed 15 (78.9%) patients within the mesocephalic range and 4 patients (21.1%) in the brachycephalic range. Detailed CSA identified 15 (78.9%) patients with subtle phenotypic changes along the scaphocephalic spectrum (ie, subtle anterior and posterior elongation with inter-parietal narrowing) and 1 patient (5.3%) with isolated overdevelopment on the posterior part of the right parietal bone. Three patients (15.8%) had a CSA close to normal. Mean ICV was 1410.5 ±â€Š192.77cc; most patients (78.9%) fell within ±2 standard deviations. CONCLUSION: Quantitative CSA revealed that most of the patients with NSC had cranial shape abnormalities, consistent with a forme fruste scaphocephaly that could not be otherwise recognized by clinical observation or CI. Given these findings, we propose the term occult scaphocephaly to describe this condition. The associated incidence of intracranial hypertension is unknown.

Detecting Potential Medication Selection Errors During Outpatient Pharmacy Processing of Electronic Prescriptions With the RxNorm Application Programming Interface: Retrospective Observational Cohort Study.

BACKGROUND: Medication errors are pervasive. Electronic prescriptions (e-prescriptions) convey secure and computer-readable prescriptions from clinics to outpatient pharmacies for dispensing. Once received, pharmacy staff perform a transcription task to select the medications needed to process e-prescriptions within their dispensing software. Later, pharmacists manually double-check medications selected to fulfill e-prescriptions before dispensing to the patient. Although pharmacist double-checks are mostly effective for catching medication selection mistakes, the cognitive process of medication selection in the computer is still prone to error because of heavy workload, inattention, and fatigue. Leveraging health information technology to identify and recover from medication selection errors can improve patient safety. OBJECTIVE: This study aimed to determine the performance of an automated double-check of pharmacy prescription records to identify potential medication selection errors made in outpatient pharmacies with the RxNorm application programming interface (API). METHODS: We conducted a retrospective observational analysis of 537,710 pairs of e-prescription and dispensing records from a mail-order pharmacy for the period January 2017 to October 2018. National Drug Codes (NDCs) for each pair were obtained from the National Library of Medicine's (NLM's) RxNorm API. The API returned RxNorm concept unique identifier (RxCUI) semantic clinical drug (SCD) identifiers associated with every NDC. The SCD identifiers returned for the e-prescription NDC were matched against the corresponding SCD identifiers from the pharmacy dispensing record NDC. An error matrix was created based on the hand-labeling of mismatched SCD pairs. Performance metrics were calculated for the e-prescription-to-dispensing record matching algorithm for both total pairs and unique pairs of NDCs in these data. RESULTS: We analyzed 527,881 e-prescription and pharmacy dispensing record pairs. Four clinically significant cases of mismatched RxCUI identifiers were detected (ie, three different ingredient selections and one different strength selection). A total of 546 less significant cases of mismatched RxCUIs were found. Nearly all of the NDC pairs had matching RxCUIs (28,787/28,817, 99.90%-525,270/527,009, 99.67%). The RxNorm API had a sensitivity of 1, a false-positive rate of 0.00104 to 0.00312, specificity of 0.99896 to 0.99688, precision of 0.00727 to 0.04255, and F1 score of 0.01444 to 0.08163. We found 872 pairs of records without an RxCUI. CONCLUSIONS: The NLM's RxNorm API can perform an independent and automatic double-check of correct medication selection to verify e-prescription processing at outpatient pharmacies. RxNorm has near-comprehensive coverage of prescribed medications and can be used to recover from medication selection errors. In the future, tools such as this may be able to perform automated verification of medication selection accurately enough to free pharmacists from having to perform manual double-checks of the medications selected within pharmacy dispensing software to fulfill e-prescriptions.

Quantification of Head Shape from Three-Dimensional Photography for Presurgical and Postsurgical Evaluation of Craniosynostosis.

BACKGROUND: Evaluation of surgical treatment for craniosynostosis is typically based on subjective visual assessment or simple clinical metrics of cranial shape that are prone to interobserver variability. Three-dimensional photography provides cheap and noninvasive information to assess surgical outcomes, but there are no clinical tools to analyze it. The authors aim to objectively and automatically quantify head shape from three-dimensional photography. METHODS: The authors present an automatic method to quantify intuitive metrics of local head shape from three-dimensional photography using a normative statistical head shape model built from 201 subjects. The authors use these metrics together with a machine learning classifier to distinguish between patients with (n = 266) and without (n = 201) craniosynostosis (aged 0 to 6 years). The authors also use their algorithms to quantify objectively local surgical head shape improvements on 18 patients with presurgical and postsurgical three-dimensional photographs. RESULTS: The authors' methods detected craniosynostosis automatically with 94.74 percent sensitivity and 96.02 percent specificity. Within the data set of patients with craniosynostosis, the authors identified correctly the fused sutures with 99.51 percent sensitivity and 99.13 percent specificity. When the authors compared quantitatively the presurgical and postsurgical head shapes of patients with craniosynostosis, they obtained a significant reduction of head shape abnormalities (p < 0.05), in agreement with the treatment approach and the clinical observations. CONCLUSIONS: Quantitative head shape analysis and three-dimensional photography provide an accurate and objective tool to screen for head shape abnormalities at low cost and avoiding imaging with radiation and/or sedation. The authors' automatic quantitative framework allows for the evaluation of surgical outcomes and has the potential to detect relapses. CLINICAL QUESTION/LEVEL OF EVIDENCE: Diagnostic, I.

Quantitative evaluation of local head malformations from three-dimensional photography: application to craniosynostosis.

The evaluation of head malformations plays an essential role in the early diagnosis, the decision to perform surgery and the assessment of the surgical outcome of patients with craniosynostosis. Clinicians rely on two metrics to evaluate the head shape: head circumference (HC) and cephalic index (CI). However, they present a high inter-observer variability and they do not take into account the location of the head abnormalities. In this study, we present an automated framework to objectively quantify the head malformations, HC, and CI from three-dimensional (3D) photography, a radiation-free, fast and non-invasive imaging modality. Our method automatically extracts the head shape using a set of landmarks identified by registering the head surface of a patient to a reference template in which the position of the landmarks is known. Then, we quantify head malformations as the local distances between the patient's head and its closest normal from a normative statistical head shape multi-atlas. We calculated cranial malformations, HC, and CI for 28 patients with craniosynostosis, and we compared them with those computed from the normative population. Malformation differences between the two populations were statistically significant (p<0.05) at the head regions with abnormal development due to suture fusion. We also trained a support vector machine classifier using the malformations calculated and we obtained an improved accuracy of 91.03% in the detection of craniosynostosis, compared to 78.21% obtained with HC or CI. This method has the potential to assist in the longitudinal evaluation of cranial malformations after surgical treatment of craniosynostosis.

Skeletal Shape Correspondence Through Entropy.

We present a novel approach for improving the shape statistics of medical image objects by generating correspondence of skeletal points. Each object's interior is modeled by an s-rep, i.e., by a sampled, folded, two-sided skeletal sheet with spoke vectors proceeding from the skeletal sheet to the boundary. The skeleton is divided into three parts: the up side, the down side, and the fold curve. The spokes on each part are treated separately and, using spoke interpolation, are shifted along that skeleton in each training sample so as to tighten the probability distribution on those spokes' geometric properties while sampling the object interior regularly. As with the surface/boundary-based correspondence method of Cates et al., entropy is used to measure both the probability distribution tightness and the sampling regularity, here of the spokes' geometric properties. Evaluation on synthetic and real world lateral ventricle and hippocampus data sets demonstrate improvement in the performance of statistics using the resulting probability distributions. This improvement is greater than that achieved by an entropy-based correspondence method on the boundary points.

Radiation-free quantification of head malformations in craniosynostosis patients from 3D photography.

The evaluation of cranial malformations plays an essential role both in the early diagnosis and in the decision to perform surgical treatment for craniosynostosis. In clinical practice, both cranial shape and suture fusion are evaluated using CT images, which involve the use of harmful radiation on children. Three-dimensional (3D) photography offers non-invasive, radiation-free, and anesthetic-free evaluation of craniofacial morphology. The aim of this study is to develop an automated framework to objectively quantify cranial malformations in patients with craniosynostosis from 3D photography. We propose a new method that automatically extracts the cranial shape by identifying a set of landmarks from a 3D photograph. Specifically, it registers the 3D photograph of a patient to a reference template in which the position of the landmarks is known. Then, the method finds the closest cranial shape to that of the patient from a normative statistical shape multi-atlas built from 3D photographs of healthy cases, and uses it to quantify objectively cranial malformations. We calculated the cranial malformations on 17 craniosynostosis patients and we compared them with the malformations of the normative population used to build the multi-atlas. The average malformations of the craniosynostosis cases were 2.68 ± 0.75 mm, which is significantly higher (p<0.001) than the average malformations of 1.70 ± 0.41 mm obtained from the normative cases. Our approach can support the quantitative assessment of surgical procedures for cranial vault reconstruction without exposing pediatric patients to harmful radiation.

Intracranial Volume Quantification from 3D Photography.

3D photography offers non-invasive, radiation-free, and anesthetic-free evaluation of craniofacial morphology. However, intracranial volume (ICV) quantification is not possible with current non-invasive imaging systems in order to evaluate brain development in children with cranial pathology. The aim of this study is to develop an automated, radiation-free framework to estimate ICV. Pairs of computed tomography (CT) images and 3D photographs were aligned using registration. We used the real ICV calculated from the CTs and the head volumes from their corresponding 3D photographs to create a regression model. Then, a template 3D photograph was selected as a reference from the data, and a set of landmarks defining the cranial vault were detected automatically on that template. Given the 3D photograph of a new patient, it was registered to the template to estimate the cranial vault area. After obtaining the head volume, the regression model was then used to estimate the ICV. Experiments showed that our volume regression model predicted ICV from head volumes with an average error of 5.81 ± 3.07% and a correlation (R2) of 0.96. We also demonstrated that our automated framework quantified ICV from 3D photography with an average error of 7.02 ± 7.76%, a correlation (R2) of 0.94, and an average estimation error for the position of the cranial base landmarks of 11.39 ± 4.3mm.

Entropy-based Correspondence Improvement of Interpolated Skeletal Models.

Statistical analysis of shape representations relies on having good correspondence across a population. Improving correspondence yields improved statistics. Point distribution models (PDMs) are often used to represent object boundaries. Skeletal representations (s-reps) model object widths and boundary directions as well as boundary positions, so they should yield better correspondence. We present two methods: one for continuously interpolating a discretely-sampled skeletal model and one for improving correspondence by using this interpolation to shift skeletal samples to new positions. The interpolation operates by an extension of the mathematics of medial structures. As with Cates' boundary-based method, we evaluate correspondence in terms of regularity and shape-feature population entropies. Evaluation on both synthetic and real data shows that our method both improves correspondence of s-rep models fit to segmented lateral ventricles and that the combined boundary-and-skeletal PDMs implied by these optimized s-reps have better correspondence than optimized boundary PDMs.

Skeletal shape correspondence via entropy minimization.

PURPOSE: Improving the shape statistics of medical image objects by generating correspondence of interior skeletal points. DATA: Synthetic objects and real world lateral ventricles segmented from MR images. METHODS: Each object's interior is modeled by a skeletal representation called the s-rep, which is a quadrilaterally sampled, folded 2-sided skeletal sheet with spoke vectors proceeding from the sheet to the boundary. The skeleton is divided into three parts: up-side, down-side and fold-curve. The spokes on each part are treated separately and, using spoke interpolation, are shifted along their skeletal parts in each training sample so as to tighten the probability distribution on those spokes' geometric properties while sampling the object interior regularly. As with the surface-based correspondence method of Cates et al., entropy is used to measure both the probability distribution tightness and sampling regularity. The spokes' geometric properties are skeletal position, spoke length and spoke direction. The properties used to measure the regularity are the volumetric subregions bounded by the spokes, their quadrilateral sub-area and edge lengths on the skeletal surface and on the boundary. RESULTS: Evaluation on synthetic and real world lateral ventricles demonstrated improvement in the performance of statistics using the resulting probability distributions, as compared to methods based on boundary models. The evaluation measures used were generalization, specificity, and compactness. CONCLUSIONS: S-rep models with the proposed improved correspondence provide significantly enhanced statistics as compared to standard boundary models.

Fitting Skeletal Object Models Using Spherical Harmonics Based Template Warping.

We present a scheme that propagates a reference skeletal model (s-rep) into a particular case of an object, thereby propagating the initial shape-related layout of the skeleton-to-boundary vectors, called spokes. The scheme represents the surfaces of the template as well as the target objects by spherical harmonics and computes a warp between these via a thin plate spline. To form the propagated s-rep, it applies the warp to the spokes of the template s-rep and then statistically refines. This automatic approach promises to make s-rep fitting robust for complicated objects, which allows s-rep based statistics to be available to all. The improvement in fitting and statistics is significant compared with the previous methods and in statistics compared with a state-of-the-art boundary based method.