Applying Radiomics to Predict Pathology of Postchemotherapy Retroperitoneal Nodal Masses in Germ Cell Tumors.

PURPOSE: After chemotherapy, approximately 50% of patients with metastatic testicular germ cell tumors (GCTs) who undergo retroperitoneal lymph node dissections (RPNLDs) for residual masses have fibrosis. Radiomics uses image processing techniques to extract quantitative textures/features from regions of interest (ROIs) to train a classifier that predicts outcomes. We hypothesized that radiomics would identify patients with a high likelihood of fibrosis who may avoid RPLND. PATIENTS AND METHODS: Patients with GCT who had an RPLND for nodal masses > 1 cm after first-line platinum chemotherapy were included. Preoperative contrast-enhanced axial computed tomography images of retroperitoneal ROIs were manually contoured. Radiomics features (n = 153) were used to train a radial basis function support vector machine classifier to discriminate between viable GCT/mature teratoma versus fibrosis. A nested 10-fold cross-validation protocol was used to determine classifier accuracy. Clinical variables/restricted size criteria were used to optimize the classifier. RESULTS: Seventy-seven patients with 102 ROIs were analyzed (GCT, 21; teratoma, 41; fibrosis, 40). The discriminative accuracy of radiomics to identify GCT/teratoma versus fibrosis was 72 ± 2.2% (area under the curve [AUC], 0.74 ± 0.028); sensitivity was 56.2 ± 15.0%, and specificity was 81.9 ± 9.0% ( P = .001). No major predictive differences were identified when data were restricted by varying maximal axial diameters (AUC range, 0.58 ± 0.05 to 0.74 ± 0.03). The prediction algorithm using clinical variables alone identified an AUC of 0.76. When these variables were added to the radiomics signature, the best performing classifier was identified when axial masses were limited to diameter < 2 cm (accuracy, 88.2 ± 4.4; AUC, 0.80 ± 0.05; P = .02). CONCLUSION: A predictive radiomics algorithm had a discriminative accuracy of 72% that improved to 88% when combined with clinical predictors. Additional independent validation is required to assess whether radiomics allows patients with a high predicted likelihood of fibrosis to avoid RPLND.

Assessing cerebrovascular reactivity by the pattern of response to progressive hypercapnia.

Cerebral blood flow responds to a carbon dioxide challenge, and is often assessed as cerebrovascular reactivity, assuming a linear response over a limited stimulus range or a sigmoidal response over a wider range. However, these assumed response patterns may not necessarily apply to regions with pathophysiology. Deviations from sigmoidal responses are hypothesised to result from upstream flow limitations causing competition for blood flow between downstream regions, particularly with vasodilatory stimulation; flow is preferentially distributed to regions with more reactive vessels. Under these conditions, linear or sigmoidal fitting may not fairly describe the relationship between stimulus and flow. To assess the range of response patterns and their prevalence a survey of healthy control subjects and patients with cerebrovascular disease was conducted. We used a ramp carbon dioxide challenge from hypo- to hypercapnia as the stimulus, and magnetic resonance imaging to measure the flow responses. We categorized BOLD response patterns into four types based on the signs of their linear slopes in the hypo- and hypercapnic ranges, color coded and mapped them onto their respective anatomical scans. We suggest that these type maps complement maps of linear cerebrovascular reactivity by providing a better indication of the actual response patterns. Hum Brain Mapp 38:3415-3427, 2017. © 2017 Wiley Periodicals, Inc.

A New 3D Reconstruction Method To Assess Anatomical Restoration In Vertebral Compression Fractures.

Deformation of the spine following vertebral compression fracture (VCF) can be associated with severe spinal dysfunction. Anatomical restoration is described as the complete reduction and stable fixation of the fracture with the goal to restore the pre-fracture local and global spine properties. The essential need to obtain endplate reduction, and consider the whole vertebral body in the management of vertebral fractures is pointed out. A new 3D imaging technique is described. A new assessment method based on CT scans 3D reconstructions was developed. This new method was used to assess the anatomical restoration of the fractured vertebrae. A procedure consisting in placing two expandable titanium implants for fracture reduction, together with stabilization with a high viscosity PMMA cement was used to demonstrate the method. The cases presented here are demonstrating that this 3D-mapping software is a tool which can address the lack of a valid, reproducible and user-friendly method to evaluate the anatomical restoration in VCF. The evaluation of the anatomical restoration of the fractured vertebrae raises some issues in terms of interpretation because of a lack of consensus in the existing reporting methods. A 3D-reconstruction method, using CT scans and a newly-developed software, is proposed to evaluate the anatomical restoration in a clear, valid, reproducible and user-friendly way.

Longitudinal Study of Postconcussion Syndrome: Not Everyone Recovers.

We examined recovery from postconcussion syndrome (PCS) in a series of 285 patients diagnosed with concussion based on international sport concussion criteria who received a questionnaire regarding recovery. Of 141 respondents, those with postconcussion symptoms lasting less than 3 months, a positive computed tomography (CT) and/or magnetic resonance imaging (MRI), litigants, and known Test of Memory Malingering (TOMM)-positive cases were excluded, leaving 110 eligible respondents. We found that only 27% of our population eventually recovered and 67% of those who recovered did so within the first year. Notably, no eligible respondent recovered from PCS lasting 3 years or longer. Those who did not recover (n = 80) were more likely to be non-compliant with a do-not-return-to-play recommendation (p = 0.006) but did not differ from the recovered group (n = 30) in other demographic variables, including age and sex (p ≥ 0.05). Clustergram analysis revealed that symptoms tended to appear in a predictable order, such that symptoms later in the order were more likely to be present if those earlier in the order were already present. Cox proportional hazards model analysis showed that the more symptoms reported, the longer the time to recovery (p = 7.4 × 10-6), with each additional symptom reducing the recovery rate by approximately 20%. This is the first longitudinal PCS study to focus on PCS defined specifically as a minimum of 3 months of symptoms, negative CT and/or MRI, negative TOMM test, and no litigation. PCS may be permanent if recovery has not occurred by 3 years. Symptoms appear in a predictable order, and each additional PCS symptom reduces recovery rate by 20%. More long-term follow-up studies are needed to examine recovery from PCS.

Postconcussion syndrome: demographics and predictors in 221 patients.

OBJECTIVE The objective of this study was to determine the demographics and predictors of postconcussion syndrome (PCS) in a large series of patients using a novel definition of PCS. METHODS The authors conducted a retrospective cohort study of 284 consecutive concussed patients, 221 of whom had PCS on the basis of at least 3 symptoms persisting at least 1 month. This definition of PCS was uniformly employed and is unique in accepting an expanded list of symptoms, in shortening the postconcussion interval to 1 month from 3 months, and in excluding those with focal injuries such as hemorrhages and contusions. RESULTS The 221 cases showed considerable heterogeneity in clinical features of PCS. They averaged 3.3 concussions, with a range of 0 to 12 or more concussions, and 62.4% occurred during sports and recreation. The median duration of PCS was 7 months at the time of examination, with 11.8% lasting more than 2 years, and 23.1% with PCS had only 1 concussion. The average patient age was 27 years (range 10-74 years). The average number of persistent symptoms was 8.1; 26.2% had a previous psychiatric condition, attention-deficit disorder/attention-deficit hyperactivity disorder, a learning disability, or previous migraine headaches. The prevalence of arachnoid cysts and Chiari malformation in PCS exceeded the general population. Additionally, involvement in litigation, presence of extracranial injuries, amnesia and/or loss of consciousness, and female sex were predictive of reporting a high number of symptoms. A prior history of psychiatric conditions or migraines, cause of injury, number of previous concussions, and age did not significantly predict symptom number. Only the number of symptoms reported predicted the duration of PCS. To predict the number of symptoms for those who fulfilled PCS criteria according to the International Classification of Diseases, 10th Revision (ICD-10), and the Diagnostic and Statistical Manual of Mental Disorders, 4th Edition (DSM-IV), the number of previous concussions was significant. CONCLUSIONS PCS is commonly associated with multiple concussions, but 23.1% in the present series occurred after only 1 concussion. Most patients with PCS had multiple symptoms persisting for months or years. The median duration of PCS was 7 months, with a range up to 26 years. In only 11.3%, the PCS had ended at the time of consultation. Not all predictors commonly cited in the literature align with the findings in this study. This is likely due to differences in the definitions of PCS used in research. These results suggest that the use of ICD-10 and DSM-IV to diagnose PCS may be biased toward those who are vulnerable to concussions or with more severe forms of PCS. It is thus important to redefine PCS based on evidence-based medicine.

Measuring cerebrovascular reactivity: the dynamic response to a step hypercapnic stimulus.

We define cerebral vascular reactivity (CVR) as the ratio of the change in blood oxygen level-dependent (BOLD) magnetic resonance imaging (MRI) signal (S) to an increase in blood partial pressure of CO2 (PCO2): % Δ S/Δ PCO2 mm Hg. Our aim was to further characterize CVR into dynamic and static components and then study 46 healthy subjects collated into a reference atlas and 20 patients with unilateral carotid artery stenosis. We applied an abrupt boxcar change in PCO2 and monitored S. We convolved the PCO2 with a set of first-order exponential functions whose time constant τ was increased in 2-second intervals between 2 and 100 seconds. The τ corresponding to the best fit between S and the convolved PCO2 was used to score the speed of response. Additionally, the slope of the regression between S and the convolved PCO2 represents the steady-state CVR (ssCVR). We found that both prolongations of τ and reductions in ssCVR (compared with the reference atlas) were associated with the reductions in CVR on the side of the lesion. τ and ssCVR are respectively the dynamic and static components of measured CVR.

A Computerized System to Assess Axillary Lymph Node Malignancy from Sonographic Images.

A computational approach to classifying axillary lymph node metastasis in sonographic images is described. One hundred five ultrasound images of axillary lymph nodes from patients with breast cancer were evaluated (81 benign and 24 malignant), and each lymph node was manually segmented, delineating both the whole lymph node and internal hilum surfaces. Normalized signed distance transforms were computed from the segmented boundaries of both structures, and each pixel was then assigned coordinates in a 3-D feature space according to the pixel's intensity, its signed distance to the node boundary and its signed distance to the hilum boundary. Three-dimensional histograms over the feature space were accumulated for each node by summing over all pixels, and the bin counts served as predictor inputs to a support vector machine learning algorithm. Repeated random sampling of 80/25 train/test splits was used to estimate generalization performance and generate receiver operating characteristic curves. The optimal classifier had an area under the receiver operating characteristic curve of 0.95 and sensitivity and specificity of 0.90 and 0.90. Our results indicate the feasibility of axillary nodal staging with computerized analysis.

Intra-vascular blood velocity and volumetric flow rate calculated from dynamic 4D CT angiography using a time of flight technique.

We examine a time of flight (TOF) approach for the analysis of contrast enhanced 4D volumetric CT angiography scans to derive and display blood velocity in arteries. Software was written to divide blood vessels into a series of cross sections and to track contrast bolus TOF along the central vessel axis, which was defined by a user, from 4D CT source data. Time density curves at each vessel cross section were fit with quadratic, Gaussian, and gamma variate functions to determine bolus time to peak (TTP). A straight line was used to plot TTP versus vessel path length for all three functions and the slope used to calculate intraluminal velocity. Software was validated in a simulated square channel and non-pulsatile flow phantom prior to the calculation of blood velocity in the major cerebral arteries of 8 normal patients. The TOF algorithm correctly calculates intra-luminal fluid velocity in eight flow conditions of the CT flow phantom where quadratic functions were used. Across all conditions, in phantoms and in vivo, the success of calculations depended strongly on having a sufficiently long path length to make measurements and avoiding venous contamination. Total blood flow into the brain was approximately 17 % of a normal 5 L cardiac output. The technique was explored in vivo in a patient with subclavian steal syndrome, in the pulmonary arteries and in the iliac artery from clinical 4D CT source data. Intravascular blood velocity and flow may be calculated from 4D CT angiography using a TOF approach.

BROCCOLI: Software for fast fMRI analysis on many-core CPUs and GPUs.

Analysis of functional magnetic resonance imaging (fMRI) data is becoming ever more computationally demanding as temporal and spatial resolutions improve, and large, publicly available data sets proliferate. Moreover, methodological improvements in the neuroimaging pipeline, such as non-linear spatial normalization, non-parametric permutation tests and Bayesian Markov Chain Monte Carlo approaches, can dramatically increase the computational burden. Despite these challenges, there do not yet exist any fMRI software packages which leverage inexpensive and powerful graphics processing units (GPUs) to perform these analyses. Here, we therefore present BROCCOLI, a free software package written in OpenCL (Open Computing Language) that can be used for parallel analysis of fMRI data on a large variety of hardware configurations. BROCCOLI has, for example, been tested with an Intel CPU, an Nvidia GPU, and an AMD GPU. These tests show that parallel processing of fMRI data can lead to significantly faster analysis pipelines. This speedup can be achieved on relatively standard hardware, but further, dramatic speed improvements require only a modest investment in GPU hardware. BROCCOLI (running on a GPU) can perform non-linear spatial normalization to a 1 mm(3) brain template in 4-6 s, and run a second level permutation test with 10,000 permutations in about a minute. These non-parametric tests are generally more robust than their parametric counterparts, and can also enable more sophisticated analyses by estimating complicated null distributions. Additionally, BROCCOLI includes support for Bayesian first-level fMRI analysis using a Gibbs sampler. The new software is freely available under GNU GPL3 and can be downloaded from github (https://github.com/wanderine/BROCCOLI/).

Medical image processing on the GPU - past, present and future.

Graphics processing units (GPUs) are used today in a wide range of applications, mainly because they can dramatically accelerate parallel computing, are affordable and energy efficient. In the field of medical imaging, GPUs are in some cases crucial for enabling practical use of computationally demanding algorithms. This review presents the past and present work on GPU accelerated medical image processing, and is meant to serve as an overview and introduction to existing GPU implementations. The review covers GPU acceleration of basic image processing operations (filtering, interpolation, histogram estimation and distance transforms), the most commonly used algorithms in medical imaging (image registration, image segmentation and image denoising) and algorithms that are specific to individual modalities (CT, PET, SPECT, MRI, fMRI, DTI, ultrasound, optical imaging and microscopy). The review ends by highlighting some future possibilities and challenges.