Area Under ROC Curve Values Research Articles

Predicting the potential distribution of main malaria vectors Anopheles stephensi, An. culicifacies s.l. and An. fluviatilis s.l. in Iran based on maximum entropy model

Malaria is considered as a major public health problem in southern areas of Iran. The goal of this study was to predict best ecological niches of three main malaria vectors of Iran: Anopheles stephensi, Anopheles culicifacies s.l. and Anopheles fluviatilis s.l. A databank was created which included all published data about Anopheles species of Iran from 1961 to 2015. The suitable environmental niches for the three above mentioned Anopheles species were predicted using maximum entropy model (MaxEnt). AUC (area under Roc curve) values were 0.943, 0.974 and 0.956 for An. stephensi, An. culicifacies s.l. and An. fluviatilis s.l respectively, which are considered as high potential power of model in the prediction of species niches. The biggest bioclimatic contributor for An. stephensi and An. fluviatilis s.l. was bio 15 (precipitation seasonality), 25.5% and 36.1% respectively, followed by bio 1 (annual mean temperature), 20.8% for An. stephensi and bio 4 (temperature seasonality) with 49.4% contribution for An. culicifacies s.l. This is the first step in the mapping of the country’s malaria vectors. Hence, future weather situation can change the dispersal maps of Anopheles. Iran is under elimination phase of malaria, so that such spatio-temporal studies are essential and could provide guideline for decision makers for IVM strategies in problematic areas.

Two-dimensional fractal analysis of retinal tissue of healthy and diabetic eyes with optical coherence tomography

In the ophthalmic research, the measurement of the retinal thickness is usually employed for characterizing the structural changes of the retinal tissue. However, changes in the fractal dimension (FD) may provide additional information regarding the structure of the retinal layers and their early damage in ocular diseases. In the present paper, we investigated the possibility of detecting changes in the structure of the cellular layers of the retina by applying a two-dimensional fractal analysis to optical coherence tomography (OCT) images. OCT images were obtained from diabetic patients without retinopathy (DM, n = 38 eyes) and with mild diabetic retinopathy (MDR, n = 43 eyes) as well as in healthy subjects (Controls, n = 74 eyes). The two-dimensional fractal dimension was calculated using the differentiate box counting methodology. We evaluated the usefulness of quantifying the fractal dimension of layered structures in the detection of retinal damage. Generalized estimating equations considering within-subject inter-eye relations were used to test for differences between the groups. An adjusted p-value of <0.001 was considered statistically significant. Receiver operating characteristic (ROC) curves were constructed to describe the ability of the fractal dimension to discriminate between the eyes of DM, MDR, and healthy eyes. Lower values of the fractal dimension were observed in all layers in the MDR eyes compared with controls except in the inner nuclear layer (INL). Lower values of the fractal dimension were also found in all layers in the MDR eyes compared with DM eyes. The highest area under receiver operating characteristic curve (AUROC) values estimated for the fractal dimension were observed for the outer plexiform layer (OPL) and outer segment photoreceptors (OS) when comparing MDR eyes with controls. The highest AUROC value estimated for the fractal dimension were also observed for the retinal nerve fiber layer (RNFL) and OS when comparing MDR eyes with DM eyes. Our results suggest that fractal dimension of the intraretinal layers may provide useful information to differentiate pathological from healthy eyes. Further research is warranted to determine how this approach may be used to aid diagnosis of retinal neurodegeneration at the early stage.

Different medical data mining approaches based prediction of ischemic stroke

AimMedical data mining (also called knowledge discovery process in medicine) processes for extracting patterns from large datasets. In the current study, we intend to assess different medical data mining approaches to predict ischemic stroke. Materials and methodsThe collected dataset from Turgut Ozal Medical Centre, Inonu University, Malatya, Turkey, comprised the medical records of 80 patients and 112 healthy individuals with 17 predictors and a target variable. As data mining approaches, support vector machine (SVM), stochastic gradient boosting (SGB) and penalized logistic regression (PLR) were employed. 10-fold cross validation resampling method was utilized, and model performance evaluation metrics were accuracy, area under ROC curve (AUC), sensitivity, specificity, positive predictive value and negative predictive value. The grid search method was used for optimizing tuning parameters of the models. ResultsThe accuracy values with 95% CI were 0.9789 (0.9470–0.9942) for SVM, 0.9737 (0.9397–0.9914) for SGB and 0.8947 (0.8421–0.9345) for PLR. The AUC values with 95% CI were 0.9783 (0.9569–0.9997) for SVM, 0.9757 (0.9543–0.9970) for SGB and 0.8953 (0.8510–0.9396) for PLR. ConclusionsThe results of the current study demonstrated that the SVM produced the best predictive performance compared to the other models according to the majority of evaluation metrics. SVM and SGB models explained in the current study could yield remarkable predictive performance in the classification of ischemic stroke.

C-Peptide-Based Index Is More Related to Incident Type 2 Diabetes in Non-Diabetic Subjects than Insulin-Based Index.

BackgroundDiabetes can be efficiently prevented by life style modification and medical therapy. So, identification for high risk subjects for incident type 2 diabetes is important. The aim of this study is to identify the best β-cell function index to identify high risk subjects in non-diabetic Koreans.MethodsThis is a retrospective longitudinal study. Total 140 non-diabetic subjects who underwent standard 2-hour 75 g oral glucose tolerance test from January 2007 to February 2007 at Kangbuk Samsung Hospital and followed up for more than 1 year were analyzed (mean follow-up, 54.9±16.4 months). The subjects were consist of subjects with normal glucose tolerance (n=44) and subjects with prediabetes (n=97) who were 20 years of age or older. Samples for insulin and C-peptide levels were obtained at 0 and 30 minutes at baseline.ResultsThirty subjects out of 140 subjects (21.4%) developed type 2 diabetes. When insulin-based index and C-peptide-based index are compared between progressor and non-progressor to diabetes, all C-peptide-based indices were statistically different between two groups, but only insulinogenic index and disposition index among insulin-based index were statistically different. C-peptide-based index had higher value of area under receiver operating characteristic curve (AROC) value than that of insulin-based index. "C-peptidogenic" index had highest AROC value among indices (AROC, 0.850; 95% confidence interval, 0.761 to 0.915). C-peptidogenic index had significantly higher AROC than insulinogenic index (0.850 vs. 0.731 respectively; P=0.014).ConclusionC-peptide-based index was more closely related to incident type 2 diabetes in non-diabetic subjects than insulin-based index.

Serum levels of B-cell activating factor in chronic hepatitis B virus infection: association with clinical diseases.

B-lymphocyte activation is a common characteristic of chronic hepatitis B virus (HBV) infection. B cell-activating factor (BAFF) plays a crucial role in the development and activation of B lymphocytes. This study investigated serum BAFF levels in 232 patients with different clinical diseases of chronic HBV infection [33 chronic asymptomatic HBV carrier (ASC), 53 chronic hepatitis (CH), 72 liver cirrhosis (LC), and 74 hepatocellular carcinoma (HCC)] and 61 gender- and age-matched healthy controls. Serum BAFF levels in HBV patients were significantly elevated compared with healthy controls (P<0.001). HCC patients had significantly higher levels of serum BAFF than ASC, CH, and LC (all P<0.001). Serum levels of BAFF in LC were significantly higher than in ASC (P<0.001) and CH (P=0.002). Serum level of BAFF was an independent variable associated with the presence of HCC in comparison with other disease groups in multivariate analysis. The area under receiver-operating characteristic curve (AUC) value of BAFF levels was 0.914 for HCC versus ASC, 0.825 for HCC versus CH, and 0.607 for HCC versus LC, respectively. The AUC value of BAFF levels was 0.854 for LC versus ASC and 0.748 for LC versus CH, respectively. The AUC value of BAFF (0.888) for HCC was higher than that of alpha-fetoprotein (0.776). We first demonstrate that serum BAFF levels in chronic HBV infection are elevated, correlated with clinical diseases, and could be used as a biomarker for indicating disease mechanisms, activity, and diagnosis.

Conventional and functional MR imaging of peripheral nerve sheath tumors: initial experience.

Differentiating benign from malignant peripheral nerve sheath tumors can be very challenging using conventional MR imaging. Our aim was to test the hypothesis that conventional and functional MR imaging can accurately diagnose malignancy in patients with indeterminate peripheral nerve sheath tumors. This institutional review board-approved, Health Insurance Portability and Accountability Act-compliant study retrospectively reviewed 61 consecutive patients with 80 indeterminate peripheral nerve sheath tumors. Of these, 31 histologically proved peripheral nerve sheath tumors imaged with conventional (unenhanced T1, fluid-sensitive, contrast-enhanced T1-weighted sequences) and functional MR imaging (DWI/apparent diffusion coefficient mapping, dynamic contrast-enhanced MR imaging) were included. Two observers independently assessed anatomic (size, morphology, signal) and functional (ADC values, early arterial enhancement by dynamic contrast-enhanced MR) features to determine interobserver agreement. The accuracy of MR imaging for differentiating malignant from benign was also determined by receiver operating characteristic analysis. Of 31 peripheral nerve sheath tumors, there were 9 malignant (9%) and 22 benign ones (81%). With anatomic sequences, average tumor diameter (6.3 ± 1.8 versus 3.9 ± 2.3 mm, P = .009), ill-defined/infiltrative margins (77% versus 32%; P = .04), and the presence of peritumoral edema (66% versus 23%, P = .01) were different for malignant peripheral nerve sheath tumors and benign peripheral nerve sheath tumors. With functional sequences, minimum ADC (0.47 ± 0.32 × 10(-3) mm(2)/s versus 1.08 ± 0.26 × 10(-3) mm(2)/s; P < .0001) and the presence of early arterial enhancement (50% versus 11%; P = .03) were different for malignant peripheral nerve sheath tumors and benign peripheral nerve sheath tumors. The minimum ADC (area under receiver operating characteristic curve was 0.89; 95% confidence interval, 0.73-0.97) and the average tumor diameter (area under the curve = 0.8; 95% CI, 0.66-0.94) were accurate in differentiating malignant peripheral nerve sheath tumors from benign peripheral nerve sheath tumors. With threshold values for minimum ADC ≤ 1.0 × 10(-3) mm(2)/s and an average diameter of ≥4.2 cm, malignancy could be diagnosed with 100% sensitivity (95% CI, 66.4%-100%). Average tumor diameter and minimum ADC values are potentially important parameters that may be used to distinguish malignant peripheral nerve sheath tumors from benign peripheral nerve sheath tumors.

Accuracy of Macular Ganglion-Cell Complex Thickness to Total Retina Thickness Ratio to Detect Glaucoma in White Europeans

The aim of this study was to evaluate the diagnostic accuracy of macular ganglion-cell complex (GCC) thickness to total retinal thickness ratio (G/T ratio), a parameter found particularly accurate in Japanese eyes to detect glaucoma and recently proposed for the RTVue-100 Fourier-domain optical coherence tomography (RTVue-100 OCT), and to compare the diagnostic accuracy of this parameter to that of parameters provided by the instrument software in European eyes. Retinal nerve fiber layer thickness (RNFLT) and macular GCC thickness parameters of 1 randomly selected eye of 177 white European patients (50 healthy, 28 ocular hypertensive, 33 preperimetric glaucoma, and 66 perimetric glaucoma eyes) were measured with an RTVue-100 OCT. The RNFLT and GCC parameter values of the healthy group were significantly higher than those of the various other groups (P<0.001 for all comparisons). For all separation (normal vs. all other eyes; normal vs. preperimetric and perimetric glaucoma eyes; and normal vs. perimetric glaucoma eyes) average RNFLT, inferior RNFLT, average GCC, and inferior GCC showed consistently higher area under receiver operating characteristics curve (AUROC) than G/T ratio. In the normal versus perimetric glaucoma separation, the AUROC values were 0.977, 0.982, 0.969, 0.976, and 0.959 for average RNFLT, inferior RNFLT, average GCC, inferior GCC, and G/T ratio, respectively. In contrast to Japanese eyes, in white Europeans G/T ratio does not improve separation of glaucomatous and healthy eyes.

MR elastography of liver disease: State of the art

Hepatic fibrosis is a common result of different chronic liver diseases caused by viral infections, alcohol abuse, nonalcoholic fatty liver disease (NAFLD), autoimmune disease and metabolic/genetic disorders. Hepatic fibrosis is a dynamic process which can, in some cases, be reversible with effective treatment (1, 2). Otherwise, it can progress to more advanced stages including cirrhosis where liver function is impaired and severe complications can occur such as variceal bleeding, ascites, portal hypertension, hepatocellular carcinoma and death (3). In 2009, chronic liver disease and cirrhosis were responsible for 143,000 hospitalizations, 11,000 in-hospital deaths (7.7% death rate), and 6.7 billion dollars of hospital charges in total (4). Accurately detecting and staging fibrosis is important in the management of chronic liver disease for initiating treatment and monitoring its effects on the liver. Chronic viral hepatitis C (HCV) infection is a significant cause of hepatic disease and its treatment may depend on the presence of substantial fibrosis (5, 6). It is well known that liver biopsy is the current reference method for the detection and staging of fibrosis. However, it is an invasive method and can cause undesired complications including bleeding (1.7%) (7) and death (0.01-1%) (7-9). The accuracy of liver biopsy is affected by sampling errors (20-33% of patients can have fibrosis stage misclassified by 1 category) (10, 11), and intra- and inter-operator inconsistency (Kappa coefficient 0.62 – 0.89, where 1 means perfect agreement) (11, 12). Furthermore, this approach is not well accepted by patients and physicians alike (13). Conventional noninvasive imaging techniques (ultrasound, computed-tomography, MRI) can visualize the morphologic changes of liver due to fibrosis and cirrhosis. However, the sensitivity and accuracy of these modalities are low for detecting earlier stages of fibrosis and thus are not suitable for staging liver fibrosis over its entire spectrum (14). Over the last several years, new noninvasive methods of assessing liver fibrosis have been developed and investigated, including serum tests, diffusion-weighted MR imaging (DWI), contrast enhanced MR imaging (CE-MRI), ultrasound based transient elastography (UTE) and MR elastography (MRE). Accumulated data showed that the AUROC (area under receiver operating characteristic curve) values are between 77%-91% for serum tests (15-20), 88% for DWI (21), and 83% for CE-MRI (22) to detect the presence of liver fibrosis due to different disease causes. As one of the two existing methods of measuring liver stiffness, UTE has higher AUROC values (0.79-0.98) (23-25), yet it is less accurate for detecting early fibrosis stages and has a high technical failure rate in certain patient subsets (ascites, narrow intercostals space, obesity) (26-28). The other method of measuring liver stiffness, MR elastography (MRE) is the most accurate (AUROC = 92% - 100%) method for detecting and staging liver fibrosis to date, especially in early stages of fibrosis because of its capability of measuring cross-sectional areas of hepatic parenchyma, and the technique is not affected by the ascites and obesity (3, 21, 26, 27, 29-31). In one study, MRE even detected necroinflammation before the onset of fibrosis in patients with NAFLD (32). The detection of early stage fibrosis is of clinical significance as treatments for liver disease are thought to be more effective when compared to later stages of fibrosis. Given its technical advantages and high diagnostic accuracy, this article reviews the application of MR elastography to address clinical questions involving HCV, NAFLD, and methotrexate (MTX) exposure.

Aspartate Aminotransferase to Platelet Ratio Index and FibroScan for Predicting Liver Fibrosis with Chronic Hepatitis B

Background: Non-invasive test have been constructed and evaluated mainly for binary diagnoses. The accuracy of non-invasive tests such as aspartate aminotransferase to platelet ratio index (APRI) and transient elastography/FibroScan should be evaluated especially in clinical practice. The aim of the study was to evaluate the accuracy of detailed fibrosis classification available for APRI and FibroScan to liver biopsy in chronic hepatitis B patients. Method: A cross sectional study was conducted in 51 patients with chronic hepatitis B. The patients underwent laboratory test, FibroScan and liver biopsy between April 2011 and July 2013 at Adam Malik Hospital, Medan. Liver biopsy was assessed based on the METAVIR score. Area under receiver operating characteristic curve (AUROC) predictive value was used to evaluate the accuracy of APRI and FibroScan. All data were analyzed using SPSS 20.0. Results : APRI versus METAVIR diagnosed severe fibrosis and cirrhosis with sensitivity 40% and specificity 83.9%, positive predictive value (PPV) 61.5%, negative predictive value (NPV) 68.4%, positive likelihood ratio (LR) 2.48 and negative LR 0.72 with diagnostic accuracy 66.7%. The AUROC value was 0.619 (95% CI = 0.446 – 0.715); kappa = 0.255; p > 0.05. FibroScan versus METAVIR predictive value with sensitivity 75% and specificity 67.6% were PPV 60%, NPV 67.7%, positive LR 2.31 and negative LR 0.36 with diagnostic accuracy 70.6%. The AUROC value was 0.714 (95% CI = 0.567–0.861); kappa 0.409; p < 0.05. Conclusion : FibroScan has better accuracy than APRI for predicting severe fibrosis and cirrhosis in patients with chronic hepatitis B. Keywords : APRI, FibroScan, liver biopsy, chronic hepatitis B

Reconstructing In-Silico Gene Regulatory Network Based on Reproducing Kernel Hilbert Space

In the fourth Dialogue for Reverse Engineering Assessments and Methods(DREAM4) competition,In-Silico dataset(Challenge 2) was generated with a 'true' biological gene network. The aim of this work is to reconstruct gene network structure from the data. Here, the authors presented a statistical independent measurement method based on reproducing kernel Hilbert space — Hilbert-Schmidt independence criteria(HSIC) to identify the gene regulatory network. Instead of data fitting, HSIC provides a criterion to measure the statistical dependence. Besides, it is a nonparametric method, has no assumption on the data distribution and computationally efficient. Comparative experimental results showed that the HSIC achieved a better performance than several classical gene regulatory network modeling methods. For size 10 network, the area under receiver operating characteristic curve(AUROC) value obtained by HSIC was 16 percent higher than ordinary differential equations(ODE), 7 percent higher than granger causality(GC), 2.4 percent and 1.4 percent higher than the best algorithm in dynamic Bayesian network(DBN) and nonlinear dynamical systems(NDS), respectively. For size 100 network, the AUROC value of HSIC was 16 percent higher than ODE,14.2 percent higher than GC, 5 percent and 1.4 percent higher than the best algorithm in DBN and NDS.These results reveal that the HSIC method has stronger capability of structural identification and is feasible for construct complex gene regulatory network.

Clinical review: Predictive value of neutrophil gelatinase-associated lipocalin for acute kidney injury in intensive care patients

Neutrophil gelatinase-associated lipocalin (NGAL) may be an early marker of acute kidney injury (AKI), but elevated NGAL occurs in a wide range of systemic diseases. Because intensive care patients have high levels of comorbidity, our objective was to conduct a systematic review of the literature to evaluate the value of plasma and urinary NGAL to predict AKI in these patients. We conducted a systematic electronic literature search of MEDLINE through PubMed, EMBASE, and Cochrane Library for all English language research publications evaluating the predictive value of plasma or urinary NGAL (or both) for AKI in adult intensive care patients. Two authors independently extracted data by using a standardized extraction sheet including study characteristics, type of NGAL measurements, and type of outcome measures. The primary summary measure was area under receiver operating characteristic curve (AuROC) for NGAL to predict study outcomes. Eleven studies with a total of 2,875 (range of 20 to 632) participants were included: seven studies assessed urinary NGAL and six assessed plasma NGAL. The included studies varied in design, including observation period from NGAL sampling to AKI follow-up (range of 12 hours to 7 days), definition of baseline creatinine value, and urinary NGAL quantification method (normalizing to urinary creatinine or absolute concentration). AuROC values for the prediction of AKI ranged from 0.54 to 0.98. Five studies reported AuROC for use of renal replacement therapy ranging from 0.73 to 0.89, and four studies reported AuROC for mortality ranging from 0.58 to 0.83. There were no differences in the predictive values of urinary and plasma NGAL. The heterogeneity in study design and results made it difficult to evaluate the value of NGAL to predict AKI in intensive care patients. NGAL seems to have reasonable value in predicting use of renal replacement therapy but not mortality.

L1pred: A Sequence-Based Prediction Tool for Catalytic Residues in Enzymes with the L1-logreg Classifier

To understand enzyme functions, identifying the catalytic residues is a usual first step. Moreover, knowledge about catalytic residues is also useful for protein engineering and drug-design. However, to experimentally identify catalytic residues remains challenging for reasons of time and cost. Therefore, computational methods have been explored to predict catalytic residues. Here, we developed a new algorithm, L1pred, for catalytic residue prediction, by using the L1-logreg classifier to integrate eight sequence-based scoring functions. We tested L1pred and compared it against several existing sequence-based methods on carefully designed datasets Data604 and Data63. With ten-fold cross-validation, L1pred showed the area under precision-recall curve (AUPR) and the area under ROC curve (AUC) of 0.2198 and 0.9494 on the training dataset, Data604, respectively. In addition, on the independent test dataset, Data63, it showed the AUPR and AUC values of 0.2636 and 0.9375, respectively. Compared with other sequence-based methods, L1pred showed the best performance on both datasets. We also analyzed the importance of each attribute in the algorithm, and found that all the scores contributed more or less equally to the L1pred performance.

The Effects of Ambient Lighting in Chest Radiology Reading Rooms

Under typical dark chest radiography reading room conditions, a radiologist's pupils contract and dilate as their visual focus intermittently shifts between the high luminance monitor and the darker background wall, resulting in increased visual fatigue and degradation of diagnostic performance. A controlled increase of ambient lighting may minimize these visual adjustments and potentially improve comfort and accuracy. This study was designed to determine the effect of a controlled increase of ambient lighting on chest radiologist nodule detection performance. Four chest radiologists read 100 radiographs (50 normal and 50 containing a subtle nodule) under low (E=1 lx) and elevated (E=50 lx) ambient lighting levels on a DICOM-calibrated, medical-grade liquid crystal display. Radiologists were asked to identify nodule locations and rate their detection confidence. A receiver operating characteristic (ROC) analysis of radiologist results was performed and area under ROC curve (AUC) values calculated for each ambient lighting level. Additionally, radiologist selection times under both illuminance conditions were determined. Average AUC values did not significantly differ (p>0.05) between ambient lighting levels (estimated mean difference=-0.03; 95% CI, (-0.08, 0.03)). Average selection times decreased or remained constant with increased illuminance. The most considerable decreases occurred for false positive identification times (35.4±18.8 to 26.2±14.9 s) and true positive identification times (29.7±18.3 to 24.5±15.5 s). No performance differences were statistically significant. Study findings suggest that a controlled increase of ambient lighting within darkly lit chest radiology reading rooms, to a level more suitable for performance of common radiological tasks, does not appear to have a statistically significant effect on nodule detection performance.

Classifying DNA repair genes by kernel-based support vector machines.

Human longevity is a complex phenotype that has a significant genetic predisposition. Like other biological processes, ageing process is governed through the regulation of signaling pathways and transcription factors. The DNA damage theory of ageing suggests that ageing is a consequence of un-repaired DNA damage accumulation. Intensive research has been carried out to elucidate the role of DNA repair systems in the ageing process. Decision Trees and Naive Bayesian Algorithm are two data-mining based classification methods for systematically analyzing data about human DNA repair genes. In this paper we develop a linearly combined kernel with Support Vector Machine (SVM) to analyze the ageing related data. The popular supervised learning algorithm enables better discrimination between ageing-related and non-ageing-related DNA repair genes. The linear combination of linear kernel and polynomial kernel of degree 3 in conjunction with SVM allows better classification accuracy in DNA repair gene data set. Compared to Decision Trees and Naive Bayesian Algorithm, SVM with the proposed kernel can achieve 65% AUC (Area Under ROC Curve) values, in contrast to 51.1% and 52.1% respectively. More importantly, we obtain 5 significant ageingrelated genes selected through the training on the whole data set and they are PCNA, PARP, APEX1, MLH1 and XRCC6. Different from the two methods, we can identify another important gene PCNA in the pathways the two methods targeted, while they failed to. And two novel genes PARP, MLH1 are selected as well. The two genes might provide potential insights for biologists in ageing research. SVM is a powerful and robust classification algorithm that can yield higher predictive accuracies. The selection of proper kernel plays a more important role in fulfilling the classification task. The important genes identified not only can target critical pathways related to ageing but also detected genes that may reveal possible related ageing biomarkers.

WE‐G‐BRA‐01: A Framework to Determine Risk of Pneumonitis Following Radiotherapy for Lung Cancer

Purpose: To assess external validity of a framework, which was previously developed to predict the incidence of radiation‐induced pneumontis in patients following radiotherapy for lung cancer, by testing it on an alternate dataset (other than the one used to develop it). Methods: The study employed two independent datasets: 1) a Duke University dataset of 219 patients, of whom 34 were diagnosed with post‐radiotherapy pneumonitis; and 2) a Washington University (WashU) dataset of 219 patients, of whom 52 developed pneumonitis post‐radiotherapy. While the Duke dataset was used to develop the framework, the WashU dataset was used to assess its validity. The framework was based on a predictive model that combined results from four machine learning models into a single fused model using a Bayesian decision fusion methodology. The results of testing the framework on the WashU dataset was measured in terms of the area under ROC curve (AUC) and compared to the AUC value obtained by Mean Lung Dose (MLD) — a metric commonly used to discriminate between patients with and without pneumonitis. Results: The framework yielded an overall AUC of 0.84 for cross‐validated testing on Duke data. Six patient features were identified as most significant. They consisted of two dose features: EUD (a=1.3) and V30; and four non‐dose features: female gender, tumor in central lung location, adenocarcinoma, and chemotherapy prior to radiotherapy. Most notably, when tested on the WashU dataset, the framework yielded an AUC value of 0.73. In comparison, the AUC value obtained by employing MLD values was only 0.68. Conclusions: Based on testing with an independent dataset, the framework to predict the incidence of radiation‐induced pneumonitis appears to be fairly robust. The framework may potentially be used in the clinic to improve the treatment decision‐making process.

Multimodal microscopy for automated histologic analysis of prostate cancer

BackgroundProstate cancer is the single most prevalent cancer in US men whose gold standard of diagnosis is histologic assessment of biopsies. Manual assessment of stained tissue of all biopsies limits speed and accuracy in clinical practice and research of prostate cancer diagnosis. We sought to develop a fully-automated multimodal microscopy method to distinguish cancerous from non-cancerous tissue samples.MethodsWe recorded chemical data from an unstained tissue microarray (TMA) using Fourier transform infrared (FT-IR) spectroscopic imaging. Using pattern recognition, we identified epithelial cells without user input. We fused the cell type information with the corresponding stained images commonly used in clinical practice. Extracted morphological features, optimized by two-stage feature selection method using a minimum-redundancy-maximal-relevance (mRMR) criterion and sequential floating forward selection (SFFS), were applied to classify tissue samples as cancer or non-cancer.ResultsWe achieved high accuracy (area under ROC curve (AUC) >0.97) in cross-validations on each of two data sets that were stained under different conditions. When the classifier was trained on one data set and tested on the other data set, an AUC value of ~0.95 was observed. In the absence of IR data, the performance of the same classification system dropped for both data sets and between data sets.ConclusionsWe were able to achieve very effective fusion of the information from two different images that provide very different types of data with different characteristics. The method is entirely transparent to a user and does not involve any adjustment or decision-making based on spectral data. By combining the IR and optical data, we achieved high accurate classification.

Quantitative prediction of MHC-II binding affinity using particle swarm optimization

Helper T-cell epitopes (Th epitopes) are the basic units which activate helper T-cell's immune response, and they are helpful for understanding the immune mechanism and developing vaccines. Peptide and major histocompatibility complex class II (MHC-II) binding is an important prerequisite event for helper T-cell immune response, and the binding peptides are usually recognized as Th epitopes, therefore we can identify Th epitopes by predicting MHC-II binding peptides. Recently, instead of differentiating the peptides as binder or non-binder, researchers are more interested in predicting binding affinities between MHC-II molecules and peptides. Motivated by the collective search strategy of the particle swarm optimization algorithm (PSO), a method was developed to make the direct prediction of peptide binding affinity. In our paper, PSO was utilized to search for the optimal position-specific scoring matrices (PSSM) from the experimentally derived allele-related peptides, and then the prediction models were constructed based on the matrices. Moreover, we evaluated several factors influencing the binding affinity, including peptide length and flanking residue length, and incorporated them into our models. The performance of our models was evaluated on three MHC-II alleles from AntiJen database and 14 MHC-II alleles from IEDB database. When compared to the existing popular quantitative methods such as MHCPred, SVRMHC, ARB and SMM-align, our method can give out better performance in terms of correlation coefficient (r) and area under ROC curve (AUC). In addition, the results demonstrated that the performance of models was further improved by incorporating the global length information, achieving average AUC value of 0.7534 and average r value of 0.4707. Quantitative prediction of MHC-II binding affinity can be modeled as an optimization problem. Our PSO based method can find the optimal PSSM, which will then be used for identifying the binding cores and scoring the binding affinities of the peptides. The experiment results show that our method is promising for the prediction of MHC-II binding affinity.

MO‐EE‐A2‐04: A Framework to Determine the Risk of Developing Pneumonitis in Lung Cancer Patients Undergoing Radiotherapy

Purpose: To determine which dose and non‐dose features of patients undergoing thoracic radiotherapy best predict the incidence of possible lung pneumonitis and to estimate the corresponding risk of developing pneumonitis. Method and Materials: The study employed a database of 235 patients, out of which 34 were diagnosed with Grade 2+ pneumonitis. A total of 94 dose and non‐dose features of each patient were analyzed. To determine the most significant features likely to affect patient outcome, a two‐fold optimization approach was developed. First, the feature values were evaluated by four machine learning models for accurate prediction. The model accuracy was measured in terms of tenfold cross‐validation‐based area under ROC curve (AUC). Next, the ROCs of the four models were combined. Towards that end, a Bayesian decision fusion methodology was developed. Finally, based on the combined ROC, the fusion model isolated the set of most significant features and also computed patient‐specific risk of injury. Results: Out of 94 original features, 27 were selected by the four models. AUC values corresponding to the four models were 0.75, 0.70, 0.79, and 0.73, respectively. Most notably, when the four models were combined, the AUC improved to 0.85. Furthermore, the fusion model identified only six features as most significant. They consisted of two dose features: EUD (a=2) and V30; and four non‐dose features: gender, tumor location, histology type, and chemotherapy schedule. The corresponding probabilities for injured patients were found to be significantly higher than those for non‐injured patients, demonstrating robustness of the over all predictive model. Conclusion: A framework was developed to predict the incidence of pneumonitis in lung cancer patients undergoing radiotherapy. The predicted probabilities of injury were found to correspond well with the known injury status of the patients. The framework may potentially be used in the clinic to improve the treatment decision‐making process.

Model for end stage of liver disease (MELD) is better than the Child-Pugh score for predicting in-hospital mortality related to esophageal variceal bleeding

Aim: The Child Pugh and MELD are good methods for predicting mortality in patients with chronic liver disease. We investigated their performance as risk factors for failure to control bleeding, in-hospital overall mortality and death related to esophageal variceal bleeding episodes. Methods: From a previous collected database, 212 cirrhotic patients with variceal bleeding admitted to our hospital were studied. The predictive capability of Child Pugh and MELD scores were compared using c statistics. Results: The Child-Pugh and MELD scores showed marginal capability for predicting failure to control bleeding (the area under receiver operating characteristics curve (AUROC) values were < 0.70 for both). The AUROC values for predicting inhospital overall mortality of Child-Pugh and MELD score were similar: 0.809 (CI 95%, 0.710 - 0.907) and 0.88 (CI 95% 0.77-0.99,) respectively. There was no significant difference between them (p > 0.05). The AU-ROC value of MELD for predicting mortality related to variceal bleeding was higher than the Child-Pugh score: 0.905 (CI 95% 0.801-1.00) vs 0.794 (CI 95% 0.676 - 0.913) respectively (p < 0.05). Conclusions: MELD and Child-Pugh were not efficacious scores for predicting failure to control bleeding. The Child-Pugh and MELD scores had similar capability for predicting in-hospital overall mortality. Nevertheless, MELD was significantly better than Child-Pugh score for predicting in-hospital mortality related to variceal bleeding.

Impact of Mitral E/A Ratio on the Accuracy of Different Echocardiographic Indices to Estimate Left Ventricular End-Diastolic Pressure

The objective was to determine the influence of left ventricular (LV) inflow pattern on the accuracy of different echocardiographic indices for estimation of LV end-diastolic pressure (LVEDP). Echocardiography with color tissue Doppler imaging (TDI) and LVEDP measurements using fluid-filled catheters were performed in 176 consecutive patients on the same day. Mitral peak diastolic velocities (E, A) and the difference in duration between pulmonary venous retrograde velocity and mitral A-velocity (PV R–A) were recorded by pulsed Doppler. Propagation velocity of the early mitral inflow (V P) was assessed using color M-mode. Early diastolic longitudinal (E′ lat) and radial (E′ radial) velocities of mitral annulus were measured by TDI. Area under ROC curve (AUC) for prediction of elevated LVEDP (≥15 mm Hg) was computed for each parameter. For E/A ≥1 (98 patients, 46 with elevated LVEDP), the AUC values were: PV R–A: 0.914; E/E′ lat: 0.780; E/E′ radial: 0.729; E/V P: 0.712 ( p < 0.001). When E/A <1 (78 patients, 26 with elevated LVEDP), only PV R–A reached statistical significance (AUC = 0.893, p < 0.001). The conclusions were: PV R–A enabled the most accurate noninvasive estimation of LVEDP irrespective of LV filling profile and combined indices E/V P, E/E′ lat and E/E′ radial represent more feasible alternatives for patients with mitral E/A–1. (E-mail: tpoerner@googlemail.com)