Deep learning in cardiovascular imaging

Delta-Radiomics Approach Using Contrast-Enhanced and Noncontrast-Enhanced Computed Tomography Images for Predicting Distant Metastasis in Patients With Borderline Resectable Pancreatic Carcinoma

BackgroundBleeding complications in patients with acute ST-segment elevation myocardial infarction (STEMI) have been associated with increased risk of subsequent adverse consequences.ObjectiveThe objective of our study was to develop and externally validate a diagnostic model of in-hospital bleeding.MethodsWe performed multivariate logistic regression of a cohort for hospitalized patients with acute STEMI in the emergency department of a university hospital. Participants: The model development data set was obtained from 4262 hospitalized patients with acute STEMI from January 2002 to December 2013. A set of 6015 hospitalized patients with acute STEMI from January 2014 to August 2019 were used for external validation. We used logistic regression analysis to analyze the risk factors of in-hospital bleeding in the development data set. We developed a diagnostic model of in-hospital bleeding and constructed a nomogram. We assessed the predictive performance of the diagnostic model in the validation data sets by examining measures of discrimination, calibration, and decision curve analysis (DCA).ResultsIn-hospital bleeding occurred in 112 of 4262 participants (2.6%) in the development data set. The strongest predictors of in-hospital bleeding were advanced age and high Killip classification. Logistic regression analysis showed differences between the groups with and without in-hospital bleeding in age (odds ratio [OR] 1.047, 95% CI 1.029-1.066; P<.001), Killip III (OR 3.265, 95% CI 2.008-5.31; P<.001), and Killip IV (OR 5.133, 95% CI 3.196-8.242; P<.001). We developed a diagnostic model of in-hospital bleeding. The area under the receiver operating characteristic curve (AUC) was 0.777 (SD 0.021, 95% CI 0.73576-0.81823). We constructed a nomogram based on age and Killip classification. In-hospital bleeding occurred in 117 of 6015 participants (1.9%) in the validation data set. The AUC was 0.7234 (SD 0.0252, 95% CI 0.67392-0.77289).ConclusionsWe developed and externally validated a diagnostic model of in-hospital bleeding in patients with acute STEMI. The discrimination, calibration, and DCA of the model were found to be satisfactory.Trial RegistrationChiCTR.org ChiCTR1900027578; http://www.chictr.org.cn/showprojen.aspx?proj=45926

BackgroundBroncho-pleural fistula (BPF) and respiratory failure (RF) are life-threatening complications after lung cancer surgery and can result in long-term hospitalization and decreased quality of life. Risk assessments for BPF and RF in addition to mortality and major morbidities are indispensable in surgical decision-making and perioperative care.MethodsThe characteristics and operative data of 80,095 patients who had undergone lung cancer surgery were derived from the 2014 and 2015 National Clinical Database (NCD) of Japan datasets. After excluding 1501 patients, risk models were developed from these data and validated by another dataset for 42,352 patients derived from the 2016 NCD dataset. Receiver operating characteristic curves were generated for postoperative BPF and RF development. The concordance-index was used to assess the discriminatory ability and validity of the model.ResultsBPF and RF occurred in 259 (0.3%) and 420 patients (0.5%), respectively, in the model development dataset and in 129 (0.3%) and 198 patients (0.5%), respectively, in the model validation dataset. Characteristic variables including types of surgery and comorbidities were identified as risk factors for BPF and RF, respectively. The concordance indexes of assessments for BPF and RF were 0.847 (p < 0.001) and 0.848 (p < 0.001), respectively, for the development dataset and 0.850 (p < 0.001) and 0.844 (p < 0.001), respectively, for the validation dataset.ConclusionsThese models are satisfactory for predicting BPF and RF after lung cancer surgery in Japan and could guide preoperative assessment and optimal measures for preventing BPF and RF.

A probabilistic framework for incorporating uncertainty in air quality models is described. The quantitative dependence of the uncertainty in calculated air quality concentrations on the uncertainty in the input meteorological data is illustrated using a simple Second-order Closure integrated Model Plume in combination with the EPRI Plume Model Validation and Development Data Set. Evaluation of the model results demonstrate that even though individual hourly samples cannot be deterministically predicted downwind of a powerplant stack, statistical representations of the observed cumulative distribution of the sample values are quite predictable. We discuss the data needed to improve the definition of the range of meteorological uncertainty within an ensemble of flows defined by given meteorological data, and thus provide for improvements in predictability models of the type illustrated. We argue that attempts to collect the data needed to define more precisely the variance within the ensemble of compatible flows will prove more productive than attempts to eliminate meteorological uncertainties in given datasets.

Necrotizing enterocolitis (NEC) is a major cause of neonatal morbidity and mortality. Preventive and therapeutic research, surveillance, and quality improvement initiatives are hindered by variations in case definitions. To develop a gestational age (GA)-specific case definition for NEC. We conducted a prospective 34-month population study using clinician-recorded findings from the UK National Neonatal Research Database between December 2011 and September 2014 across all 163 neonatal units in England. We split study data into model development and validation data sets and categorized GA into groups (group 1, less than 26 weeks' GA; group 2, 26 to less than 30 weeks' GA; group 3, 30 to less than 37 weeks' GA; group 4, 37 or more weeks' GA). We entered GA, birth weight z score, and clinical and abdominal radiography findings as candidate variables in a logistic regression model, performed model fitting 1000 times, averaged the predictions, and used estimates from the fitted model to develop an ordinal NEC score and cut points to develop a dichotomous case definition based on the highest area under the receiver operating characteristic curves [AUCs] and positive predictive values [PPVs]. Abdominal radiography performed to investigate clinical concerns. Ordinal NEC likelihood score, dichotomous case definition, and GA-specific probability plots. Of the 3866 infants, the mean (SD) birth weight was 2049.1 (1941.7) g and mean (SD) GA was 32 (5) weeks; 2032 of 3663 (55.5%) were male. The total included 2978 infants (77.0%) without NEC and 888 (23.0%) with NEC. Infants with NEC in group 1 were less likely to present with pneumatosis (31.1% vs 47.2%; P = .01), blood in stool (11.8% vs 29.6%; P < .001), or mucus in stool (2.1% vs 5.6%; P = .048) but more likely to present with gasless abdominal radiography findings (6.3% vs 0.9%; P = .009) compared with infants with NEC in group 3. In the ordinal NEC score analysis, we allocated 3 points to pneumatosis, 2 points to blood in stool, and 1 point each to abdominal tenderness and abdominal discoloration; 1 point was assigned if 1 or more of pneumoperitoneum, fixed loop, and portal venous gas were present, and 1 point was assigned if both increased and/or bilious aspirates and abdominal distension were present. The cutoff scores for the dichotomous GA-specific case definition were 2 or greater for infants in groups 1 and 2, 3 or greater for infants in group 3, and 4 or greater for infants in group 4. The ordinal NEC score and dichotomous case definition discriminated well between infants with (AUC, 87%) and without (AUC, 80%) NEC. The case definition has a sensitivity of 66.2% (95% CI, 63.0-69.4), a specificity of 94.4% (95% CI, 93.2-95.4), an AUC of 80.0% (95% CI, 79-82), and a PPV of 85.5% (95% CI, 82.6-88.1). Applying the cut points to the 431 infants who underwent a laparotomy yielded a sensitivity of 76.5% (95% CI, 70.0-82.1), a specificity of 74.4% (95% CI, 68.3-80.0), an AUC of 75.0% (95% CI, 71.0- 80.0), and a PPV of 72.9% (95% CI, 66.4-78.7). The risk of NEC and clinical presentation are associated with GA. Adoption of a consistent GA-specific case definition would strengthen global efforts to reduce the population burden of this devastating neonatal disease.

Applicability of day-to-day variation in behavior for the automated detection of lameness in dairy cows

Taurus cedar (Cedrus libani A. Rich.) is one of the most important tree speciesin Turkey. Therefore, the information is necessary about growth and yield ofthe species for developing future management and planning strategies. The one of the essential building blocks in forestgrowth and yield prediction models is the equations for estimating individualtree volume. In thisstudy, tree volume equations were developed for Taurus cedar stands in WestMediterranean Region. The tested models were compared using six performancecriteria (Fit index, average bias, average absolute residuals, absolute maximumbias, and Akaike Information criteria) for model development and validationdataset. According to relative ranks of models, the best volume equations areTakata (1958) and Schumacher-Hall (1933) for Taurus cedar. As a result, treevolume can be estimated with high precision using Schumacher-Hall (1933)’sequation for natural cedar stands in this region.

BACKGROUND Intracranial germ cell tumors (iGCTs) are divided into germinomas (GEs) and nongerminomatous germ cell tumors (NGGCTs), each having distinct treatments and prognoses. However, predictive biomarkers for differentiation are limited. Here, we aim to develop a deep learning (DL) model that can automatically differentiate GEs from NGGCTs using preoperative magnetic resonance (MR) images and analyzing how our model improved clinical decision making among five physicians. METHODS: This multicenter study included patients with pathology-confirmed iGCTs for the developed set (total n = 280), which was split into training (n = 200) and testing datasets (n= 80), and three independent validation datasets (n = 64), respectively. We developed a DL model based on 3D nnU-Net and preoperative MR images to differentiate GEs from NGGCTs. To evaluate the clinical value of the DL model, five physicians' accuracy and level of consistency in identifying tumor types were compared in the testing and three validation datasets, with and without referring to the DL model’s output. RESULTS The DL model achieved AUCs of 0.950, 0.921, 0.869, and 0.905 in the testing and three validation datasets, respectively, which is superior to using conventional information including MR features and clinical information (AUC=0.768, P=0.009). The physicians’ diagnostic performance improved from an average AUC of 0.695 without reference to the DL output to 0.842 using the DL output in the testing and independent validation datasets (21.2% improvement, P=0.014). CONCLUSIONS We developed and validated a DL model using a large dataset to accurately differentiate GEs from NGGCTs using preoperative MR images.

To establish a nomogram model to predict early cognitive impairment after supratentorial spontaneous intracranial hematoma in adult patients.A retrospective cohort study was held between January 2016 and October 2018. One hundred twenty seven out of 170 consecutive patients with supratentorial spontaneous intracranial hematoma were enrolled in this study. They were divided into development (n = 92) and validation (n = 35) dataset according to their admission time. Mini-mental State Examination (MMSE) was conducted between the third and the sixth month after the onset of stroke. MMSE ≤ 24 was considered as cognitive impairment. Univariate and multivariate logistic regression was used to screen for independent risk factors which correlate with cognitive impairment on the development dataset. A nomogram was built based on Akaike Information Criterion (AIC). Receiver operating characteristic (ROC) curve and calibration curve on development and validation dataset was drawn with each area under the curves (AUC) calculated. The decision curve analysis was also conducted with the development dataset.The bleeding volume, Glasgow Coma Scale (GCS), and intraventricular hemorrhage (IVH) are the most significant risk factors which may cause cognitive impairment both in the univariate and multivariate analysis. The finial model performed good discrimination ability on both development and validation dataset with AUC 0.911 and 0.919. Most patients would benefit from the model according to the decision curve analysis.A nomogram, constructed based on bleeding volume, GCS, and IVH can provide a feasible tool to evaluate cognitive impairment after supratentorial spontaneous intracranial hematoma in adult patients.

The aim was to develop a multivariable risk prediction model for 30-day mortality following elective abdominal aortic aneurysm (AAA) repair. Data collected prospectively on 2765 consecutive patients undergoing elective open and endovascular AAA repair from September 1999 to October 2009 in the North West of England were split randomly into development (1936 patients) and validation (829) data sets. Logistic regression analysis was undertaken to identify risk factors for 30-day mortality. Ninety-eight deaths (5·1 per cent) were recorded in the development data set. Variables associated with 30-day mortality included: increasing age (P = 0·005), female sex (P = 0·002), diabetes (P = 0·029), raised serum creatinine level (P = 0·006), respiratory disease (P = 0·031), antiplatelet medication (P < 0·001) and open surgery (P = 0·002). The area under the receiver operating characteristic (ROC) curve for predicted probability of 30-day mortality in the development and validation data sets was 0·73 and 0·70 respectively. Observed versus expected 30-day mortality was 3·2 versus 2·0 per cent (P = 0·272) in low-risk, 6·1 versus 5·1 per cent (P = 0·671) in medium-risk and 11·1 versus 10·7 per cent (P = 0·879) in high-risk patients. This multivariable model for predicting 30-day mortality following elective AAA repair can be used clinically to calculate patient-specific risk and is useful for case-mix adjustment. The model predicted well across all risk groups in the validation data set.

IntroductionPatients with sepsis who present to an emergency department (ED) have highly variable underlying disease severity, and can be categorized from low to high risk. Development of a risk stratification tool for these patients is important for appropriate triage and early treatment. The aim of this study was to develop machine learning models predicting 31-day mortality in patients presenting to the ED with sepsis and to compare these to internal medicine physicians and clinical risk scores.MethodsA single-center, retrospective cohort study was conducted amongst 1,344 emergency department patients fulfilling sepsis criteria. Laboratory and clinical data that was available in the first two hours of presentation from these patients were randomly partitioned into a development (n = 1,244) and validation dataset (n = 100). Machine learning models were trained and evaluated on the development dataset and compared to internal medicine physicians and risk scores in the independent validation dataset. The primary outcome was 31-day mortality.ResultsA number of 1,344 patients were included of whom 174 (13.0%) died. Machine learning models trained with laboratory or a combination of laboratory + clinical data achieved an area-under-the ROC curve of 0.82 (95% CI: 0.80–0.84) and 0.84 (95% CI: 0.81–0.87) for predicting 31-day mortality, respectively. In the validation set, models outperformed internal medicine physicians and clinical risk scores in sensitivity (92% vs. 72% vs. 78%;p<0.001,all comparisons) while retaining comparable specificity (78% vs. 74% vs. 72%;p>0.02). The model had higher diagnostic accuracy with an area-under-the-ROC curve of 0.85 (95%CI: 0.78–0.92) compared to abbMEDS (0.63,0.54–0.73), mREMS (0.63,0.54–0.72) and internal medicine physicians (0.74,0.65–0.82).ConclusionMachine learning models outperformed internal medicine physicians and clinical risk scores in predicting 31-day mortality. These models are a promising tool to aid in risk stratification of patients presenting to the ED with sepsis.

It can be challenging in some situations to distinguish primary central nervous system lymphoma (PCNSL) from glioblastoma (GBM) based on magnetic resonance imaging (MRI) scans, especially those involving the corpus callosum. The objective of this study was to assess the diagnostic performance of deep learning (DL) models between PCNSLs and GBMs in corpus callosal tumors. The axial T1-weighted gadolinium-enhanced MRI scans of 274 individuals with pathologically confirmed PCNSL (n = 94) and GBM (n = 180) were examined. After image pooling, pre-operative MRI scans were randomly split with an 80/20 procedure into a training dataset (n = 709) and a testing dataset (n = 177) for DL model development. Therefore, the DL model was deployed as a web application and validated with the unseen images (n = 114) and area under the receiver operating characteristic curve (AUC); other outcomes were calculated to assess the discrimination performance. The first baseline DL model had an AUC of 0.77 for PCNSL when evaluated with unseen images. The 2nd model with ridge regression regularization and the 3rd model with drop-out regularization increased an AUC of 0.83 and 0.84. In addition, the last model with data augmentation yielded an AUC of 0.57. DL with regularization may provide useful diagnostic information to help doctors distinguish PCNSL from GBM.

A comparison of turbine mass flow models based on pragmatic identification data sets for turbogenerator model development

ObjectivePoor sleep quality is prevailing, deleterious, but heterogeneous in older adults. This study aimed to develop a simplified instrument to screen and classify poor sleep quality in community-dwelling older adults, by which stepped care with needs-based interventions could be implemented.MethodsCohorts of adults aged 65 years and older were used to develop the Rapid Classification Scale for Sleep Quality (RCSSQ). Poor sleep quality was defined with the Pittsburgh Sleep Quality Index (PSQI). Established subgroups of poor sleep quality in the development dataset (n = 2622) were used as the criterion standard. Two independent validation datasets (n = 964 and 193, respectively) were used to examine the external validity. Questions in the PSQI were examined by the stepwise multinomial logistic regressions to determine the optimal numbers of items in the RCSSQ. On the premise of item parsimony and instrument validity, the optimal combination of reduced items was determined.ResultsIn the development dataset, the 4-item RCSSQ (RCSSQ-4) was the optimal predictive model. In terms of internal validity, the accuracy rates to identify PSQI-defined poor sleep quality and its subgroups in the developmental dataset by the RCSSQ-4 were 89.0% and 79.9%, respectively. Meanwhile, the RCSSQ-4 also had good external validity in the validation datasets to detect PSQI-defined poor sleep quality (accuracy rates: 89.1–90.7%). Furthermore, the profiles of PSQI component scores and comorbid conditions for the predicted subgroups in the validation dataset were comparable with the criterion standard.ConclusionThe RCSSQ-4 is a valid instrument for screening and subgrouping poor sleep quality in community-dwelling older adults. The RCSSQ-4 may help guide tailored interventions under the context of stepped care in the community.

Purpose—This study was conducted to develop an automated detection algorithm for screening fundus abnormalities, including age-related macular degeneration (AMD), diabetic retinopathy (DR), epiretinal membrane (ERM), retinal vascular occlusion (RVO), and suspected glaucoma among health screening program participants. Methods—The development dataset consisted of 43,221 retinal fundus photographs (from 25,564 participants, mean age 53.38 ± 10.97 years, female 39.0%) from a health screening program and patients of the Kangbuk Samsung Hospital Ophthalmology Department from 2006 to 2017. We evaluated our screening algorithm on independent validation datasets. Five separate one-versus-rest (OVR) classification algorithms based on deep convolutional neural networks (CNNs) were trained to detect AMD, ERM, DR, RVO, and suspected glaucoma. The ground truth for both development and validation datasets was graded at least two times by three ophthalmologists. The area under the receiver operating characteristic curve (AUC), sensitivity, and specificity were calculated for each disease, as well as their macro-averages. Results—For the internal validation dataset, the average sensitivity was 0.9098 (95% confidence interval (CI), 0.8660–0.9536), the average specificity was 0.9079 (95% CI, 0.8576–0.9582), and the overall accuracy was 0.9092 (95% CI, 0.8769–0.9415). For the external validation dataset consisting of 1698 images, the average of the AUCs was 0.9025 (95% CI, 0.8671–0.9379). Conclusions—Our algorithm had high sensitivity and specificity for detecting major fundus abnormalities. Our study will facilitate expansion of the applications of deep learning-based computer-aided diagnostic decision support tools in actual clinical settings. Further research is needed to improved generalization for this algorithm.