Higher Area Under The Curve Values Research Articles

Deep multiscale convolutional feature learning for intracranial hemorrhage classification and weakly supervised localization

ObjectiveThis study evaluated the performance of attentional fusion model-based multiscale features in classifying intracerebral hemorrhage and the localization of bleeding focus based on weakly supervised target localization. MethodsA publicly available dataset provided by the American College of Neuroradiology (ASNR) was used, consisting of 750,000 computed tomography (CT) scans of the brain, manually marked by radiologists for intracranial hemorrhage and five hemorrhage subtypes. A multiscale feature classification and weakly supervised localization framework based on an attentional fusion mechanism were applied, which could be annotated at the slice level and provided intracranial hemorrhage classification and hemorrhage focus localization. ResultsThe designed framework achieved excellent performance for classification and localization. The area under the curve (AUC) for predicting bleeding was 0.973. High AUC values were observed for the five hemorrhage subtypes (epidural AUC = 0.891, subdural AUC = 0.991, subarachnoid AUC = 0.983, intraventricular AUC = 0.995, intraparenchymal AUC = 0.990). This model outperformed the average entry-level radiology trainee compared to previously reported data. ConclusionThe designed method quickly and accurately detected intracerebral hemorrhage, classifying hemorrhage subtypes and locating bleeding points with image-level annotation alone. The results indicate that this framework can significantly reduce diagnostic time while improving the detection of intracerebral hemorrhage in emergencies. It can thus be integrated into the diagnostic radiology workflow in the future.

Association between serum creatinine to albumin ratio and short- and long-term all-cause mortality in patients with acute pancreatitis admitted to the intensive care unit: a retrospective analysis based on the MIMIC-IV database.

Serum creatinine (Cr) and albumin (Alb) are important predictors of mortality in individuals with various diseases, including acute pancreatitis (AP). However, most previous studies have only examined the relationship between single Cr or Alb levels and the prognosis of patients with AP. To our knowledge, the association between short- and long-term all-cause mortality in patients with AP and the blood creatinine to albumin ratio (CAR) has not been investigated. Therefore, this study aimed to evaluate the short- and long-term relationships between CAR and all-cause mortality in patients with AP. We conducted a retrospective study utilizing data from the Medical Information Market for Intensive Care (MIMIC-IV) database. The study involved analyzing various mortality variables and obtaining CAR values at the time of admission. The X-tile software was used to determine the optimal threshold for the CAR. Kaplan-Meier (K-M) survival curves and multivariate Cox proportional hazards regression models were used to assess the relationship between CAR and both short- and long-term all-cause mortality. The predictive power, sensitivity, specificity, and area under the curve (AUC) of CAR for short- and long-term mortality in patients with AP after hospital admission were investigated using Receiver Operating Characteristic analysis. Additionally, subgroup analyses were conducted. A total of 520 participants were included in this study. The CAR ideal threshold, determined by X-tile software, was 0.446. The Cox proportional hazards model revealed an independent association between CAR≥0.446 and all-cause mortality at 7-day (d), 14-d, 21-d, 28-d, 90-d, and 1-year (y) before and after adjustment for confounders. K-M survival curves showed that patients with CAR≥0.446 had lower survival rates at 7-d, 14-d, 21-d, 28-d, 90-d, and 1-y. Additionally, CAR demonstrated superior performance, with higher AUC values than Cr, Alb, serum total calcium, Glasgow Coma Scale, Systemic Inflammatory Response Syndrome score, and Sepsis-related Organ Failure Assessment score at 7-d, 14-d, 21-d, 28-d, 90-d, and 1-y intervals. Subgroup analyses showed that CAR did not interact with a majority of subgroups. The CAR can serve as an independent predictor for short- and long-term all-cause mortality in patients with AP. This study enhances our understanding of the association between serum-based biomarkers and the prognosis of patients with AP.

Construction of a prognostic model for hepatocellular carcinoma patients receiving transarterial chemoembolization treatment based on the Tumor Burden Score

BackgroundPatients with hepatocellular carcinoma (HCC) who undergo transarterial chemoembolization (TACE) may have varied outcomes based on their liver function and tumor burden diversity. This study aims to assess the prognostic significance of the tumor burden score (TBS) in these patients and develop a prognostic model for their overall survival.MethodsThe study involved a retrospective analysis of 644 newly diagnosed HCC patients undergoing TACE treatment. The individuals were assigned randomly to a training cohort (n = 452) and a validation cohort (n = 192). We utilized a multivariate Cox proportional risk model to identify independent preoperative predictive factors. We then evaluated model performance using the area under the curve (AUC), consistency index (c-index), calibration curve, and decision curve analysis (DCA) methods.ResultsThe multivariate analysis revealed four prognostic factors associated with overall survival: Tumor Burden Score, Tumor Extent, Types of portal vein invasion (PVI), and Child-Pugh score. The total score was calculated based on these factors. The model demonstrated strong discriminative ability with high AUC values and c-index, providing high net clinical benefits for patients. Based on the model’s scoring results, patients were categorized into high, medium, and low-risk groups. These results were validated in the validation cohort.ConclusionsThe tumor burden score shows promise as a viable alternative prognostic indicator for assessing tumor burden in cases of HCC. The new prognostic model can place patients in one of three groups, which will estimate their individual outcomes. For high-risk patients, it is suggested to consider alternative treatment options or provide the best supportive care, as they may not benefit significantly from TACE treatment.

Clinical application of machine learning-based pathomics signature of gastric atrophy.

The diagnosis of gastric atrophy is highly subjective, and we aimed to establish a model of gastric atrophy based on pathological features to improve diagnostic consistency. We retrospectively collected the HE-stained pathological slides of gastric biopsies and used CellProfiler software for image segmentation and feature extraction of ten representative images for each sample. Subsequently, we employed the Least absolute shrinkage and selection operator (LASSO) to select features and different machine learning (ML) algorithms to construct the diagnostic models for gastric atrophy. We selected 289 gastric biopsy specimens for training, testing, and external validation. We extracted 464 pathological features and screened ten features by LASSO to establish the diagnostic model for moderate-to-severe atrophy. The range of area under the curve (AUC) for various machine learning algorithms was 0.835-1.000 in the training set, 0.786-0.949 in the testing set, and 0.689-0.818 in the external validation set. LR model had the highest AUC value, with 0.900 (95% CI: 0.852-0.947) in the training set, 0.901 (95% CI: 0.807-0.996) in the testing set, and 0.818 (95% CI: 0.714-0.923) in the external validation set. The atrophy pathological score based on the LR model was associated with endoscopic atrophy grading (Z=-2.478, P=0.013) and gastric cancer (GC) (OR=5.70, 95% CI: 2.63-12.33, P<0.001). The ML model based on pathological features could improve the diagnostic consistency of gastric atrophy, which is also associated with endoscopic atrophy grading and GC.

Diffusion-Weighted MRI-Based Virtual Elastography and Shear-Wave Elastography for the Assessment of Breast Lesions.

Diffusion-weighted imaging (DWI)-based virtual MR elastography (DWI-vMRE) in the assessment of breast lesions is still in the research stage. To investigate the usefulness of elasticity values on DWI-vMRE in the evaluation of breast lesions, and the correlation with the values calculated from shear-wave elastography (SWE). Prospective. 153 patients (mean age ± standard deviation: 55 ± 12 years) with 153 pathological confirmed breast lesions (24 benign and 129 malignant lesions). 1.5-T MRI, multi-b readout segmented echo planar imaging (b-values of 0, 200, 800, and 1000 sec/mm2 ). For DWI-vMRE assessment, lesions were manually segmented using apparent diffusion coefficient (ADC0-1000 ) map, then the region of interests were copied to the map of shifted-ADC (sADC200-800 , sADC 200-1500 ). For SWE assessment, the shear modulus of the lesions was measured by US elastic modulus (μUSE ). Intraclass/interclass kappa coefficients were calculated to measure the consistency. Pearson's correlation was used to assess the relationship between sADC and μUSE . A receiver operating characteristic analysis with the area under the curve (AUC) was performed to compare the diagnostic accuracy between benign and malignant breast lesions of sADC and μUSE . A P value <0.05 was considered statistically significant. There were significant differences between benign and malignant breast lesions of μUSE (24.17 ± 10.64 vs. 37.20 ± 12.61), sADC200-800 (1.38 ± 0.31 vs. 0.97 ± 0.18 × 10-3 mm2 /sec), and sADC200-1500 (1.14 ± 0.30 vs. 0.78 ± 0.13 × 10-3 mm2 /sec). In all breast lesions, a moderate but significant correlation was observed between μUSE and sADC200-800 /sADC200-1500 (r = -0.49/-0.44). AUC values to differentiate benign from malignant lesions were as follows: μUSE , 0.78; sADC200-800 , 0.89; sADC200-1500 , 0.89. Both SWE and DWI-vMRE could be used for the differentiation of benign versus malignant breast lesions. Furthermore, DWI-vMRE with the use of sADC show relatively higher AUC values than SWE. 4 TECHNICAL EFFICACY: Stage 2.

Preoperative frailty assessment with ultrasound in elderly patients scheduled for total knee arthroplasty.

Frailty, a decline in physical and cognitive reserve capacity, renders patients susceptible to various stressors and has been linked to adverse outcomes and increased healthcare utilization. This study aimed to determine whether ultrasound measurements of the rectus abdominis (RA) and biceps brachii (BB) could predict frailty in patients scheduled for total knee arthroplasty. Frailty was assessed using the Clinical Frailty Scale in adults aged ≥60 years. Ultrasound measurements of the rectus abdominis, BB, and quadriceps femoris muscles, along with thigh circumference measurements, were obtained before surgery. The predictive ability of the unadjusted and BMI- and body surface area (BSA)-adjusted measurements were evaluated using receiver operating characteristic curve analysis and area under the curve (AUC) values. Postoperative outcomes, such as admission to the intensive care unit or skilled nursing facility, delirium, falls, re-hospitalization, and 30-day mortality were recorded. We analyzed data from 148 patients. BB thickness provided a fair prediction of frailty. Average measurements of both BB adjusted for BMI (0.708, 95% CI 0.602-0.814; P<0.001), and BSA (0.708, 95% CI 0.598-0.817; P<0.001) had the highest AUC values. RA muscle measurements could not discriminate frailty. The BMI-adjusted measurements for: right quadriceps femoris thickness (AUC 0.614, 95% CI 0.503-0.725; P=0.044), left thigh circumference (AUC 0.648, 95% CI 0.528-0.769; P=0.016), and average thigh circumference (AUC 0.630, 95% CI 0.511-0.750; P=0.033) had statistically significant but poor AUC values. Preoperative ultrasound measurements of the bilateral BB can fairly predict frailty in patients scheduled for total knee arthroplasty.

18F]SynVesT-1 and [18F]FDG quantitative PET imaging in the presurgical evaluation of MRI-negative children with focal cortical dysplasia type II.

MRI-negative children with focal cortical dysplasia type II (FCD II) are one of the most challenging cases in surgical epilepsy management. We aimed to utilize quantitative positron emission tomography (QPET) analysis to complement [18F]SynVesT-1 and [18F]FDG PET imaging and facilitate the localization of epileptogenic foci in pediatric MRI-negative FCD II patients. We prospectively enrolled 17 MRI-negative children with FCD II who underwent [18F]SynVesT-1 and [18F]FDG PET before surgical resection. The QPET scans were analyzed using statistical parametric mapping (SPM) with respect to healthy controls. The sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and area under the curve (AUC) of [18F]SynVesT-1 PET, [18F]FDG PET, [18F]SynVesT-1 QPET, and [18F]FDG QPET in the localization of epileptogenic foci were assessed. Additionally, we developed a multivariate prediction model based on dual trace PET/QPET assessment. The AUC values of [18F]FDG PET and [18F]SynVesT-1 PET were 0.861 (sensitivity = 94.1%, specificity = 78.2%, PPV = 38.1%, NPV = 98.9%) and 0.908 (sensitivity = 82.4%, specificity = 99.2%, PPV = 93.3%, NPV = 97.5%), respectively. [18F]FDG QPET showed lower sensitivity (76.5%) and NPV (96.6%) but higher specificity (95.0%) and PPV (68.4%) than visual assessment, while [18F]SynVesT-1 QPET exhibited higher sensitivity (94.1%) and NPV (99.1%) but lower specificity (97.5%) and PPV (84.2%). The multivariate prediction model had the highest AUC value (AUC = 0.996, sensitivity = 100.0%, specificity = 96.6%, PPV = 81.0%, NPV = 100%). The multivariate prediction model based on [18F]SynVesT-1 and [18F]FDG PET/QPET assessments holds promise in noninvasively identifying epileptogenic regions in MRI-negative children with FCD II. Furthermore, the combination of visual assessment and QPET may improve the sensitivity and specificity of diagnostic tests in localizing epileptogenic foci and achieving a preferable surgical outcome in MRI-negative FCD II.

Predicting intraoperative hemorrhage during curettage treatment of cesarean scar pregnancy using free-breathing GRASP DCE-MRI

ObjectiveTo explore the feasibility of the golden-angle radial sparse parallel (GRASP) dynamic magnetic resonance imaging (MRI) technique in predicting the intraoperative bleeding risk of scar pregnancy.MethodsA total of 49 patients with cesarean scar pregnancy (CSP) who underwent curettage and GRASP-MRI imaging were retrospectively selected between January 2021 and July 2022. The pharmacokinetic parameters, including Wash-in, Wash-out, time to peck (TTP), initial area under the curve (iAUC), the transfer rate constant (Ktrans), constant flow rate (Kep), and volume of extracellular space (Ve), were calculated. The amount of intraoperative bleeding was recorded by a gynecologist who performed surgery, after which patients were divided into non-hemorrhage (blood loss ≤ 200 mL) and hemorrhage (blood loss > 200 mL) groups. The measured pharmacokinetic parameters were statistically compared using the t-test or Mann–Whitney U test with a significant level set to be p < 0.05. The receiver operating characteristic (ROC) curve was constructed, and the area under the curve (AUC) was calculated to evaluate each parameter’s capability in intraoperative hemorrhage subgroup classification.ResultsTwenty patients had intraoperative hemorrhage (blood loss > 200 mL) during curettage. The hemorrhage group had larger Wash-in, iAUC, Ktrans, Ve, and shorter TTP than the non-hemorrhage group (all P > 0.05). Wash-in had the highest AUC value (0.90), while Ktrans had the lowest value (0.67). Wash-out and Kep were not significantly different between the two groups.ConclusionGRASP DCE-MRI has the potential to forecast intraoperative hemorrhage during curettage treatment of CSP, with Wash-in exhibiting the highest predictive performance. This data holds promise for advancing personalized treatment. However, further study is required to compare its effectiveness with other risk factors identified through anatomical MRI and ultrasound.

Distribution of denitrifiers predicted by correlative niche modeling of changing environmental conditions and future climatic scenarios across the Baltic Sea

Denitrifying microbial communities provide an important ecosystem function in aquatic systems. Yet, knowledge on predictive and modeling of these complex and changing communities is limited. The emergently challenging question of how the geographical distribution of denitrifiers responds to ongoing and future environmental change is not yet fully understood. In our study we use metadata-based correlative niche modeling to analyze the geographical distribution of selected putative denitrifiers in the genus Sphingomonas, Mycoplana, Shewanella, and Alteromonas at different predicted environmental conditions and future climatic scenarios across the Baltic Sea. Using the predictive power of an ensemble modeling approach and eight different machine-learning algorithms, habitat suitability and the distribution of the selected denitrifiers were evaluated using geophysical and bioclimatic variables, benthic conditions, and four Representative Concentration Pathway (RCP) trajectories of future global warming scenarios. All algorithms provided successful prediction capabilities both for variable importance, and for habitat suitability with Area Under the Curve (AUC) values between 0.89 and 1.00. Model findings revealed that salinity and nitrate concentrations significantly explained the variation in distribution of the selected denitrifiers. Rising temperatures of 0.8 to 1.8 °C at future RCP60–2050 trajectories are predicted to diminish or eliminate the bioclimatic suitable habitats for denitrifier distributions across the Baltic Sea. Multi-collated terrestrial and marine environmental variables contributed to the successful prediction of denitrifier distributions within the study area. The correlative niche modeling approach with high AUC values presented in the study allowed for accurate projections of the future distributions of the selected denitrifiers. The modeling approach can be used to improve our understanding of how ongoing and predicted future environmental changes may affect habitat suitability for organisms with denitrification capacity across the Baltic Sea.

Landslide susceptibility prediction using machine learning and remote sensing: Case study in Thua Thien Hue province, Vietnam

Landslides lead to widespread devastation and significant loss of life in mountainous regions around the world. Susceptibility assessments can provide critical data to help decision‐makers, for example, local authorities and other organizations, mitigating the landslide risk, although the accuracy of existing studies needs to be improved. This study aims to assess landslide susceptibility in the Thua Thien Hue province of Vietnam using deep neural networks (DNNs) and swarm‐based optimization algorithms, namely Adam, stochastic gradient descent (SGD), Artificial Rabbits Optimization (ARO), Tuna Swarm Optimization (TSO), Sand Cat Swarm Optimization (SCSO), Honey Badger Algorithm (HBA), Marine Predators Algorithm (MPA) and Particle Swarm Optimization (PSO). The locations of 945 landslides occurring between 2012 and 2022, along with 14 conditioning factors, were used as input data to build the DNN and DNN‐hybrid models. The performance of the proposed models was evaluated using the statistical indices receiver operating characteristic curve, area under the curve (AUC), root mean square error, mean absolute error (MAE), R2 and accuracy. All proposed models had a high accuracy of prediction. The DNN‐MPA model had the highest AUC value (0.95), followed by DNN‐HBA (0.95), DNN‐ARO (0.95), DNN‐Adam (0.95), DNN‐SGD (0.95), DNN‐TSO (0.93), DNN‐PSO (0.9) and finally DNN‐SCSO (0.83). High‐precision models have identified that the majority of the western region of Thua Thien Hue province is very highly susceptible to landslides. Models like the aforementioned ones can support decision‐makers in updating large‐scale sustainable land‐use strategies.

Value of disc topographic measurements for the discrimination of glaucoma suspects from early primary open angle glaucoma cases: A thin discriminative line

PurposeTo determine the diagnostic value of optic nerve head (ONH) topographic parameters measured by swept-source optical coherence tomography (SS-OCT) in differentiating between early glaucoma cases, glaucoma suspects (GSs), and healthy eyes. MethodsThe files of GSs, those with an early primary open angle glaucoma (POAG) diagnosis, and healthy eyes were retrospectively screened. Demographic characteristics, retinal nerve fibre measurements, visual field examinations, and ONH topographic parameters, including rim area, disc area, vertical and horizontal cup-to-disc ratios (CDRs), and cup volume, were compared between the groups. A receiver operating characteristic (ROC) curve analysis was performed on the ONH parameters. ResultsA total of 170 eyes from 85 GSs, 114 eyes from 57 patients with a diagnosis of early POAG, and 70 healthy eyes from 48 subjects were included in the study. The median age was 52.1 ± 0.9 years for the POAG group, 60.2 ± 1.1 years for the GS group, and 60.5 ± 1.6 for the controls. In the comparison of the ONH parameters between the groups, only rim area statistically significantly differed between the POAG and GS groups. However, all ONH parameters statistically significantly differed between the GS and control groups (p< 0.05). The mean rim areas of the GS, POAG, and control groups were 1.278 ± 0.055 mm2, 1.073 ± 0.065 mm2, and 1.446 ± 0.055 mm2, respectively, being statistically significantly higher in the controls and lower in the POAG group (p < 0.001). The highest area under the curve (AUC) value belonged to vertical CDR for the discrimination of the eyes with POAG from healthy eyes [0.806 (0.728–0.869, p < 0.001] and rim area for the discrimination of POAG cases and GSs [0.728 (0.650–0.797, p < 0.001]. ConclusionThere are significant differences in all ONH parameters between the eyes of GSs and healthy eyes. For the discrimination of early POAG and GS cases, topographic disc parameters may have value. In this study, rim area significantly differed between the GS and POAG groups, and it had the highest AUC value amongst the evaluated ONH parameters in the discrimination of early POAG cases S GSs.

Parametric maps of spatial two-tissue compartment model for prostate dynamic contrast enhanced MRI - comparison with the standard tofts model in the diagnosis of prostate cancer.

The spatial two-tissue compartment model (2TCM) was used to analyze prostate dynamic contrast enhanced (DCE) MRI data and compared with the standard Tofts model. A total of 29 patients with biopsy-confirmed prostate cancer were included in this IRB-approved study. MRI data were acquired on a Philips Achieva 3T-TX scanner. After T2-weighted and diffusion-weighted imaging, DCE data using 3D T1-FFE mDIXON sequence were acquired pre- and post-contrast media injection (0.1 mmol/kg Multihance) for 60 dynamic scans with temporal resolution of 8.3s/image. The 2TCM has one fast ([Formula: see text] and [Formula: see text]) and one slow ([Formula: see text] and [Formula: see text]) exchanging compartment, compared with the standard Tofts model parameters (Ktrans and kep). On average, prostate cancer had significantly higher values (p < 0.01) than normal prostate tissue for all calculated parameters. There was a strong correlation (r = 0.94, p < 0.001) between Ktrans and [Formula: see text] for cancer, but weak correlation (r = 0.28, p < 0.05) between kep and [Formula: see text]. Average root-mean-square error (RMSE) in fits from the 2TCM was significantly smaller (p < 0.001) than the RMSE in fits from the Tofts model. Receiver operating characteristic (ROC) analysis showed that fast [Formula: see text] had the highest area under the curve (AUC) than any other individual parameter. The combined four parameters from the 2TCM had a considerably higher AUC value than the combined two parameters from the Tofts model. The 2TCM is useful for quantitative analysis of prostate DCE-MRI data and provides new information in the diagnosis of prostate cancer.

Ultrasound-based radiomic analysis of the peripheral nerves for differentiation between CIDP and POEMS syndrome.

Demyelinating peripheral neuropathy is characteristic of both polyneuropathy, organomegaly, endocrinopathy, M-protein, and skin changes (POEMS) syndrome and chronic inflammatory demyelinating polyneuropathy (CIDP). We hypothesized that the different pathogeneses underlying these entities would affect the sonographic imaging features. To investigate whether ultrasound (US)-based radiomic analysis could extract features to describe the differences between CIDP and POEMS syndrome. In this retrospective study, we evaluated nerve US images from 26 with typical CIDP and 34 patients with POEMS syndrome. Cross-sectional area (CSA) and echogenicity of the median and ulnar nerves were evaluated in each US image of the wrist, forearm, elbow, and mid-arm. Radiomic analysis was performed on these US images. All radiomic features were examined using receiver operating characteristic analysis. Optimal features were selected using a three-step feature selection method and were inputted into XGBoost to build predictive machine-learning models. The CSAs were more enlarged in patients with CIDP than in those with POEMS syndrome without significant differences, except for that of the ulnar nerve at the wrist. Nerve echogenicity was significantly more heterogeneous in patients with CIDP than in those with POEMS syndrome. The radiomic analysis yielded four features with the highest area under the curve (AUC) value of 0.83. The machine-learning model showed an AUC of 0.90. US-based radiomic analysis has high AUC values in differentiating POEM syndrome from CIDP. Machine-learning algorithms further improved the discriminative ability.

Clinical and inflammation marker features of cancer patients with COVID-19: data of Istanbul, Turkey multicenter cancer patients (2020–2022)

Objective We aimed to identify a rapid, accurate, and accessible biomarker in the early stages of COVID-19 that can determine the prognosis of the disease in cancer patients. Methods A total number of 241 patients with solid cancers who had a COVID-19 diagnosis between March 2020 and February 2022 were included in the study. Factors and ten different markers of inflammation were analyzed by year of diagnosis of COVID-19 and grouped by severity of infection. Results Hospitalization, referral to the intensive care unit (ICU), mechanical ventilation, and death were more frequent in 2020 than in 2021 and 2022 (mortality rates, respectively, were 18.8%, 3.8%, and 2.5%). Bilateral lung involvement and chronic lung disease were independent risk factors for severe disease in 2020. In 2021–2022, only bilateral lung involvement was found as an independent risk factor for severe disease. The neutrophil-to-lymphocyte platelet ratio (NLPR) with the highest area under the curve (AUC) value in 2020 had a sensitivity of 71.4% and specificity of 73.3% in detecting severe disease (cut-off > 0.0241, Area Under the Curve (AUC) = 0.842, p <.001). In 2021–2022, the sensitivity of the C-reactive protein-to-lymphocyte ratio (CRP/L) with the highest AUC value was 70.0%, and the specificity was 73.3% (cut-off > 36.7, AUC = 0.829, p = .001). Conclusions This is the first study to investigate the distribution and characteristics of cancer patients, with a focus on the years of their COVID-19 diagnosis. Based on the data from our study, bilateral lung involvement is an independent factor for severe disease, and the CRP/L inflammation index appears to be the most reliable prognostic marker.

A Comparative Study of Sentiment Analysis Methods for Detecting Fake Reviews in E-Commerce

The popularity of the e-commerce system has increased, especially under the COVID scenario. Consumer product reviews from the past have had a significant impact on influencing consumers' purchasing decisions. Fake reviews—those written by humans and computers that engage in dishonest behavior—are consequently generated to increase product sales. The fake reviews hurt consumers and are dishonest. The goal of this research is to examine and evaluate the performance of various methods for identifying fake reviews. The well-known and widely-used Amazon Review Data (2018) dataset was used for this research. The first 10 product categories on Amazon.com with favorable feedback will be provided in the data section. After that, perform fundamental data preparation procedures such as special character trimming, bag of words, TF-IDF, etc. The models are trained to create a dataset for detecting fake reviews. This research compares the performance of four different models: GPT-2, NBSVM, BiLSTM, and RoBERTa. The hyperparameters of the models are also tuned to find the optimal values. The research concludes that the RoBERTa model performs the best overall, with an accuracy of 97%. GPT-2 has an overall accuracy of 82%, NBSVM has an overall accuracy of 95%, and BiLSTM has an overall accuracy of 92%. The research also calculates the Area Under the Curve (AUC) for each model and finds that RoBERTa has an AUC of 0.9976, NBSVM has an AUC of 0.9888, BiLSTM has an AUC of 0.9753, and GPT-2 has an AUC of 0.9226. It can be observed that the RoBERTa model has the highest AUC value, which is close to 1. Therefore, it can be concluded that this model provides the most accurate prediction for detecting fake reviews, which is the main focus of this research. Doi: 10.28991/HIJ-2023-04-02-08 Full Text: PDF

Preoperative prediction of vessel invasion in locally advanced gastric cancer based on computed tomography radiomics and machine learning.

Vessel invasion (VI) is an important factor affecting the prognosis of gastric cancer (GC), and the accurate determination of preoperative VI for locally advanced GC is of great clinical significance. Traditional methods for the evaluation of VI require postoperative pathological examination. Noninvasive preoperative evaluation of VI is therefore crucial to determine the best treatment strategy. To determine the value of preoperative prediction of gastric VI based on portal venous phase computed tomography (CT) radiomic features and machine-learning models, a retrospective analysis of 296 patients with locally advanced GC confirmed through pathological examination was performed. They were divided into two groups, VI+ (n=213) and VI- (n=83), based on pathological results. Using pyradiomics to extract two-dimensional radiomic features of the portal venous stage of locally advanced GC, data were divided into training (n=207) and validation sets (n=89), with a ratio of 7:3, and three feature selection methods were cascaded and merged. Finally, least absolute shrinkage and selection operator (LASSO) regression was used for feature screening to obtain the optimal feature subset. Four current representative machine-learning algorithms were used to construct the prediction model, the receiver operating characteristic curve was constructed to evaluate the predictive performance of the model, and the area under the curve (AUC), accuracy, sensitivity, and specificity were calculated. The differentiation degree, and the Lauren's and CA199 classifications were independent risk factors for locally advanced GC VI. Pyradiomics extracted 864 quantitative features of portal vein images of locally advanced GC. After filtering out low variance features using R, 236 features remained. Next, 18 features were screened using the LASSO algorithm. Extreme gradient boosting (XGBoost), logistic regression, Gaussian naive Bayes, and support vector machine models were constructed based on the 18 best features screened out of the portal venous CT images of advanced GC and three independent risk factors of GC VI in clinical features predicted the training set AUC values of 0.914, 0.897, 0.880, and 0.814, respectively. The predicted validation set AUC values were 0.870, 0.877, 0.859, and 0.773, respectively. The DeLong test results indicated no statistically significant difference in AUC values between the XGBoost and logistic regression models in the training and validation sets. The four machine-learning models showed high predictive performance. The logistic regression model had the highest AUC value in the validation set (0.877), and the accuracy and F1 score were 77 and 87.6%, respectively. CT radiomic features and machine-learning models based on the portal venous phase can be used as a noninvasive imaging method for the preoperative prediction of VI in locally advanced GC. The logistic regression model exhibited the highest diagnostic performance.

Corneal layer thickness in keratoconus using optical coherence tomography

ABSTRACT Clinical relevance Accurate thickness measurement of corneal layers using anterior segment OCT can be used to improve visual outcomes. Understanding its applications is essential for optometric practices to enhance eye care procedures. Background To evaluate the thicknesses of different corneal layers for identifying keratoconus (KCN) and subclinical keratoconus (SKCN) using spectral-domain optical coherence tomography (SD-OCT). Methods This prospective study analyzed 60 eyes with KCN, 48 eyes with SKCN, and 53 normal eyes. The central corneal thickness (CCT) and thicknesses of the epithelium, Bowman, stroma, and Descemet-endothelium layers were measured using SD-OCT. One way analysis of variance and the area under the curve (AUC) were used to evaluate the parameters. The Delong method was used to compare AUCs. Results In KCN, CCT and thicknesses of epithelium, Bowman, stroma, and Descemet-endothelium layers were 495.5 ± 41.7, 52.6 ± 6.4,11.5 ± 1.4, 415.5 ± 38.9, and 12.3 ± 1.7 µm, respectively. These thickness values were respectively 524.5 ± 33.3, 56.8 ± 6.8, 11.5 ± 1.6, 439.8 ± 30.6, and 12.4 ± 1.7 µm in SKCN and 563.8 ± 37.9, 57.7 ± 6.9, 12.2 ± 1.6, 469.5 ± 33.7, and 12.8 ± 2.1µm in normal group. Total cornea and stroma in KCN and SKCN, and epithelium in KCN were significantly thinner compared to the normal group (P < 0.001). The highest AUC values were observed for CCT in KCN (AUC 0.90) and SKCN (AUC 0.782). The diagnostic accuracy was significantly higher for stromal thickness in KCN (sensitivity 81.7%, specificity 73.6%, AUC 0.871) and SKCN (sensitivity 80.0%, specificity 56.6%, AUC 0.751) than other individual corneal layers (Delong, P < 0.001) . Conclusion CCT can accurately distinguish keratoconus from normal eyes. However, central corneal stromal thinning was the most sensitive diagnostic index for early detection of SKCN. Developing standardized stromal maps may be helpful for detecting SKCN.

Assessing the accuracy of sensitivity analysis: an application for a cellular automata model of Bogota’s urban wetland changes

This study analyzes the outcomes of Cellular Automata (CA) with different neighborhood sizes and spatial resolution configurations on the performance of the Future Land Use Simulation (FLUS) model. The analysis is executed using three analogic images to extract the land use/land cover in Bogota, Colombia, for three years: 1998, 2004, and 2010. The FLUS model has an Artificial Neuronal Network model, which was used for calculating the relationships between the land uses and the associated drivers and to estimate the probability of occurrence of each land use. Whenever a CA is used to model and simulate, sensitivity analysis (SA) becomes a crucial step in CA modeling to understand better the influence of parameters’ changes in the simulation outcomes. Therefore, the SA is conducted by varying the neighborhood sizes between 3 × 3, 5 × 5, and 7 × 7 for 5 and 30 meters. In addition, cross-classification maps, Area Under the Curve (AUC) of the Total Operating Characteristic, landscape metrics, the figure of merit, Fuzzy Kappa, and disagreement metrics were calculated to assess how well the model performed. High AUC values and low disagreement results show that, in general, the model performed well, and the accuracy of the outputs improves with a 3 × 3 neighborhood size and 5 meters spatial resolution. This study provides a broad assessment approach to the different methods that must be considered to evaluate the sensitivity of CA models in the simulation of urban wetlands’ spatial-temporal evolution.

Abstract Number ‐ 51: Automated Versus Human Hyperdense Vessel Sign Detection Using Non‐Contrast Computed Tomography Scans

Introduction Rapid detection of large vessel occlusion (LVO) is very crucial in triaging stroke patients potentially candidates for mechanical thrombectomy (MT). Hyperdense vessel sign (HDVS) is one of the earliest ischemic changes in non‐contrast CT scan (NCCT) indicating LVO stroke. Artificial intelligence emerged to detect HDVS with the advantages of faster acquisition, less variation, and a lower need for experience than the usual detection. We aimed to identify the diagnostic performance of automated software (e‐Stroke, Brainomix) in HDVS detection. Methods A prospectively collectedMT database from March 2020 to August 2021 was reviewed. Patients were included if they had intracranial internal carotid artery or middle cerebral artery M1 or M2 occlusion. Cases with HDVS were identified through the routine 2.5‐mm slice thickness NCCT scans after being correlated with patients’ clinical information and confirmed with CT angiography (CTA) scans. NCCT scans were classified according to slice thickness into two groups: 2.5‐mm scans and 0.625‐mm generated scans. All NCCT scans were read by e‐Stroke software, then deidentified and reviewed by two stroke neurologists who were blinded to any clinical, other imaging, or therapeutic information. They were required to record the presence/laterality of HDVS before and after observing other NCCT early ischemic changes like gaze deviation, loss of insular ribbon, caudate or lentiform hypodensity. ROC curve analysis was used to estimate sensitivity and specificity and the area under the curve (AUC) was compared using DeLong’s test. Inter‐rater agreement between the two readers’ final reads, e‐Stroke, and the standard read was measured using the Fleiss Kappa test. Results Among 304 patients included in the study, 37.7% had HDVS. Approximately 44% of the scans had 2.5‐mm slice thickness and 56% had 0.625‐mm slice thickness. The e‐Stroke software identified HDVS with a sensitivity of 63% and a specificity of 71% (Table 1). The mean AUC value of e‐Stroke HDVS detection (0.67[0.61‐0.74]) was similar to reader‐1 (0.68[0.62‐0.74];p = 0.87) and reader‐2 (0.63[0.57‐0.70];p = 0.56). HDVS detection improved by reader‐1(0.78[0.72‐0.83];p = 0.03) after observing other early ischemic changes on the same scans, but reader‐2 performance remained similar to e‐Stroke (0.69[0.63‐0.76];p = 0.71). AUC, sensitivity and specificity ofHDVS detection by e‐Stroke were significantly higher using 2.5‐mm compared to 0.625‐mm sliced NCCT scans (0.78[0.70‐0.86],sensitivity 70%,specificity 86%;p&lt; 0.001) vs (0.58[0.50‐0.67],sensitivity 56%,specificity 61%;p = 0.06) respectively;p = 0.01. The readers also had higher AUC values with 2.5‐mm scans but not statistically significant, (0.74[0.66‐0.83] vs 0.64[0.56‐0.73];p = 0.18) for reader‐1 and (0.68[0.59‐0.77] vs 0.57[0.48‐0.66];p = 0.23) for reader‐2. The same after the final read, (0.85[0.78‐0.92] vs 0.75[0.67‐0.82];p = 0.08) for reader‐1 and (0.73[0.65‐0.82] vs 0.67[0.58‐0.76];p = 0.43) for reader‐2. Similarly, inter‐rater agreement was higher using 2.5‐mm sliced scans, k = 0.50(0.43‐0.75) compared to0.625‐mm scans,k = 0.27(0.21‐0.33). Conclusions Artificial intelligence (e‐Stroke software) has comparable sensitivity and specificity to human readers in HDVS detection. For e‐Stroke software, 2.5‐mm sliced CT scans are better to identifyHDVS compared to 0.625‐mm scans.

Classification of multiple sclerosis women with voiding dysfunction using machine learning: Is functional connectivity or structural connectivity a better predictor?

Machine learning (ML) is an established technique that uses sets of training data to develop algorithms and perform data classification without using human intervention/supervision. This study aims to determine how functional and anatomical brain connectivity (FC and SC) data can be used to classify voiding dysfunction (VD) in female MS patients using ML. Twenty-seven ambulatory MS individuals with lower urinary tract dysfunction were recruited and divided into two groups (Group 1: voiders [V, n= 14]; Group 2: VD [n= 13]). All patients underwent concurrent functional MRI/urodynamics testing. Best-performing ML algorithms, with highest area under the curve (AUC), were partial least squares (PLS, AUC =0.86) using FC alone and random forest (RF) when using SC alone (AUC =0.93) and combined (AUC =0.96) as inputs. Our results show 10 predictors with the highest AUC values were associated with FC, indicating that although white matter was affected, new connections may have formed to preserve voiding initiation. MS patients with and without VD exhibit distinct brain connectivity patterns when performing a voiding task. Our results demonstrate FC (grey matter) is of higher importance than SC (white matter) for this classification. Knowledge of these centres may help us further phenotype patients to appropriate centrally focused treatments in the future.