Area Under ROC Curve Research Articles

Predicting Choroidal Nevus Transformation to Melanoma Using Machine Learning

PurposeTo develop and validate machine learning (ML) models to predict choroidal nevus transformation to melanoma based on multimodal imaging at initial presentation. DesignRetrospective multicenter study. ParticipantsPatients diagnosed with choroidal nevus on the Ocular Oncology Service at Wills Eye Hospital (2007-2017) or Mayo Clinic Rochester (2015-2023). MethodsMultimodal imaging was obtained, including fundus photography, fundus autofluorescence, spectral domain OCT, and B-scan ultrasonography. ML models were created (XGBoost, LGBM, Random Forest, Extra Tree) and optimized for area under receiver operating curve (AUROC). The Wills Eye Hospital cohort was utilized for training and testing (80% training-20% testing) with 5-fold cross validation. The Mayo Clinic cohort provided external validation. Model performance was characterized by AUROC and area under the precision recall curve (AUPRC). Models were interrogated using SHapley Additive exPlanations (SHAP) to identify the features most predictive of conversion from nevus to melanoma. Differences in AUROC and AUPRC between models were tested using 10,000 bootstrap samples with replacement and results. Main Outcome MeasuresAUROC and AUPRC for each ML model. ResultsThere were 2,870 nevi included in the study, with conversion to melanoma confirmed in 128 cases. Simple AI Nevus Transformation System (SAINTS; XGBoost) was the top performing model in the test cohort [pooled AUROC 0.864 (95% confidence interval (CI): 0.864-0.865), pooled AUPRC 0.244 (0.243-0.246)] and in the external validation cohort [pooled AUROC 0.931 (95% CI: 0.930-0.931), pooled AUPRC 0.533 (0.531-0.535)]. Other models also had good discriminative performance: LGBM (test set pooled AUROC 0.831, validation set pooled AUROC 0.815), Random Forest (test set pooled AUROC 0.812, validation set pooled AUROC 0.866), and Extra Tree (test set pooled AUROC 0.826, validation set pooled AUROC 0.915). A model including only nevi with at least 5 years of follow up demonstrated the best performance in AUPRC (test: pooled 0.592 (95% CI: 0.590-0.594); validation: pooled 0.656 (95% CI: 0.655-0.657)). The top five features in SAINTS by SHAP values were: tumor thickness, largest tumor basal diameter, tumor shape, distance to optic nerve, and subretinal fluid extent. ConclusionsWe demonstrate accuracy and generalizability of a ML model for predicting choroidal nevus transformation to melanoma based on multimodal imaging.

Quantitative Volumetric Analysis of Retinal Ischemia with an Oxygen Diffusion Model and OCT Angiography

PurposeRetinal ischemia is a major feature of diabetic retinopathy (DR). Traditional nonperfused areas measured by optical coherence tomography angiography (OCTA) measure blood supply but not ischemia. We propose a novel 3-dimensional (3D) quantitative method to derive ischemia measurements from OCTA data. DesignCross-sectional study. ParticipantsWe acquired 223 macular OCTA volumes from 33 healthy eyes, 33 diabetic eyes without retinopathy, 7 eyes with non-referable DR, 17 eyes with referable but non-vision-threatening DR, and 133 eyes with vision-threatening DR. MethodsEach eye was scanned using a spectral-domain OCTA system (Avanti RTVue-XR, Visionix/Optovue, Inc) with 1.6 mm scan depth in a 3 × 3-mm region (640 × 304 × 304 voxels) centered on the fovea. For each scanned OCTA volume, a custom algorithm removed flow projection artifacts. We then enhanced, binarized, and skeletonized the vasculature in each OCTA volume and generated a 3D oxygen tension map using a zero-order kinetics oxygen diffusion model. Each volume was scaled to the average retina thickness in healthy controls after foveal registration and flattening of the Bruch membrane. Finally, we extracted 3D ischemia maps by comparison with a reference map established from scans of healthy eyes using the same processing. To assess the ability of the ischemia maps to grade DR severity, we constructed receiver-operating characteristic (ROC) curves for diagnosing diabetes, referable DR, and vision-threatening DR. Main outcome measuresSpearman correlation coefficient and area under ROC curve (AUC) were used to quantify the ability of the ischemia maps to DR. ResultsThe ischemia maps showed that the ischemic tissues were at or near pathologically non-perfused areas, but not the normally non-vascular tissue, such as the foveal avascular zone. We found multiple novel metrics, including inferred 3D-oxygen tension, ischemia index, and ischemic volume ratio, were strongly correlated with DR severity. The AUCs of ischemia index measured were 0.95 for diabetes, 0.89 for DR, 0.88 for referable DR, and 0.85 for vision threatening DR. ConclusionsA quantitative method to infer 3D oxygen tension and ischemia using OCTA in diabetic eyes can identify ischemic tissue that are more specific to pathologic changes in DR.

Comparative Analysis of Frequency Ratio, Logistic Regression and Deep Learning Methods for Landslide Susceptibility Mapping in Tokat Province on the North Anatolian Fault Zone (Turkey)

In the current investigation, a Geographic Information System (GIS) and machine learning-based software were employed to generate and compare landslide susceptibility maps (LSMs) for the city center of Tokat, which is situated within the North Anatolian Fault Zone (NAFZ) in the Central Black Sea Region of Turkey, covering an area of approximately 2003 km2. 294 landslides were identified within the study area, with 258 (70%) randomly selected for modeling and the remaining 36 (30%) used for model validation. Three distinct methodologies were used to generate LSMs, namely Frequency Ratio (FR), Logistic Regression (LR), and Deep Learning (DL), using nine parameters, including slope, aspect, curvature, elevation, lithology, rainfall, distance to fault, distance to road, and distance to stream. The susceptibility maps produced in this study were categorized into five classes based on the level of susceptibility, ranging from very low to very high. This study used the area under receiver operating characteristic curve (AUC-ROC), overall accuracy, and precision methods to validate the results of the generated LSMs and compare and evaluate the performance. DL outperformed all validation methods compared to the others. Finally, it is concluded that the generated LSMs will assist decision-makers in mitigating the damage caused by landslides in the study area.

An optical photothermal infrared investigation of lymph nodal metastases of oral squamous cell carcinoma

In this study, optical photothermal infrared (O-PTIR) spectroscopy combined with machine learning algorithms were used to evaluate 46 tissue cores of surgically resected cervical lymph nodes, some of which harboured oral squamous cell carcinoma nodal metastasis. The ratios obtained between O-PTIR chemical images at 1252 cm−1 and 1285 cm−1 were able to reveal morphological details from tissue samples that are comparable to the information achieved by a pathologist’s interpretation of optical microscopy of haematoxylin and eosin (H&E) stained samples. Additionally, when used as input data for a hybrid convolutional neural network (CNN) and random forest (RF) analyses, these yielded sensitivities, specificities and precision of 98.6 ± 0.3%, 92 ± 4% and 94 ± 5%, respectively, and an area under receiver operator characteristic (AUC) of 94 ± 2%. Our findings show the potential of O-PTIR technology as a tool to study cancer on tissue samples.

Flood Susceptibility Mapping Using Information Fusion Paradigm Integrated with Decision Trees

Accurate estimation of flood-damaged zones in a watershed is prominent in guiding a framework for developing sustainable strategies. For these purposes, several flood conditioning factor values at flooded and non-flooded points are extracted, and those points are analyzed using decision tree algorithms and eight novel information fusion techniques to get more reliable flood susceptibility mapping. The belief function values of flood susceptibility values at leaf nodes of the tree are fused by several techniques named Dempster-Shafer (DS), Fuzzy Gamma Overlay (FGO), Hesitant Fuzzy Weighted Averaging (HFWA), Hesitant Fuzzy Weighted Geometric (HFWG), Hesitant Fuzzy Weighted Ordered Averaging (HFWOA), HFWOG, Closeness coefficient (Cc) using Euclidean and Manhattan distances. The flood susceptibility values are extracted from the generated maps and are validated by receiver operating characteristics (ROC) curve parameters, and the seed cell area index (SCAI) of classified flood levels. The area under ROC (AUROC) values of training process are 0.997 for DS, HFWA, HFWOA, and Cc-Euclidean, 0.996 for Cc-Manhattan, 0.995 for FGO and 0.994 for HFWG and HFWOG. The AUROC values of the testing process are 0.951 for DS, HFWA, HFWOA, Cc-Euclidean, and Cc-Manhattan, 0.945 for FGO, 0.943 for HFWG, and 0.941 for HFWOG. True Skill Statistics values are 0.962 and 0.870 for training and testing processes. Although these techniques present excellent performance, the SCAI values versus flood susceptibility classes are fitted to assess the prediction capabilities of the techniques further. HFWA and HFWOG have the first- and second-best performances on the estimations. Hence, information fusion paradigm can be employed to combine flood conditioning factors based on a robust classification method to get reliable predictions of flood potential levels and utilize them for land use and construction planning and management.

Serum Mac-2 binding protein glycosylation isomer dynamics in patients achieving sustained virologic response for hepatitis C virus.

Understanding the dynamics of serum Mac-2 binding protein glycosylation isomer (M2BPGi) remains pivotal for hepatitis C virus (HCV) patients' post-sustained virologic response (SVR12) through direct-acting antivirals (DAAs). We compared areas under receiver operating characteristic curves (AUROCs) of M2BPGi, FIB-4, and APRI and assess M2BPGi cutoff levels in predicting fibrosis stages of ≥F3 and F4 utilizing transient elastography in 638 patients. Variations in M2BPGi levels from pretreatment to SVR12 and their association with pretreatment alanine transaminase (ALT) levels and fibrosis stage were investigated. The AUROCs of M2BPGi were comparable to FIB-4 in predicting ≥F3 (0.914 vs 0.902, P=0.48) and F4 (0.947 vs 0.915, P=0.05) but were superior to APRI in predicting ≥F3 (0.914 vs 0.851, P=0.001) and F4 (0.947 vs 0.857, P<0.001). Using M2BPGi cutoff values of 2.83 and 3.98, fibrosis stages of ≥F3 and F4 were confirmed with a positive likelihood ratio ≥10. The median M2BPGi change was -0.55. Patients with ALT levels ≥5 times ULN or ≥F3 demonstrated more pronounced median decreases in M2BPGi level compared to those with ALT levels 2-5 times ULN and <2 times ULN (-0.97 vs -0.68 and -0.44; P<0.001) or with < F3 (-1.52 vs -0.44; P<0.001). Serum M2BPGi is a reliable marker for advanced hepatic fibrosis. Following viral clearance, there is a notable M2BPGi decrease, with the extent of reduction influenced by ALT levels and fibrosis stage.

Predicting negative attitudes towards suicide in social media texts: prediction model development and validation study.

Implementing machine learning prediction of negative attitudes towards suicide may improve health outcomes. However, in previous studies, varied forms of negative attitudes were not adequately considered, and developed models lacked rigorous external validation. By analyzing a large-scale social media dataset (Sina Weibo), this paper aims to fully cover varied forms of negative attitudes and develop a classification model for predicting negative attitudes as a whole, and then to externally validate its performance on population and individual levels. 938,866 Weibo posts with relevant keywords were downloaded, including 737,849 posts updated between 2009 and 2014 (2009-2014 dataset), and 201,017 posts updated between 2015 and 2020 (2015-2020 dataset). (1) For model development, based on 10,000 randomly selected posts from 2009 to 2014 dataset, a human-based content analysis was performed to manually determine labels of each post (non-negative or negative attitudes). Then, a computer-based content analysis was conducted to automatically extract psycholinguistic features from each of the same 10,000 posts. Finally, a classification model for predicting negative attitudes was developed on selected features. (2) For model validation, on the population level, the developed model was implemented on remaining 727,849 posts from 2009 to 2014 dataset, and was externally validated by comparing proportions of negative attitudes between predicted and human-coded results. Besides, on the individual level, similar analyses were performed on 300 randomly selected posts from 2015 to 2020 dataset, and the developed model was externally validated by comparing labels of each post between predicted and actual results. For model development, the F1 and area under ROC curve (AUC) values reached 0.93 and 0.97. For model validation, on the population level, significant differences but very small effect sizes were observed for the whole sample (χ 2 1 = 32.35, p < 0.001; Cramer's V = 0.007, p < 0.001), men (χ 2 1 = 9.48, p = 0.002; Cramer's V = 0.005, p = 0.002), and women (χ 2 1 = 25.34, p < 0.001; Cramer's V = 0.009, p < 0.001). Besides, on the individual level, the F1 and AUC values reached 0.76 and 0.74. This study demonstrates the efficiency and necessity of machine learning prediction of negative attitudes as a whole, and confirms that external validation is essential before implementing prediction models into practice.

Individualized, dynamic, and full-course vancomycin dosing prediction: a study on the customized dose model.

The single-point trough-based therapeutic drug monitoring (TDM) and Bayesian forecasting approaches are still limited in individualized and dynamic vancomycin delivery. Until recently, there has not yet been enough focus on the direct integration of pharmacokinetic/pharmacodynamic (PK/PD) and TDM to construct a customized dose model (CDM) for vancomycin to achieve individualized, dynamic, and full-course dose prediction from empirical to follow-up treatment. This study sought to establish CDM for vancomycin, test its performance and superiority in clinical efficacy prediction, formulate a CDM-driven full-course dosage prediction strategy to overcome the above challenge, and predict the empirical vancomycin dosages for six Staphylococci populations and four strains in patients with various creatinine clearance rates (CLcr). The PK/PD and concentration models derived from our earlier research were used to establish CDM. The receiver operating characteristic (ROC) curve, with the area under ROC curve (AUCR) as the primary endpoint, for 21 retrospective cases was applied to test the performance and superiority of CDM in clinical efficacy prediction by comparison to the current frequently-used dose model (FDM). A model with an AUCR of at least 0.8 was considered acceptable. Based on the availability of TDM, the strategy of CDM-driven individualized, dynamic, and full-course dose prediction for vancomycin therapy was formulated. Based on the CDM, Monte Carlo simulation was used to predict the empirical vancomycin dosages for the target populations and bacteria. Four CDMs and the strategy of CDM-driven individualized, dynamic, and full-course dose prediction for vancomycin therapy from empirical to follow-up treatment were constructed. Compared with FDM, CDM showed a greater AUCR value (0.807 vs. 0.688) in clinical efficacy prediction. The empirical vancomycin dosages for six Staphylococci populations and four strains in patients with various CLcr were predicted. CDM is a competitive individualized dose model. It compensates for the drawbacks of the existing TDM technology and Bayesian forecasting and offers a straightforward and useful supplemental approach for individualized and dynamic vancomycin delivery. Through mathematical modeling of the vancomycin dosage, this study achieved the goal of predicting doses individually, dynamically, and throughout, thus promoting "mathematical knowledge transfer and application" and also providing reference for quantitative and personalized research on similar drugs.

P-707 Discovery of novel serum biomarkers for premature ovarian insufficiency through machine-learning bioinformatic analysis

Abstract Study question Can we develop predictive, practical biomarkers for premature ovarian insufficiency (POI) from normal ovaries and human blood, tested and validated by machine-learning (ML) algorithm? Summary answer Applying random forest and XGBoost algorithms, we pinpointed 60 significant genes from ovarian tissues; transcriptome analysis identified RPM1 as the most significant biomarker. What is known already Accelerated ovarian aging has been suggested as one of the possible underlying mechanisms of POI; previous literature has focused on identifying its cause in genetic realm, commonly adopting first-generation sequencing (Sanger) and next-generation sequencing (NGS). However, such powerful tools are still limited in accurately depicting the complete transcript information of ovarian tissue. Moreover, theoretical value of such disease-predictive gene candidates needs to be validated using human tissues and blood, and finally, underlying mechanisms should be confirmed with ideal animal model and cell line. Study design, size, duration To overcome the existing limitations and maximize clinical relevance of disease-predictive gene analysis and theoretical biomarker exploration, the current study incorporated the transcriptome expression data from both young and old ovaries obtained from the Genotype-Tissue Expression (GTEx) database. Using supervised machine learning techniques, potential marker genes were identified, and a meaningful predictive index was subsequently developed using human plasma samples of control and POI women with and without pregnancy history. Participants/materials, setting, methods The performance evaluation of two ML models were achieved using the area under receiver operating characteristic curve (AUROC), accuracy, kappa, and F1 score, among various age groups. The expression profiles of final candidate marker genes were analyzed within the GTEx datasets, in relation to their correlation pattern with age. Finally, the potential transcriptome biomarkers were quantitatively evaluated using human blood samples from 92 controls and 108 POI women visiting a tertiary center in 2023. Main results and the role of chance We found 41 significant genes using the random forest algorithm and 19 significant genes using the XGboost algorithm. We then tested the ML-characterized 7 final candidate biomarkers in human blood samples (all p &amp;lt; 0.05). The AUROC and interventional predictive index results demonstrated the potential of RPM1 as a predictive marker with AUROC of 0.784, comparable to the AUROC of anti-mullerian hormone (AMH) as 0.659 and inhibin B (INHB) as 0.646. Limitations, reasons for caution To avoid potential biases, RPMs should be tested in a larger, multinational cohort with different genetic and geographic backgrounds, possibly incorporating Mendelian randomization method to estimate the causal effect of RPMs on POI. Finally, to ultimately confirm their therapeutic potential, preclinical studies should be considered using RPM recombinant proteins. Wider implications of the findings Our results proposed that predictive biomarkers for POI could be associated with premature, accelerated ovarian aging of the disease mechanism, and these biomarkers can be developed as evaluation metrics for the efficacy, progress, and prognosis of interventions and treatment modalities assigned to POI women seeking fertility. Trial registration number 202302240001

Acute kidney injury in critically ill obstetric patients: Incidence and role of neutrophil gelatinase-associated lipocalcin - A prospective observational cohort study.

Data focussing on acute kidney injury (AKI) in obstetric patients admitted to the intensive care unit (ICU) are scarce and even more so regarding the role of neutrophil gelatinase-associated lipocalcin (NGAL) in detecting AKI or predicting outcomes in these patients. Hence, we aim to evaluate the incidence of AKI in obstetric ICU patients and validate the role of urinary and serum NGAL in predicting the onset of AKI and mortality. This prospective observational cohort included 45 obstetric patients admitted in ICU, excluding those with prior renal dysfunction. Serum creatinine and urine output were monitored for the occurrence of AKI during the ICU stay. The outcome of the patient (survival or death) in the ICU and hospital was recorded, and serum and urinary NGAL were determined at the time of ICU admission. AKI occurred in 32 [71.1%; 95% confidence interval (CI): 55.4%, 86.8%] patients during their ICU stay. Serum NGAL showed an area under receiver operating characteristic curve (AUROCC) of 0.630 (95% CI: 0.417, 0.842) (P = 0.231) for AKI and 0.486 (95% CI: 0.295, 0.676) (P = 0.883) for ICU mortality. Urinary NGAL showed AUROC = 0.472 (95% CI: 0.285, 0.660) (P = 0.772) to predict AKI and 0.430 (95% CI: 0.268, 0.652) (P = 0.684) for ICU mortality. AKI is common amongst critically ill obstetric ICU patients. However, serum and urinary NGAL cannot be advocated to discriminate between patients with or without AKI or between survivors and non-survivors in critically ill obstetric patients.

Predicting novel mosquito-associated viruses from metatranscriptomic dark matter.

The exponential growth of metatranscriptomic studies dedicated to arboviral surveillance in mosquitoes has yielded an unprecedented volume of unclassified sequences referred to as the virome dark matter. Mosquito-associated viruses are classified based on their host range into Mosquito-specific viruses (MSV) or Arboviruses. While MSV replication is restricted to mosquito cells, Arboviruses infect both mosquito vectors and vertebrate hosts. We developed the MosViR pipeline designed to identify complex genomic discriminatory patterns for predicting novel MSV or Arboviruses from viral contigs as short as 500 bp. The pipeline combines the predicted probability score from multiple predictive models, ensuring a robust classification with Area Under ROC (AUC) values exceeding 0.99 for test datasets. To assess the practical utility of MosViR in actual cases, we conducted a comprehensive analysis of 24 published mosquito metatranscriptomic datasets. By mining this metatranscriptomic dark matter, we identified 605 novel mosquito-associated viruses, with eight putative novel Arboviruses exhibiting high probability scores. Our findings highlight the limitations of current homology-based identification methods and emphasize the potentially transformative impact of the MosViR pipeline in advancing the classification of mosquito-associated viruses. MosViR offers a powerful and highly accurate tool for arboviral surveillance and for elucidating the complexities of the mosquito RNA virome.

Metagenomics reveals unique gut mycobiome biomarkers in psoriasis.

In present, the diagnosis of psoriasis is mainly based on the patient's typical clinical manifestations, dermoscopy and skin biopsy, and unlike other immune diseases, psoriasis lacks specific indicators in the blood. Therefore, we are required to search novel biomarkers for the diagnosis of psoriasis. In this study, we analyzed the composition and the differences of intestinal fungal communities composition between psoriasis patients and healthy individuals in order to find the intestinal fungal communities associated with the diagnosis of psoriasis. We built a machine learning model and identified potential microbial markers for the diagnosis of psoriasis. The results of AUROC (area under ROC) showed that Aspergillus puulaauensis (AUROC=0.779), Kazachstania africana (AUROC=0.750) and Torulaspora delbrueckii (AUROC=0.745) had high predictive ability (AUROC>0.7) for predicting psoriasis, While Fusarium keratoplasticum (AUROC=0.670) was relatively lower (AUROC<0.7). The strategy based on the prediction of intestinal fungal communities provides a new idea for the diagnosis of psoriasis and is expected to become an auxiliary diagnostic method for psoriasis.

Genetic and non-genetic factors in prediction of early pubertal development in Chinese girls.

The objective of this study is to develop a combined predictive model for early pubertal development (EPD) in girls based on both non-genetic and genetic factors. The case-control study encompassed 147 girls diagnosed with EPD and 256 girls who exhibited normal pubertal development. The non-genetic risk score (NGRS) was calculated based on 6 independent biochemical predictors screened by multivariate logistic regressions, and the genetic risk score (GRS) was constructed using 28 EPD related single-nucleotide polymorphisms (SNPs). Area under receiver operator characteristic curve (AROC), net reclassification optimization index (NRI) and integration differentiation index (IDI) were used to evaluate the improvement of adding genetic variants to the non-genetic risk model. Overweight (OR=2.74), longer electronic screen time (OR=1.79) and higher ratio of plastic bottled water (OR=1.01) were potential risk factors, and longer exercise time (OR=0.51) and longer day sleeping time (OR=0.97) were protective factors for EPD, and the AROC of NGRS model was 83.6% (79.3-87.9%). The GRS showed a significant association with EPD (OR=1.90), and the AROC of GRS model was 65.3% (59.7-70.8%). After adding GRS to the NGRS model, the AROC significantly increased to 85.7% (81.7-89.6%) (P=0.020), and the reclassification significantly improved, with NRI of 8.19% (P= 0.023) and IDI of 4.22% (P <0.001). We established a combined prediction model of EPD in girls. Adding genetic variants to the non-genetic risk model brought modest improvement. However, the non-genetic factors such as overweight and living habits have higher predictive utility.

Deltoid tuberosity index for proximal humerus fracture: reliability and a predictor of systemic osteoporosis in an Asian population

Our study investigates the reliability of deltoid tuberosity index (DTI) and DTI as a predictor of systemic osteoporosis. The proximal humerus is a common fragility fracture. Current literature suggests that poor local bone density is a significant predictor for surgical fixation failure. The DTI is a simple radiographical tool that is strongly correlates with local humeral BMD aiding surgical planning to consider adjuncts or arthroplasty. However, there is a lack of data in the reliability of assessment of DTI, as well as its correlation to systemic osteoporosis. Respective cohort of patients with PHF treated at a trauma center in Singapore from August 2017 to July 2018 were recruited. Four raters at different levels of varying clinical seniority measured DTI using shoulder radiographs. The dual energy X-ray Absorptiometry (DEXA) bone mineral density (BMD) scan of the hip and lumbar spine was used to diagnose osteoporosis. Area under receiver operating characteristics (AUROC) analysis was conducted to study the diagnostic utility of DTI to predict the risk of osteoporosis. Our study had 87 patients consisting 18 males and 69 females, mainly of Chinese ethnicity (84%) and mean age of 69.7 years (SD 9.52, range 39-92yrs). For assessment of DTI, there was good intra-rater reliability amongst four raters (correlation coefficient range 0.805- 0.843) and excellent inter-rater reliability between al raters (intraclass correlation coefficient = 0.898; 95% CI 0.784-0.950, p-value <0.001). Based on BMD, 55.2% (n=48) were osteoporotic using T-score <-2.5. The highest correlation of DTI to BMD was with femoral neck density at 0.580. The DTI cut-off of 1.6 had the highest combined sensitivity and false positive rate, with area under curve (AUC) = 0.682 (95% CI, 0.564-0.799) for the overall population and AUC =0.706 (95% CI, 0.569-0.842) for patients <75 years. The DTI is a simple and reliable tool, strengthening its applicability in clinical practice to enhance preoperative planning in the surgical fixation of PHF. DTI with a cut off of 1.6 may be helpful tool prompting clinicians to workup and manage underlying osteoporosis.

Establishment and evaluation of a nomogram model for predicting the risk of sepsis in diabetic foot patients

To establish a nomogram model for predicting the risk of sepsis in diabetic foot patients, and to provide reference for clinical prevention and treatment. The clinical data of 430 patients with diabetic foot who were hospitalized in Chu Hsien-I Memorial Hospital of Tianjin Medical University from January 2022 to March 2023 were reviewed and collected, including age, gender, past medical history, smoking and drinking history, family history, diabetes course, Texas grade of diabetic foot and laboratory indicators within 24 hours after admission. Patients were divided into sepsis group and non-sepsis group according to the presence or absence of sepsis during hospitalization. The differences in clinical data between the two groups were compared. Multivariate Logistic regression analysis was used to screen the influencing factors of sepsis in patients with diabetic foot during hospitalization, and a nomogram predictive model was established. The performance of the prediction model was evaluated by receiver operator characteristic curve (ROC curve), calibration curve and decision curve analysis (DCA). Internal validation was performed by using Bootstrap method. A total of 430 patients were enrolled, among which 90 patients developed sepsis during hospitalization and 340 patients did not. There were statistically significant differences in diabetes course, Texas grade of diabetic foot, white blood cell count (WBC), neutrophil count (NEU), lymphocyte count (LYM), neutrophil to lymphocyte ratio (NLR), hemoglobin (Hb), albumin (Alb), glycosylated hemoglobin (HbA1c), C-reactive protein (CRP), and blood urea nitrogen (BUN) between the two groups. Multivariate Logistic regression analysis showed that diabetes course [odds ratio (OR) = 2.774, 95% confidence interval (95%CI) was 1.053-7.308, P = 0.039], Texas grade of diabetic foot (OR = 2.312, 95%CI was 1.014-5.273, P = 0.046), WBC (OR = 1.160, 95%CI was 1.042-1.291, P = 0.007), HbA1c (OR = 1.510, 95%CI was 1.278-1.784, P < 0.001), CRP (OR = 1.007, 95%CI was 1.000-1.014, P = 0.036) were independent risk factors for sepsis in patients with diabetic foot during hospitalization, while Alb was a protective factor (OR = 0.885, 95%CI was 0.805-0.972, P = 0.011). A nomogram predictive model was constructed based on the above 6 indicators. The ROC curve showed that the area under ROC curve (AUC) of the nomogram predictive model for identifying the sepsis patients was 0.919 (95%CI was 0.889-0.948). The AUC of the nomogram predictive model after internal verification was 0.918 (95%CI was 0.887-0.946). Hosmer-Lemeshow test showed χ 2 = 2.978, P = 0.936, indicating that the calibration degree of the predictive model was good. Calibration curve showed that the predicted probability of sepsis was in good agreement with the actual probability. DCA curve showed that the nomogram predictive model had good clinical usefulness. The nomogram predictive model based on the risk factors of diabetes course, Texas grade of diabetic foot, WBC, HbA1c, CRP and Alb has good predictive value for the occurrence of sepsis in patients with diabetic foot during hospitalization, which is helpful for clinical screening of the possibility of diabetic foot patients progressing to sepsis, and timely personalized intervention for different patients.

Multiparametric MR characterization for human epithelial growth factor receptor 2 expression in bladder cancer: an exploratory study.

To investigate the application value of multiparametric MRI in evaluating the expression status of human epithelial growth factor receptor 2 (HER2) in bladder cancer (BCa). From April 2021 toJuly 2023, preoperative imaging manifestations of 90 patients with pathologically confirmed BCa were retrospectively collected and analyzed. All patients underwent multiparametric MRI including synthetic MRI, DWI, from which the T1, T2, proton density (PD) and apparent diffusion coefficient (ADC) values were obtained. The clinical and imaging characteristics as well as quantitative parameters (T1, T2, PD and ADC values) between HER2-positive and -negative BCa were compared using student t test and chi-square test. The diagnostic efficacy of parameters in predicting HER2 expression status was evaluated by calculating the area under ROC curve (AUC). In total, 76 patients (mean age, 63.59years ± 12.84 [SD]; 55 men) were included: 51 with HER2-negative and 25 with HER2-positiveBCa. HER2-positive group demonstrated significantly higher ADC, T1, and T2 values than HER2-negative group (all P < 0.05). The combination of ADC values and tumor grade yielded the best diagnostic performance in evaluating HER2 expression level with an AUC of 0.864. The multiparametric MR characterization can accurately evaluate the HER2 expression status in BCa, which may further guide the determination of individualized anti-HER2 targeted therapy strategies.

Novel decision tree models predict the overall survival of patients with submandibular gland cancer.

While the accurate prediction of the overall survival (OS) in patients with submandibular gland cancer (SGC) is paramount for informed therapeutic planning, the development of reliable survival prediction models has been hindered by the rarity of SGC cases. The purpose of this study is to identify key prognostic factors for OS in SGC patients using a large database and construct decision tree models to aid the prediction of survival probabilities in 12, 24, 60 and 120 months. We performed a retrospective cohort study using the Surveillance, Epidemiology and End Result (SEER) program. Demographic and peri-operative predictor variables were identified. The outcome variables overall survival at 12-, 24-, 60, and 120 months. The C5.0 algorithm was utilized to establish the dichotomous decision tree models, with the depth of tree limited within 4 layers. To evaluate the performances of the novel models, the receiver operator characteristic (ROC) curves were generated, and the metrics such as accuracy rate, and area under ROC curve (AUC) were calculated. A total of 1,705, 1,666, 1,543, and 1,413 SGC patients with a follow up of 12, 24, 60 and 120 months and exact survival status were identified from the SEER database. Predictor variables of age, sex, surgery, radiation, chemotherapy, tumor histology, summary stage, metastasis to distant lymph node, and marital status exerted substantial influence on overall survival. Decision tree models were then developed, incorporating these vital prognostic indicators. Favorable consistency was presented between the predicted and actual survival statuses. For the training dataset, the accuracy rates for the 12-, 24-, 60- and 120-month survival models were 0.866, 0.767, 0.737 and 0.797. Correspondingly, the AUC values were 0.841, 0.756, 0.725, and 0.774 for the same time points. Based on the most important predictor variables identified using the large, SEER database, decision tree models were established that predict OS of SGC patients. The models offer a more exhaustive evaluation of mortality risk and may lead to more personalized treatment strategies.

Development and validation of an automated machine learning model for the multi-class classification of diabetic retinopathy, central retinal vein occlusion and branch retinal vein occlusion based on color fundus photographs

IntroductionAutomated machine learning (AutoML) is a novel artificial intelligence (AI) strategy that enables clinicians without coding experience to develop their own AI models. This study assessed the discriminative performance of AutoML in differentiating diabetic retinopathy (DR), central retinal vein occlusion (CRVO) and branch retinal vein occlusion (BRVO) from normal fundi using color fundus photographs (CFPs). MethodsWe carried out AutoML model design using CFPs retrieved from a publicly available CFP data set (3200 labelled images). The retrieved CFPs were reviewed for quality and then uploaded to the Google Cloud Vertex AI platform for AutoML training and testing. We trained a multi-class classification model to differentiate DR, CRVO, BRVO from normal fundi using 875 CFPs and externally validated the model using 210 CFPs obtained from another dataset. Performance metrics, including area under receiver operator curve (AUROC) and sensitivity were reported. We then compared the AutoML model to state-of-the-art deep learning (DL)-based DR and RVO multi-class models identified through a literature search. ResultsOur AutoML model showed high discriminative performance in the multi-class classification of DR, CRVO and BRVO based on CFPs, with an AUROC, precision and recall reaching 0.995, 95.4% and 95.4% respectively at the 0.5 confidence threshold. The per-label sensitivity and specificity, respectively, were normal fundi (97.5%, 100%), DR (100%, 93.88%), CRVO (66.67%, 100%) and BRVO (71.43%, 98.73%). Our AutoML model generally showed similar performance to the state-of-the-art DL classifiers. ConclusionOur AutoML model can detect DR, CRVO, and BRVO in CFPs with good diagnostic accuracy and is a potentially useful screening tool.

Development of a grading diagnostic model for schistosomiasis-induced liver fibrosis based on radiomics and clinical laboratory indicators

To investigate the feasibility of developing a grading diagnostic model for schistosomiasis-induced liver fibrosis based on B-mode ultrasonographic images and clinical laboratory indicators. Ultrasound images and clinical laboratory testing data were captured from schistosomiasis patients admitted to the Second People's Hospital of Duchang County, Jiangxi Province from 2018 to 2022. Patients with grade I schistosomiasis-induced liver fibrosis were enrolled in Group 1, and patients with grade II and III schistosomiasis-induced liver fibrosis were enrolled in Group 2. The machine learning binary classification tasks were created based on patients'radiomics and clinical laboratory data from 2018 to 2021 as the training set, and patients'radiomics and clinical laboratory data in 2022 as the validation set. The features of ultrasonographic images were labeled with the ITK-SNAP software, and the features of ultrasonographic images were extracted using the Python 3.7 package and PyRadiomics toolkit. The difference in the features of ultrasonographic images was compared between groups with t test or Mann-Whitney U test, and the key imaging features were selected with the least absolute shrinkage and selection operator (LASSO) regression algorithm. Four machine learning models were created using the Scikit-learn repository, including the support vector machine (SVM), random forest (RF), linear regression (LR) and extreme gradient boosting (XGBoost). The optimal machine learning model was screened with the receiver operating characteristic curve (ROC), and features with the greatest contributions to the differentiation features of ultrasound images in machine learning models with the SHapley Additive exPlanations (SHAP) method. The ultrasonographic imaging data and clinical laboratory testing data from 491 schistosomiasis patients from 2019 to 2022 were included in the study, and a total of 851 radiomics features and 54 clinical laboratory indicators were captured. Following statistical tests (t = -5.98 to 4.80, U = 6 550 to 20 994, all P values < 0.05) and screening of key features with LASSO regression, 44 features or indicators were included for the subsequent modeling. The areas under ROC curve (AUCs) were 0.763 and 0.611 for the training and validation sets of the SVM model based on clinical laboratory indicators, 0.951 and 0.892 for the training and validation sets of the SVM model based on radiomics, and 0.960 and 0.913 for the training and validation sets of the multimodal SVM model. The 10 greatest contributing features or indicators in machine learning models included 2 clinical laboratory indicators and 8 radiomics features. The multimodal machine learning models created based on ultrasound-based radiomics and clinical laboratory indicators are feasible for intelligent identification of schistosomiasis-induced liver fibrosis, and are effective to improve the classification effect of one-class data models.

Machine Learning–Based Models for Advanced Fibrosis and Cirrhosis Diagnosis in Chronic Hepatitis B Patients With Hepatic Steatosis

Background and AimsThe global rise of chronic hepatitis B (CHB) superimposed on hepatic steatosis (HS) warrants non-invasive, precise tools for assessing fibrosis progression. This study leveraged machine learning (ML) to develop diagnostic models for advanced fibrosis and cirrhosis in this patient population. MethodsTreatment-naive CHB patients with concurrent HS who underwent liver biopsy in ten medical centers were enrolled as a training cohort and an independent external validation cohort (NCT05766449). Six ML models were implemented to predict advanced fibrosis and cirrhosis. The final models, derived from Shapley Additive exPlanations, were compared to Fibrosis-4 Index (FIB-4), Nonalcoholic fatty liver disease Fibrosis Score (NFS), and Aspartate transaminase to platelet ratio index (APRI) using the area under receiver operating characteristic curve (AUROC), and decision curve analysis (DCA). ResultsOf 1,198 eligible patients, the random forest (RF) model achieved AUROCs of 0.778 [95% confidence interval (CI) 0.749-0.807] for diagnosing advanced fibrosis (RF-AF model) and 0.777 (95%CI 0.748-0.806) for diagnosing cirrhosis (RF-C model) in the training cohort, and maintained high AUROCs in the validation cohort. In the training cohort, the RF-AF model obtained an AUROC of 0.825 (95% CI 0.787-0.862) in patients with HBV DNA ≥105 IU/ml, and RF-C model had an AUROC of 0.828 (95% CI 0.774-0.883) in female patients. The two models outperformed FIB-4, NFS, and APRI in the training cohort, and also performed well in the validation cohort. ConclusionThe RF models provide reliable, non-invasive tools for identifying advanced fibrosis and cirrhosis in CHB patients with concurrent HS, offering a significant advancement in the co-management of the two diseases.