Random Forest Machine Learning Research Articles

Deep Learning for Urban Tree Canopy Coverage Analysis: A Comparison and Case Study

Urban tree canopy (UTC) coverage, or area, is an important metric for monitoring changes in UTC over large areas within a municipality. Several methods have been used to obtain these data, but remote sensing image classification is one of the fastest and most reliable over large areas. However, most studies have tested only one or two classification methods to accomplish this while using costly satellite imagery or LiDAR data. This study seeks to compare three urban tree canopy cover classifiers by testing a deep learning U-Net convolutional neural network (CNN), support vector machine learning classifier (SVM) and a random forests machine learning classifier (RF) on cost-free 2012 aerial imagery over a small southern USA city and midsize, growing southern USA city. The results of the experiment are then used to decide the best classifier and apply it to more recent aerial imagery to determine canopy changes over a 10-year period. The changes are subsequently compared visually and statistically with recent urban heat maps derived from thermal Landsat 9 satellite data to compare the means of temperatures within areas of UTC loss and no change. The U-Net CNN classifier proved to provide the best overall accuracy for both cities (89.8% and 91.4%), while also requiring the most training and classification time. When compared spatially with city heat maps, city periphery regions were most impacted by substantial changes in UTC area as cities grow and the outer regions get warmer. Furthermore, areas of UTC loss had higher temperatures than those areas with no canopy change. The broader impacts of this study reach the urban forestry managers at the local, state/province, and national levels as they seek to provide data-driven decisions for policy makers.

Plasmodium vivax antigen candidate prediction improves with the addition of Plasmodium falciparum data

AbstractIntensive malaria control and elimination efforts have led to substantial reductions in malaria incidence over the past two decades. However, the reduction in Plasmodium falciparum malaria cases has led to a species shift in some geographic areas, with P. vivax predominating in many areas outside of Africa. Despite its wide geographic distribution, P. vivax vaccine development has lagged far behind that for P. falciparum, in part due to the inability to cultivate P. vivax in vitro, hindering traditional approaches for antigen identification. In a prior study, we have used a positive-unlabeled random forest (PURF) machine learning approach to identify P. falciparum antigens based on features of known antigens for consideration in vaccine development efforts. Here we integrate systems data from P. falciparum (the better-studied species) to improve PURF models to predict potential P. vivax vaccine antigen candidates. We further show that inclusion of known antigens from the other species is critical for model performance, but the inclusion of only the unlabeled proteins from the other species can result in misdirection of the model toward predictors of species classification, rather than antigen identification. Beyond malaria, incorporating antigens from a closely related species may aid in vaccine development for emerging pathogens having few or no known antigens.

Abstract B028: Analytical validation of tumor-informed whole genome sequencing analyses for detection of molecular residual disease in solid tumors

Abstract Introduction: Across all patients diagnosed with solid tumors, approximately two-thirds initially present with locoregional disease and are potentially eligible for curative-intent intervention. However, there is a significant subset of patients for which residual tumor cells remain afterwards, potentially leading to disease recurrence. It has been reported that these residual tumor cells, representative of molecular residual disease (MRD), can be detected through analyses of circulating tumor DNA (ctDNA), often earlier than can be achieved through radiographic imaging or clinical presentation. Methods: Labcorp Plasma Detect was utilized to analyze patient-matched tumor, germline (white blood cells), and cell-free or contrived DNA samples through whole genome sequencing at approximately 80x, 40x, and 30x depth, respectively. Raw sequencing data were demultiplexed and trimmed for read quality, then aligned to the hg19 human reference genome. Tumor-specific single nucleotide variants were identified from the tumor and germline datasets and used to determine the presence of cancer within cell-free or contrived DNA samples through a random forest machine learning model. ctDNA status was determined based on the level of signal compared to a reference cohort of noncancerous donor plasma samples (n=80). Results: Analytical specificity was determined through analysis of five noncancerous donor plasma samples with five replicates each, evaluated against 60 somatic mutation profiles obtained from patients with stage III colon cancer, demonstrating a specificity of 99.4% (835/840). Analytical sensitivity was assessed using three contrived cell line models from breast and melanoma tumors, analyzed across three tumor content levels with five replicates at each level and resulted in a limit of detection of 0.005% tumor content using a 100% hit-rate model. Precision, repeatability, and reproducibility were demonstrated within and across runs, operators, and instruments, resulting in a 100% detection rate at 0.05% tumor content with a coefficient of variation of 8.5%. Finally, analytical accuracy compared to an independent tumor-informed ctDNA MRD approach was determined through analysis of 39 clinical cases obtained from patients with bladder, breast, colorectal, and lung cancer, yielding a positive percent agreement of 91% (10/11) and negative percent agreement of 100% (24/24). The prognostic value of ctDNA MRD was demonstrated in stage III colon cancer and clinical validation results for HPV-negative head and neck squamous cell carcinoma will be presented separately. Conclusions: There is significant potential for tumor-informed, non- bespoke MRD approaches for ctDNA detection across a broad range of solid tumor types and curative-intent clinical settings. These data demonstrate the analytical performance of Labcorp Plasma Detect in a CAP/CLIA laboratory for pan-solid tumor research use together with investigational use in early stage colon cancer to support the interventional MEDOCC-CrEATE clinical trial. Citation Format: Kaitlin Victor, Andrew Georgiadis, Ellen Verner, Antoine Simmons, David Riley, James R White, Christopher Greco, Liam Cox, Jesse Fox, Jennifer B Jackson, Eric A Severson, Brian J Caveney, Marcia Eisenberg, Taylor J Jensen, Shakti Ramkissoon, Samuel V Angiuoli, Amy Greer, Kenneth Valkenburg, Mark Sausen. Analytical validation of tumor-informed whole genome sequencing analyses for detection of molecular residual disease in solid tumors [abstract]. In: Proceedings of the AACR Special Conference: Liquid Biopsy: From Discovery to Clinical Implementation; 2024 Nov 13-16; San Diego, CA. Philadelphia (PA): AACR; Clin Cancer Res 2024;30(21_Suppl):Abstract nr B028.

Abstract A043: Tumor-informed whole genome sequencing of ctDNA from lymph and plasma detects molecular residual disease in HPV-negative head and neck cancer patients

Abstract Introduction: Recurrence after tumor resection is a major cause of failure in HPV-negative head and neck squamous cell carcinoma (HNSCC), with up to 50% of patients recurring within 2 years. There is unmet need for an accurate diagnostic test that can predict risk of recurrence prior to adjuvant therapy selection. We previously demonstrated that ctDNA in lymphatic exudate collected via surgical drains (“lymph”) accurately identifies molecular residual disease (MRD) and outperforms plasma in HNSCC patients using a targeted sequencing (TS) approach. While TS assays are robust and sensitive, they have key limitations, namely limited ctDNA features, higher cost, longer turnaround times, and lower throughput. Here we evaluate whole genome sequencing (WGS) for detection of MRD in both lymph and plasma sampled 24 hours after surgery. Methods: Lymph, plasma, and whole blood were collected 24 hours postoperatively from 30 HPV-negative HNSCC patients (stages I – IV) along with resected tumor. To detect ctDNA, we utilized WGS analyses of patient matched tumor, germline (blood), and cfDNA from lymph or plasma samples at approximately 150x, 30x, and 30x depth, respectively. Tumor- specific single nucleotide variants were identified from tumor and germline datasets, from which a candidate variant set was used to determine the presence of ctDNA within lymph or plasma DNA through a random forest machine learning model. ctDNA status was determined based on the level of signal compared to a reference population of noncancerous donor plasma samples (n=80). 5 tumors and 1 lymph sample failed QC; in total, 24 patients remained, yielding 10 patients with disease recurrence and 14 with no evidence of disease with >1 year of follow-up. The Kaplan-Meier (KM) estimator with log-rank test and Cox proportional-hazards model were used for survival analyses. Results: KM survival analyses indicated that WGS of lymph ctDNA accurately predicts recurrence (sensitivity (SN) = 50%, specificity (SP) = 93%; p = 0.003, Hazard ratio (HR) = 6.0), consistent with previously published results using a TS MRD approach in this cohort (SN = 80%, SP = 64%, p = 0.02, HR = 5.3). WGS of lymph ctDNA outperformed plasma at this timepoint (SN = 30%, SP = 93%, p = 0.03, HR = 4.2). WGS of lymph ctDNA also showed enhanced performance in locoregional relapse (n=21, SN = 57%, SP = 93%; p = 0.006, HR = 6.7). Finally, the ctDNA fraction of proximal lymph was higher than paired plasma in all locoregional recurrences (n=7), while peripheral plasma had a higher ctDNA fraction in 2 out of 3 distant recurrences. Conclusions: Tumor-informed WGS analysis of lymph ctDNA represents a robust MRD approach in HPV-negative HNSCC. Lymph outperforms time-matched plasma for prediction of recurrence, particularly in patients with locoregional relapse. Accurate MRD identification in the immediate post-surgical timeframe suggests that lymph analysis has potential to augment pathology and longitudinal plasma testing to provide more personalized adjuvant treatment decision-making in patients with HPV-negative HNSCC. Citation Format: Zhuosheng Gu, Seka Lazare, Noah Earland, Marra S Francis, Adam Harmon, Megan Long, Amy Greer, Ellen Verner, Andrew Georgiadis, Mark Sausen, Shakti Ramkissoon, Jose P Zevallos, Aadel A Chaudhuri, Taylor J Jensen, Wendy Winckler. Tumor-informed whole genome sequencing of ctDNA from lymph and plasma detects molecular residual disease in HPV-negative head and neck cancer patients [abstract]. In: Proceedings of the AACR Special Conference: Liquid Biopsy: From Discovery to Clinical Implementation; 2024 Nov 13-16; San Diego, CA. Philadelphia (PA): AACR; Clin Cancer Res 2024;30(21_Suppl):Abstract nr A043.

Random forest models reveal academic and financial factors outweigh demographics in predicting completion of a year-round veterinary program

Abstract OBJECTIVE The purpose of this study was to develop random forest classifier models (a type of supervised machine learning algorithm) that could (1) predict students who will or will not complete the DVM degree requirements and (2) identify the top predictors for academic success and completion of the DVM degree. METHODS The study utilized Ross University School of Veterinary Medicine student records from 2013 to 2022. Twenty-four variables encompassing demographic (eg, age, race), academic (eg, grade point average), and financial aid (eg, outstanding balances) data were assessed in 11 cross-validated random forest machine learning models. One model was built assessing all years of data and 10 individual models were developed for each enrollment year to compare how the top predictors of success varied among the years. RESULTS Consistently, only academic and financial factors were identified as being features of importance (predictors) in all models. Demographic factors such as race were not important for predicting student success. All models performed very well to excellently based on multiple performance metrics including accuracy, ranging from 96.1% to 99%, and the areas under the receiver operating characteristic curves, ranging from 98.1% to 99.9%. CONCLUSIONS The random forest algorithm is a powerful machine learning prediction model that performs well with veterinary student academic records and is customizable such that variables important to each veterinary school’s student population can be assessed. CLINICAL RELEVANCE Identifying predictors of success as well as at-risk students is essential for providing targeted curricular interventions to increase retention and achieve timely completion of a DVM degree.

Comprehensive assessment of cascading dams-induced hydrological alterations in the lancang-mekong river using machine learning technique

The development of cascading hydropower dams in river basins has significantly altered natural flow regimes in recent decades. This study investigates hydrological alterations caused by cascading hydropower dams in the Lancang-Mekong River Basin (LMRB) by integrating the Indicators of Hydrologic Alteration (IHA) method with non-regulated flow predicted using the Random Forest (RF) machine learning (ML) technique. The analysis focuses on four hydrological stations: Chiang Saen, Mukdahan, Pakse, and Stung Treng across pre-impact (1961–1991), transition (1992–2008), and post-impact (2009–2021) periods. RF models predict streamflow altered primarily by human activities, particularly hydropower development using spatially averaged daily precipitation, cumulative precipitation, and temperature data. This approach isolates human-induced impacts, unlike previous studies that combined climate change and human activities. Results indicate that post-impact alterations were most pronounced at Chiang Saen (77.3%) and decreased downstream. Alterations increased from the transition to post-impact periods by 31.5%, 22.2%, 26.5%, and 17.6% at Chiang Saen, Mukdahan, Pakse, and Stung Treng, respectively. The median monthly flow, annual extreme conditions, and rate and frequency of flow change groups in the IHA were highly altered compared to natural conditions in the post-impact period. Human activities contributed over 50% of streamflow changes (2017–2021). This approach provides a more precise assessment of dam-induced hydrological alterations, aiding hydropower management by isolating human impacts and offering high-resolution predictions.

Cross-sectional study on the diagnostic significance of plasma exosomal miRNAs in HBV-related hepatocellular carcinoma

PurposeHepatocellular carcinoma (HCC) associated with Hepatitis B Virus (HBV) is one of the most severe malignancies in East Asia, where early diagnosis is crucial for improving patient prognosis. So we aim to identify effective early diagnostic model for HCC.Design and methodsWe enrolled 108 early-stage HCC patients and 102 non-HCC individuals underlying HBV infection, collecting plasma exosomal miRNAs (exo-miRNAs) from all participants. These patients were randomly assigned to sequencing, screening, training, and validation group. After preliminary screening of candidate exo-miRNAs by next-generation high-throughput sequencing, qPCR data from the screening group were utilized in conjunction with the random forest machine learning algorithm to identify candidate exo-miRNAs with diagnostic potential. Subsequently, logistic regression diagnostic model was constructed using the relative expression levels of candidate exo-miRNAs, alpha-fetoprotein (AFP) levels and clinical parameters of gender and the presence of cirrhosis from the training group. The diagnostic accuracy of diagnostic model was subsequently validated in the validation group.ResultsFirstly, we identified miR-212-5p, miR-1248, and miR-1250-5p as candidate exo-miRNAs with potential diagnostic value. The exo-miRNAs panel, which consisted of miR-212-5p, miR-1248, miR-1250-5p, along with clinical parameters of gender and cirrhosis, achieved an AUC of 0.8634 (95% CI: 0.8027–0.9241), demonstrating diagnostic performance non-inferior to AFP in the independent dataset. Subsequently, by combining exo-miRNAs, AFP level and clinical parameter of gender, we enhanced the diagnostic panel, miRAGe, which exhibited an AUC of 0.9499 (95% CI: 0.9192–0.9806), sensitivity of 0.8900, and specificity of 0.9468.ConclusionOur study indicates that the miRAGe panel has low rate of both missed diagnosis and misdiagnosis rates, potentially serving as a useful diagnostic tool for HBV-related HCC in early stage, which may subsequently contribute to improve the prognosis.

Usefulness of decision tree analysis of MRI features for diagnosis of placenta accreta spectrum in cases with placenta previa.

Placenta previa complicated by placenta accrete spectrum (PAS) is a life-threatening obstetrical condition; therefore, preoperative diagnosis of PAS is important to determine adequate management. Although several MRI features that suggest PAS has been reported, the diagnostic importance, as well as optimal use of each feature has not been fully evaluated. The occurrence of 11 PAS-related MRI features was investigated in MR images of 145 patients with placenta previa. The correlation between each MRI feature and pathological diagnosis of PAS was evaluated using univariate analysis. A decision tree model was constructed according to a random forest machine learning model of variable selection. Eight MRI features showed a significant correlation with PAS in univariate analysis. Among these features, placental/uterine bulge and myometrial thinning showed high odds ratios: 138.2 (95% CI: 12.7-1425.6) and 66.0 (95% CI: 18.01-237.1), respectively. A decision tree was constructed based on five selected MRI features: myometrial thinning, placental bulge, serosal hypervascularity, placental ischemic infarction/recess, and intraplacental T2 dark bands. The decision tree predicted the presence of PAS in the randomly assigned validation cohort with significance (p < 0.001). The sensitivity and the specificity of the decision tree for detecting PAS were 90.0% (95%CI: 53.2-98.9) and 95.5% (95%CI: 89.9-96.8), respectively. Among PAS-related MRI features, placental/uterine bulge and myometrial thinning showed high diagnostic values. In addition, the present decision tree model was shown to be effective in predicting the presence of PAS in cases with placenta previa.

Analysing the Impact of Natural Disasters, Pandemics and Other Crisis on Employment and Workforce Dynamics

This study investigates the impact of recent crises such as natural disasters and pandemics on employment and workforce dynamics using a Random Forest machine learning model. The data was collected through a detailed questionnaire analyzing various factors, including demographic information, employment status, the impact of crises, job security, coping mechanisms, and future outlook. The findings reveal that most respondents are young adults, with a higher percentage engaged in full-time or self-employment roles. Crises, notably COVID-19, led to significant financial impacts such as reduced salaries and layoffs, with many experiencing jobs in security and moderate to high stress levels. A substantial portion of the workforce transitioned to remote work, though many reported inadequate employer support. The analysis of the Random Forest model indicates a moderate performance in predicting workforce dynamics, with accuracy, precision, recall, and F1 scores highlighting the complexity of this issue. The study provides practical recommendations such as enhancing remote work infrastructure, improving financial and mental health support, and promoting workforce adaptability to mitigate the adverse effects of future crises and ensure a resilient workforce.

Dry mass grassland estimation using UAV ultra-wide RGB images

Abstract. Dry mass is an important parameter to optimise grassland management. Traditionally, dry mass values are estimated manually by cutting, drying, and weighing vegetation samples. In large areas of cultivation, this becomes a time-consuming and costly activity. In recent years, many researchers have studied different sensors embedded in Unmanned Aerial Vehicles (UAV) to collect spatial data and estimate biomass using machine learning algorithms for forest and agricultural applications. However, there needs to be more research dealing with estimating production indices for pasture, especially in Brazil, as stated. This study evaluates the feasibility of using the GoPro wide-angle RGB camera on UAVs (Unmanned Aerial Vehicles) to estimate the dry mass of pastures. Different data analysis methods were compared, including the combination of vegetation indices (VIs) values and three-dimensional metrics (3D) extracted from the Canopy Height Model (CHM): all metrics (ALL), three VIs plus four 3D metrics (VI3 + CHM4) and only 3D metrics. Random Forest (RF) machine learning algorithm was used to estimate dry mass. The best results were obtained when merging all the variables from the two flight campaigns, with a coefficient of determination (R2) of 0.80 for the model and a Pearson Correlation Coefficient (PCC) of 0.85 for validation, with a Root Mean Square Error (RMSE%) of 20.5%. In summary, using RGB sensors embedded in UAVs is a promising technique for estimating farm grazing parameters.

Data-driven explainable machine learning for personalized risk classification of myasthenic crisis

ObjectiveMyasthenic crisis (MC) is a critical progression of Myasthenia gravis (MG), requiring intensive care treatment and invasive therapies. Classifying patients at high-risk for MC facilitates treatment decisions such as changes in medication or the need for mechanical ventilation and helps prevent disease progression by decreasing treatment-induced stress on the patient. Here, we investigated whether it is possible to reliably classify MG patients into groups at low or high risk of MC based entirely on routine medical data using explainable machine learning (ML). MethodsIn this single-center pseudo-prospective cohort study, we investigated the precision of ML models trained with real-world routine clinical data to identify MG patients at risk for MC, and identified explainable distinctive features for the groups. 51 MG patients, including 13 MC, were used for model training based on real-world clinical data available from the hospital management system. Patients were classified to high or low risk for MC using Lasso regression or random forest ML models. ResultsThe mean cross-validated AUC classifying MG patients as high or low risk for MC based on simple or compound features derived from real-world clinical data showed a predictive accuracy of 68.8% for a regularized Lasso regression and 76.5% for a random forest model. Studying feature importance across 5100 model runs identified explainable features to distinguish MG patients at high or low risk for MC. Feature importance scores suggested that multimorbidity may play a role in risk classification. ConclusionThis study establishes feasibility and proof-of-concept for risk classification of MC based on real-world routine clinical data using ML with explainable features and variance control at the point of care. Future research on ML-based prediction of MC should include multi-center, multinational data collection, more in-depth data per patient, more patients, and an attention-based ML model to include free-text.

Global patterns in vegetation accessible subsurface water storage emerge from spatially varying importance of individual drivers

Abstract Vegetation roots play an essential role in regulating the hydrological cycle by removing water from the subsurface and releasing it to the atmosphere. However, the present understanding of the drivers of ecosystem-scale root development and their spatial variability globally is limited. This study investigates the varying roles of climate, landscape, and vegetation on the magnitude of root zone storage capacity ( S r ) worldwide, which is defined as the maximum volume of subsurface moisture accessible to vegetation roots. To this aim, we quantified S r and evaluated 21 possible climate, landscape, and vegetation controls for 3612 river catchments worldwide using a random forest machine learning model. Our findings reveal climate as primary, but spatially varying, driver of ecosystem scale S r with landscape and vegetation characteristics playing a minor role. More specifically, we found the mean inter-storm duration as most dominant control of S r globally, followed by mean temperature, mean precipitation, and mean topographic slope. While the inter-storm duration, temperature, and slope exhibit a consistent relation with S r globally, the relation between precipitation and S r varies spatially. Based on this spatial variability, we classified two different regimes: precipitation driven and energy limited. The precipitation-driven regime exhibits a positive relation between precipitation and S r for precipitation of up to 3 mm d − 1 , above which the relation flattens and eventually becomes negative. The energy-limited regime exhibits a strictly negative relation between precipitation and S r . Using the random forest model based on these three dominant climate variables and the landscape variable slope, we generated a global gridded dataset of S r , which closely resembles other global datasets of root characteristics. This suggests that our parsimonious approach based on four globally available variables to estimate S r on a global scale has the potential to be readily and easily integrated into the parameterization of S r in global hydrological and land surface models. This may enhance the accuracy of global predictions of land–atmosphere exchange fluxes and hydrological extremes by providing a robust representation of both spatial and temporal variability in vegetation root characteristics.

Geospatial Hot Spots and Cold Spots in US Cancer Disparities and Associated Risk Factors, 2004-2008 to 2014-2018.

Despite declining cancer death rates in the US, cancer remains the second deadliest disease and disparities persist. Although research has focused on identifying risk factors for cancer deaths and associated disparities, few studies have examined how these relationships vary over time and space. The primary objective of this study was to identify cancer mortality hot spots and cold spots - areas where cancer death rates decreased less than or more than neighboring areas over time. A secondary objective was to identify risk factors of cancer mortality hot spots and cold spots. We analyzed county-level cancer death rates from 2004 through 2008 and 2014 through 2018, exploring disparities in changes over time for socioeconomic and demographic variables. We used hot spot analysis to identify areas with larger decreases (cold spots) and smaller decreases (hot spots) in cancer death rates and random forest machine learning analysis to assess the relative importance of risk factors associated with hot spots and cold spots. We mapped spatial clustering areas. Geospatial analysis showed hot spots predominantly in the Plains states and Midwest and cold spots in the Southeast, Northeast, 2 Mountain West states (Utah and Idaho), and a portion of Texas. Factors with the strongest influence on hot spots and cold spots were unemployment, preventable hospital stays, mammography screening, and high school education. Geospatial disparities in changes in cancer death rates point out the critical role of access to care, socioeconomic position, and health behaviors in persistent cancer mortality disparities. Study results provide insights for interventions and policies that focus on addressing health care access and social determinants of health.

Evaluating long-term reductions in trace metal emissions from shipping in Shanghai

Shipping emissions are a major contributor to air pollution in coastal cities, and Shanghai Port, the busiest port in the world, handles over 40 million TEU annually. To mitigate shipping emissions, China introduced the Domestic Emission Control Area (DECA) policy in phases: DECA 1.0 in 2016 and DECA 2.0 in 2019. DECA 1.0 mandated low-sulfur fuels (0.5 %) at berth and near major ports, down from 1.5 %, and this requirement was extended to a 12-nautical-mile emission control zone in DECA 2.0. In this study, we conducted long-term online measurements of shipping emission tracer of vanadium (V) and nickel (Ni), at the Dian Shan Lake (DSL) supersite, located approximately 50 km from Shanghai waters. The observed V concentration exhibited a strong dependence on wind direction, with higher levels from spring to summer due to more frequent marine winds compared to other seasons. The long-term measurement showed a significant decrease in V concentrations, dropping from an annual mean of 7.08 ng m-³ during DECA 1.0 to 2.64 ng m-³ during DECA 2.0. The represents a year-on-year reduction of 63 % based on measurement. To remove the meteorological impact on the measured concentration, a Random Forest (RF) machine learning model and the Community Multiscale Air Quality (CMAQ 5.4) model were applied under a business-as-usual (BAU) scenario. Both models produced consistent results, showing a reduction of up to 82 % in V concentrations in April with more frequent marine winds, confirming regional air quality improvements post-DECA 2.0. Additionally, we found that Ni has other sources that require further control beyond shipping. The study highlights the importance of long-term measurements for understanding the impact of air quality policies on emission patterns.

Characterization of Brain Abnormalities in Lactational Neurodevelopmental Poly I:C Rat Model of Schizophrenia and Depression Using Machine-Learning and Quantitative MRI.

A recent neurodevelopmental rat model, utilizing lactational exposure to polyriboinosinic-polyribocytidilic acid (Poly I:C) leads to mimics of behavioral phenotypes resembling schizophrenia-like symptoms in male offspring and depression-like symptoms in female offspring. To identify mechanisms of neuronal abnormalities in lactational Poly I:C offspring using quantitative MRI (qMRI) tools. Prospective. Twenty Poly I:C rats and 20 healthy control rats, age 130 postnatal day. 7 T. Multiflip-angle FLASH protocol for T1 mapping; multi-echo spin-echo T2-mapping protocol; echo planar imaging protocol for diffusion tensor imaging. Nursing dams were injected with the viral mimic Poly I:C or saline (control group). In adulthood, quantitative maps of T1, T2, proton density, and five diffusion metrics were generated for the offsprings. Seven regions of interest (ROIs) were segmented, followed by extracting 10 quantitative features for each ROI. Random forest machine learning (ML) tool was employed to identify MRI markers of disease and classify Poly I:C rats from healthy controls based on quantitative features. Poly I:C rats were identified from controls with an accuracy of 82.5 ± 25.9% for females and 85.0 ± 24.0% for males. Poly I:C females exhibited differences mainly in diffusion-derived parameters in the thalamus and the medial prefrontal cortex (MPFC), while males displayed changes primarily in diffusion-derived parameters in the corpus callosum and MPFC. qMRI shows potential for identifying sex-specific brain abnormalities in the Poly I:C model of neurodevelopmental disorders. NA TECHNICAL EFFICACY: Stage 2.

Detection and risk stratification of cardiac amyloidosis patients by integration of imaging and non-imaging data using a machine learning approach

Abstract Background With the advent of amyloid-targeting therapies, early and reliable diagnosis as well as precise risk estimation of cardiac amyloidosis (CA) have become of substantial importance. While the current diagnostic approach relies on difficult-to-standardize, visual interpretation of 99mTc-scintigraphy, risk assessment is largely based on a combination of blood and imaging parameters. Purpose Aim of this study is to assess the possibility of machine learning-based integration of scintigraphy, echocardiography, and non-imaging parameters such as blood parameters and comorbidities for the detection and risk stratification of patients with CA. Methods This study included all patients who underwent 99mTc-scintigraphy at a university affiliated tertiary care centre between 2010 and 2023. Two machine learning models were developed, and the relative predictive importance of their parameters were assessed (Figure 1): First, a model to detect patients with CA-suggestive uptake (Perugini grade 2/3) on 99mTc-scintigraphy scans; Second, a model to assess the risk of patients with CA-suggestive uptake for future heart failure hospitalization (HFH). A total of 58 features were extracted from electronic health records including blood, echocardiographic, demographic parameters and comorbidities. Scintigraphy imaging features were extracted from the raw imaging data using a deep learning approach. For the time to HFH prediction, a random survival forest machine learning model was employed. Results Overall, 12 380 consecutively enrolled patients were included, 279 (2.3%) of which were affected by CA-suggestive uptake. The machine learning model showed higher accuracy for the detection of CA (AUC 0.96 [95% CI 0.95-0.97], sensitivity 0.78 [95% CI 0.77-0.80] and specificity 0.99 [95% CI 0.99-0.99]. In the outcome analysis, the machine model showed good accuracy in predicting future HFH (C-index 0.71 [95% CI 0.68-0.75]). Parameter importance for the time to event analysis revealed right ventricular diameter, previous diagnosis of chronic heart failure and creatine kinase as the three most important predictors (Figure 2). Conclusions Detection of patients with cardiac amyloidosis and their risk estimation for heart failure can be aided using machine learning-based integration of imaging and non-imaging data. Our findings may guide novel approaches for assessing disease progression and treatment response.

Random Forest Prognostication of Survival and 6-Month Outcome In Pediatric Patients Following Decompressive Craniectomy For Traumatic Brain Injury

There is a dearth of literature regarding prognostic and predictive factors for outcome following pediatric decompressive craniectomy (DC) following traumatic brain injury (TBI). The aim of this study was to develop a random forest machine learning algorithm to predict outcomes following DC in pediatrics. This is a multi-institutional retrospective study assessing the 6-month postoperative outcome in pediatric patients that underwent DC. We developed a machine learning model using classification and survival random forest algorithms (CRF and SRF respectively) for the prediction of outcomes. Data was collected on clinical signs, radiographic studies, and laboratory studies. The outcome measures for the CRF were mortality and good or bad outcome based on Glasgow Outcome Scale (GOS) at 6-months. A GOS score of 4 or higher indicated a good outcome. The outcomes for the SRF model assessed mortality at during the follow-up period. A total of 40 pediatric patients were included. There was a hospital mortality rate of 27.5%, and 75.8% of survivors had a good outcome at 6-month follow up. The CRF for 6-month mortality had a ROC AUC of 0.984; whereas, the 6-month good/bad outcome had a ROC AUC of 0.873. The SRF was trained at the 6-month timepoint with a ROC AUC of 0.921. CRF and SRF models successfully predicted 6-month outcomes and mortality following DC in pediatric TBI patients. These results suggest random forest models may be efficacious for predicting outcome in this patient population.

Predicting ECGs masquerading AS STEMI with machine learning

Abstract Background Advancements in the management of ST elevation myocardial infarction (STEMI) in the last 15 years include early STEMI identification in the prehospital setting leading to expedited cardiac catheterisation laboratory (CCL) activation or fibrinolytic therapy. We have previously reported of patients who met the Glasgow algorithm (GA) ECG criteria for STEMI; 50% did not actually have an acute STEMI according to the 4th universal definition of MI (4UDMI)1. Identifying such patients could avoid unnecessary emergency CCL activations or inappropriate fibrinolytic administration. Methods From June 2010 to October 2021, we identified 2992 patients with ECG transmissions to Liverpool Hospital, Sydney Australia which met GA criteria for STEMI. These were adjudicated according to the 4UDMI for true-positive STEMI. We used univariate and multivariate logistic regression to identify likely patient characteristics that were predictive of true-positive STEMI. We then improved model performance using receiver-operator curve (ROC) with simple random forest machine learning algorithms. Results Among 2994 patients with prehospital ECG transmissions that met GA-criteria, 1553 (51%) were adjudicated as true-positive STEMI, which is comparable to previous findings1. All standard modifiable and non-modifiable risk factors were independently predictive of true-positive STEMI; however, in multivariate logistic regression gender, hypertension and diabetes lost significance. Dyslipidemia and current smoking were positively predictive of true-positive STEMI status (OR = 1.64 and 1.32 respectively), whereas age and previous MI were negatively associated (OR = 0.98 (per year increment) and 0.4). Using these identified predictors in a random forest to predict true-positive STEMI yielded an AUC of 0.65 (Figure 1). Conclusion Specific clarification of dyslipidaemia and current smoking status at first medical contact may be the most salient in predicting a true-positive STEMI during prehospital assessment which may assist in avoiding unnecessary emergency CCL activation or inappropriate fibrinolytic administration.ROC of ST elevation STEMI recognition

Machine learning-based prediction of pulmonary embolism to reduce unnecessary computed tomography scans in gastrointestinal cancer patients: a retrospective multicenter study

This study aimed to develop a machine learning (ML) model for predicting pulmonary embolism (PE) in patients with gastrointestinal cancers, a group at increased risk for PE. We conducted a retrospective, multicenter study analyzing patients who underwent computed tomographic pulmonary angiography (CTPA) between 2010 and 2020. The study utilized demographic and clinical data, including the Wells score and D-dimer levels, to train a random forest ML model. The model’s effectiveness was assessed using the area under the receiver operating curve (AUROC). In total, 446 patients from hospital A and 139 from hospital B were included. The training set consisted of 356 patients from hospital A, with internal validation on 90 and external validation on 139 patients from hospital B. The model achieved an AUROC of 0.736 in hospital A and 0.669 in hospital B. The ML model significantly reduced the number of patients recommended for CTPA compared to the conventional diagnostic strategy (hospital A; 100.0% vs. 91.1%, P < 0.001, hospital B; 100.0% vs. 93.5%, P = 0.003). The results indicate that an ML-based prediction model can reduce unnecessary CTPA procedures in gastrointestinal cancer patients, highlighting its potential to enhance diagnostic efficiency and reduce patient burden.

Machine learning-based assessment of regional-scale variation of landslide susceptibility in central Vietnam.

Recurrent landslide events triggered by typhoons and tropical storms over Vietnam pose a longstanding threat to the nation's population and infrastructure. Changes in hydroclimatic conditions, especially the growing intensity and frequency of storms, have elevated landslide susceptibility in many parts of the country. This research examines the spatio-temporal variations in landslide susceptibility across central Vietnam over several years, using multi-temporal landslide inventories from Typhoon Ketsana (2009), Tropical Storm Podul (2013), and Typhoon Molave (2020). Additionally, the research explores the impact of individual landslide causative factors on the probabilistic occurrences of landslides. The post-event landslide susceptibility models of these three climate extreme events were developed using nine causative factors and a Random Forest machine learning algorithm. The results indicate a notable areal expansion of high to very high landslide susceptibility in the northern and eastern regions and a moderate reduction in the central and southern areas during the post-Molave period compared to the post-Ketsana period. These changes may be early indicators of increasing landslide susceptibility in response to changing hydro-climatic conditions. The research found that annual average rainfall and topographic elevation are the two most important variables influencing landslide prediction, showing a nonlinear relationship with landslide probability. The landslide susceptibility models achieved high Area Under the Receiver Operating Characteristic Curve (AUC) (>95%), accuracy (>89%), and sensitivity (>90%) scores, signifying the robustness of the models. Additionally, the uncertainty of the models was quantified and spatially mapped. This multi-temporal analysis of landslide susceptibility is crucial for understanding the regional susceptibility trends and identifying areas with increasing, decreasing, and consistently high susceptibility to landslides. These insights are invaluable for prioritizing mitigation and risk reduction strategies in landslide-prone regions and guiding appropriate land use planning.