Random Forest Machine Learning Classifier Research Articles

Feasibility of using an automated analysis of formulation effort in patients’ spoken seizure descriptions in the differential diagnosis of epileptic and nonepileptic seizures

ObjectiveThere are three common causes of Transient Loss of Consciousness (TLOC), syncope, epileptic and psychogenic nonepileptic seizures (PNES). Many individuals who have experienced TLOC initially receive an incorrect diagnosis and inappropriate treatment. Whereas syncope can be distinguished relatively easily with a small number of “yes”/”no” questions, the differentiation of the other two causes of TLOC is more challenging. Previous qualitative research based on the methodology of Conversation Analysis has demonstrated that the descriptions of epileptic seizures contain more formulation effort than accounts of PNES. This research investigates whether features likely to reflect the level of formulation effort can be automatically elicited from audio recordings and transcripts of speech and used to differentiate between epileptic and nonepileptic seizures. MethodVerbatim transcripts of conversations between patients and neurologists were manually produced from video and audio recordings of 45 interactions (21 epilepsy and 24 PNES). The subsection of each transcript containing the person's account of their first seizure was manually extracted for the analysis. Seven automatically detectable features were designed as markers of formulation effort. These features were used to train a Random Forest machine learning classifier. ResultThere were significantly more hesitations and repetitions in descriptions of epileptic than nonepileptic seizures. Using a nested leave-one-out cross validation approach, 71% of seizures were correctly classified by the Random Forest classifier. DiscussionThis pilot study provides proof of principle that linguistic features that have been automatically extracted from audio recordings and transcripts could be used to distinguish between epileptic seizures and PNES and thereby contribute to the differential diagnosis of TLOC. Future research should explore whether additional observations can be incorporated into a diagnostic stratification tool and compare the performance of these features when they are combined with additional information provided by patients and witnesses about seizure manifestations and medical history.

Evaluation of different machine learning algorithms for pearl millet discrimination using multi-sensor SAR data

This study aims to assess the potential of multi-temporal Sentinel-1 and RADARSAT-2 data for pearl millet discrimination using various classifiers—machine learning and knowledge-based decision tree (m-DT and k-DT), Support Vector Machine (SVM), and Random Forest (RF). Results show that the RF classifier outperformed the other classifiers in terms of overall accuracy (OA) and kappa coefficient. The highest Overall Accuracy (OA) of 89.1% and kappa coefficients of 0.822 and 0.814 were achieved by the RF classifier for both the Sentinel-1 and RADARSAT-2 data, respectively. The Variable Importance (VI) of the RF classifier revealed that the polarimetric parameters—volume scattering, double bounce, entropy and alpha angle were crucial for pearl millet discrimination. Polarimetric parameters, when ingested with the RF machine learning classifier, achieved better classification accuracies.

Mapping the Dynamics of Winter Wheat in the North China Plain from Dense Landsat Time Series (1999 to 2019)

Monitoring spatio-temporal changes in winter wheat planting areas is of high importance for the evaluation of food security. This is particularly the case in China, having the world’s largest population and experiencing rapid urban expansion, concurrently, it puts high pressure on food demands and the availability of arable land. The relatively high spatial resolution of Landsat is required to resolve the historical mapping of smallholder wheat fields in China. However, accurate Landsat-based mapping of winter wheat planting dynamics over recent decades have not been conducted for China, or anywhere else globally. Based on all available Landsat TM/ETM+/OLI images (~28,826 tiles) using Google Earth Engine (GEE) cloud computing and a Random Forest machine-learning classifier, we analyzed spatio-temporal dynamics in winter wheat planting areas during 1999–2019 in the North China Plain (NCP). We applied a median value of 30-day sliding windows to fill in potential data gaps in the available Landsat images, and six EVI-based phenological features were then extracted to discriminate winter wheat from other land cover types. Reference data for training and validation were extracted from high-resolution imagery available via Google Earth™ online mapping service, Sentinel-2 and Landsat imagery. We ran a sensitivity analysis to derive the optimal training sample class ratio (β = 1.8) accounting for the unbalanced distribution of land-cover types. We mapped winter wheat planting areas for 1999–2019 with overall accuracies ranging from 82% to 99% and the user’s/producer’s accuracies of winter wheat range between 90% and 99%. We observed an overall increase in winter wheat planting areas of 1.42 × 106 ha in the NCP as compared to the year 2000, with a significant increase in the Shandong and Hebei provinces (p < 0.05). This result contrasts the general discourse suggesting a decline in croplands (e.g., rapid urbanization) and climate change-induced unfavorable cropping conditions in the NCP. This suggests adjustments of the winter wheat planting area over time to satisfy wheat supply in relation to food security. This study highlights the application of Landsat images through GEE in documenting spatio-temporal dynamics of winter wheat planting areas for adequate management of cropping systems and assessing food security in China.

Accurate Classification of Pediatric Colonic Inflammatory Bowel Disease Subtype Using a Random Forest Machine Learning Classifier.

The pediatric inflammatory bowel disease (PIBD) classes algorithm was developed to bring consistency to labelling of colonic IBD, but labels are exclusively based on features atypical for ulcerative colitis (UC). The aim of the study was to develop an algorithm and identify features that discriminate between pediatric UC and colonic Crohn disease (CD). Baseline clinical, endoscopic, radiologic, and histologic data, including the PIBD class features in 74 colonic IBD (56: UC, 18: colonic CD) patients were collected. The PIBD class features and additional features common to UC were used to perform initial clustering, using similarity network fusion (SNF). We trained a Random Forest (RF) classifier on the full dataset and used a leave-one-out approach to evaluate model accuracy. The top-features were used to build a new classifier, which we tested on 15 previously unused patients. We then performed clustering with SNF on the top RF features and assessed ability to discriminate between UC and colonic-CD independent of a supervised model. The initial SNF clustering with 58 patients demonstrated 2 groups: group 1 (n = 39, 90% UC) and group 2 (n = 19, 68% colonic-CD). Our RF classifier correctly labelled 97% of the 58 patients based on leave-one-out cross validation and identified the 7 most important features (3 histological and 4 endoscopic) to clinically distinguish these groups. We trained a new RF classifier with the top 7 features and found 100% accuracy in a set of 15 held-out patients. Finally, post hoc clustering with these 7 features revealed 2 groups of patients: group 1 (n = 55, 98% UC) and group 2 (n = 18, 94% colonic-CD). A combination of supervised and unsupervised analyses identified a short list of features, which consistently distinguish UC from colonic CD. Future directions include validation in other populations.

An Efficient Phishing Website Detection Plugin Service for Existing Web Browsers Using Random Forest Classifier

An efficient phishing website detection plugin service was developed using machine learning technique based on the prevalent phishing threat while using existing web browsers in critical online transactions. The study gathered useful information from 27 published articles and dataset consisting of 11,000 data points with 30 features downloaded from phishtank. A unique architectural framework for detecting phishing websites was designed using random forest machine learning classifier based the aim and objectives of the study. The model was trained with 90% (9,900) of the dataset and tested with 10% (1,100) using Python programming language for better efficiency. Microsoft Visual Studio Code, Jupiter Notebook, Anaconda Integrated Development Environment, HTML/CSS and JavaScript was used in developing the frontend of the model for easy integration into existing web browsers. The proposed model was also modeled using use-case and sequence diagrams to test its internal functionalities. The result revealed that the proposed model had an accuracy of 0.96, error rate of 0.04, precision of 0.97, recall value of 0.99 and f1-score of 0.98 which far outperform other models developed based on literatures. Future recommendations should focus on improved security features, more phishing adaptive learning properties, and so on, so that it can be reasonably applied to other web browsers in accurately detecting real-world phishing situations using advanced algorithms such as hybridized machine learning and deep learning techniques.

Gradient and statistical features-based prediction system for COVID-19 using chest X-ray images

As per data available on WHO website, COVID-19 patients on 20 June 2020 have surpassed the figure of 8.7 million globally and around 4.6 lakhs have lost their life. The most common diagnostic test for COVID-19 detection is a Polymerase Chain Reaction (PCR) test. In highly populated developing countries like Brazil, India etc., there has been a severe shortage of PCR test-kits. Furthermore, the PCR-test is very specific and has lower sensitivity. In this research work, authors have designed a decision support system based on statistical features and edge maps of X-ray images to detect COVID-19 virus in a patient. Online available data sets of chest X-ray images have been used to train and test decision tree, K-nearest neighbour's, random forest, and multilayer perceptron machine learning classifiers. From the experimental results, it has found that the multilayer perceptron achieved 94% accuracy which is higher than the other classifiers.

An exploratory proteomic study delineating the local and systemic immune-oncologic profile of urinary bladder cancer patients

Cancer of the urinary bladder (UBC) primarily derives from the urothelium, encompassing the inner surface of the bladder. Tumors infiltrating the detrusor muscle are categorized as muscle-invasive bladder cancer (MIBC) and constitutes 20-25% of all newly diagnosed UBC. These tumors are more likely to spread to peripheral organs and lymph nodes, compared to the more prevalent (75-80%) non-muscle invasive bladder cancer subtype (NMIBC). To this end, cystoscopy and urine cytology are the current standards for diagnosing and surveying UBC, which provides limited information regarding their immunological profile. We performed a targeted human protein biomarker analysis from collected plasma and urine from a well-categorized UBC patient cohort. In short, 92 immuno-oncology related proteins were investigated by a novel Proximity Extension Assay multiplex technology (Olink). The assayed proteins depicted a heterogenic immune profile. Neither PCA or unsupervised hierarchical clustering identified distinct groups associated with clinical features. However, a direct comparison between non-invasive and invasive cases resulted in 4 differentially expressed proteins (p<0.05). MIBC patients had lower urine levels of the TNF-receptor superfamily members CD27 and CD40, and a systemic increase of matrix metallopeptidase 7 (MMP7) and C-C motif chemokine ligand 23 (CCL23). We further developed and trained a random forest machine learning classifier on our plasma samples. It consistently ranked CCL23 and MMP7 as the most important markers for invasiveness. When extended to survival analysis, we moreover found MMP levels to best predict a dismal survival status with MMP12 being the highest ranked variable. A finding supported by MMP12 being the most significant protein in individual cox-regression survival analyses. This study highlights the heterogenic nature of the immunogenic landscape in UBC and its potential role in immunotherapy treatment responses. Our findings indicate that systemic MMP levels should be further explored as a potential liquid biopsy marker for patient stratification purposes. The role of MMPs in UBC may be linked to an invasive tumor characteristic, as shown by previous UBC studies where high levels of MMPs correlated to reduced overall survival.

Extraction of built-up area using multi-sensor data—A case study based on Google earth engine in Zhejiang Province, China

ABSTRACT Accurate and up-to-date built-up area mapping is of great importance to the science community, decision-makers, and society. Therefore, satellite-based, built-up area (BUA) extraction at medium resolution with supervised classification has been widely carried out. However, the spectral confusion between BUA and bare land (BL) is the primary hindering factor for accurate BUA mapping over large regions. Here we propose a new methodology for the efficient BUA extraction using multi-sensor data under Google Earth Engine cloud computing platform. The proposed method mainly employs intra-annual satellite imagery for water and vegetation masks, and a random-forest machine learning classifier combined with auxiliary data to discriminate between BUA and BL. First, a vegetation mask and water mask are generated using NDVI (normalized differenced vegetation index) max in vegetation growth periods and the annual water-occurrence frequency. Second, to accurately extract BUA from unmasked pixels, consisting of BUA and BL, random-forest-based classification is conducted using multi-sensor features, including temperature, night-time light, backscattering, topography, optical spectra, and NDVI time-series metrics. This approach is applied in Zhejiang Province, China, and an overall accuracy of 92.5% is obtained, which is 3.4% higher than classification with spectral data only. For large-scale BUA mapping, it is feasible to enhance the performance of BUA mapping with multi-temporal and multi-sensor data, which takes full advantage of datasets available in Google Earth Engine.

Development and validation of an MRI-based radiomic nomogram to distinguish between good and poor responders in patients with locally advanced rectal cancer undergoing neoadjuvant chemoradiotherapy.

In the clinical management of patients with locally advanced rectal cancer (LARC), the early identification of poor and good responders after neoadjuvant chemoradiotherapy (N-CRT) is essential. Therefore, we developed and validated predictive models including MRI findings from the structured report template, clinical and radiomics parameters to differentiate between poor and good responders in patients with locally advanced rectal cancer who underwent neoadjuvant chemoradiotherapy. Preoperative multiparametric MRI from 183 patients with locally advanced rectal cancer (122 in the training cohort, 61 in the validation cohort) was included in this retrospective study. After preprocessing, radiomic features were extracted and two methods of feature selection was applied to reduce the number of radiomics features. Logistic regression (LR) and random forest (RF) machine learning classifiers were trained to identify good responders from poor responders. Multivariable logistic regression analysis was used to incorporate the radiomic signature and clinical risk factors into a nomogram. Classifier performance was evaluated using the area under the curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). For the differentiation of poor and good responders, the radiomics model with an LR classifier achieved AUCs of 0.869 and 0.842 for the training and validation cohorts, respectively. The nomogram showed the highest reproducibility and prognostic ability in the training and validation cohorts, with AUCs of 0.923 (95% confidence interval, 0.872-0.975) and 0.898 (0.819-0.978), respectively. Additionally, the nomogram achieved significant risk stratification of patients in respect to progression free survival (PFS). The nomogram accurately differentiated good and poor responders in patients with LARC undergoing N-CRT, and showed significant performance for predicting PFS.

Reducing False-Positive Results in Newborn Screening Using Machine Learning.

Newborn screening (NBS) for inborn metabolic disorders is a highly successful public health program that by design is accompanied by false-positive results. Here we trained a Random Forest machine learning classifier on screening data to improve prediction of true and false positives. Data included 39 metabolic analytes detected by tandem mass spectrometry and clinical variables such as gestational age and birth weight. Analytical performance was evaluated for a cohort of 2777 screen positives reported by the California NBS program, which consisted of 235 confirmed cases and 2542 false positives for one of four disorders: glutaric acidemia type 1 (GA-1), methylmalonic acidemia (MMA), ornithine transcarbamylase deficiency (OTCD), and very long-chain acyl-CoA dehydrogenase deficiency (VLCADD). Without changing the sensitivity to detect these disorders in screening, Random Forest-based analysis of all metabolites reduced the number of false positives for GA-1 by 89%, for MMA by 45%, for OTCD by 98%, and for VLCADD by 2%. All primary disease markers and previously reported analytes such as methionine for MMA and OTCD were among the top-ranked analytes. Random Forest’s ability to classify GA-1 false positives was found similar to results obtained using Clinical Laboratory Integrated Reports (CLIR). We developed an online Random Forest tool for interpretive analysis of increasingly complex data from newborn screening.

Classification of solid fuels with machine learning

In energy applications, fuels are processed in various ways according to their types. Experiments conducted with non-optimal processing procedures cause waste of the materials and may lead to inaccurate conclusions. Moreover, classifying the fuel material becomes harder when the material is already processed or collected from an environment that makes it difficult to discern. Thus, with the constantly developing and diversifying energy applications, the need for a framework that can classify solid fuels is increased. In this study, we used machine learning based approach to classify fuels with the use of proximate analysis results, i.e., fixed carbon, volatile matter and ash contents. We collected a data set from the literature and group them into four classes, i.e., coals, woods, agricultural residue and manufactured biomass. Then, K-nearest neighbor, support vector machine and random forest machine learning classifiers are employed to develop prediction models. Furthermore, hierarchical classification approach is taken to combine each classifier’s advantages with integration of expert opinion to create a complete and highly accurate classifier framework which can classify an unknown fuel into one of the four categories. K-fold cross validation is used to evaluate classifiers’ performances in unbiased manner. With hierarchical classifier, % 96 and %92 classification accuracy is achieved for training and testing phases, respectively. Source code of the proposed hierarchical classifier framework is also provided.

Land use/land cover change and land surface temperature of Ibadan and environs, Nigeria.

Rapid urbanization is having a considerable impact on various aspects of living, thereby altering the biophysical environment. This study adopted the use of remote sensing technique and geographical information system (GIS) to analyse the relationship between changing land use/land cover and land surface temperature in a rapidly urbanizing tropical cityof Ibadan between 1984 and 2019. Landsat series TM, ETM+, and OLI satellite imageries of Ibadan region city for 1984, 2002, and 2019, respectively, were obtained from the US Geological Survey (USGS) Landsat series of Earth Observation satellites accessible on the Google earth engine (GEE) platform. Supervised classification was done using a random forest (RF) machine learning classifier in the GEE platform. Surface emissivity maps were obtained from the normalized difference vegetation index (NDVI) thresholds method and land cover information. The surface emissivity based on NDVI classes was used to retrieve land surface temperature (LST). The results showed an increase in urban cover from 341.72km2 in 1984 to 520.58km2 in 2019 with an average increase in land surface temperature from 17°C to 38°C, respectively. Temperature sampling in the north-south and west-east transect revealed that highly urbanized areas located at the city centre of Ibadan have the highest LST of about 38°C. It dissipates to about 19°C at the suburb that is less built up. A significant negative relationship exists between the health condition of vegetation (NDVI) and LST with a correlation coefficient of r = - 0.95. The study confirms the potential application of GIS and remote sensing for detecting urban growth as well as relates growth impact to LST, thereby suggesting that fitting strategies will be important for the sustainable management of the urban areas.

Generalization considerations and solutions for point cloud hillslope classifiers

Point cloud classifiers have the potential to rapidly perform landscape characterization for a variety of applications. The generalization (i.e., transferability to new sites) of such classifiers could improve their accessibility and usefulness for both engineers and researchers alike, but guidelines for classifier generalization are lacking in the literature. This study develops and applies a Random Forest machine learning classifier for Terrestrial Laser Scanning (TLS) point clouds, and generalizes the classifier to point clouds from several different locations. The classifier is trained to identify basic hillslope topographic features, including vegetation, soil, talus, and bedrock using multi-scale geometric features of the point cloud. Four rock and soil slopes in western Colorado were scanned using TLS. Generalization experiments were performed testing point density, occlusion, and between-site domain variance factors, and all factors showed a significant influence on generalization accuracy. Several methods for improving classifier generalization accuracy were tested and compared, including combining training data from multiple sites, imposing probability thresholds, and a Domain Adaptation methodology known as Active Learning. It was found that incorporating data from multiple sites resulted in improved generalization accuracy, but in most cases the largest improvements in accuracy were associated with adding new training data from the target site. In this case, using Active Learning resulted in significant accuracy improvements with an over 90% reduction in the number of added training points. The results suggest that scanning characteristics are important factors in classifier generalization accuracy, but their effects can be mitigated by using the techniques described herein.

Inflammatory lesions and brain tumors: is it possible to differentiate them based on texture features in magnetic resonance imaging?

Background:Neuroimaging strategies are essential to locate, to elucidate the etiology, and to the follow up of brain disease patients. Magnetic resonance imaging (MRI) provides good cerebral soft-tissue contrast detection and diagnostic sensitivity. Inflammatory lesions and tumors are common brain diseases that may present a similar pattern of a cerebral ring enhancing lesion on MRI, and non-enhancing core (which may reflect cystic components or necrosis) leading to misdiagnosis. Texture analysis (TA) and machine learning approaches are computer-aided diagnostic tools that can be used to assist radiologists in such decisions.Methods:In this study, we combined texture features with machine learning (ML) methods aiming to differentiate brain tumors from inflammatory lesions in magnetic resonance imaging. Retrospective examination of 67 patients, with a pattern of a cerebral ring enhancing lesion, 30 with inflammatory, and 37 with tumoral lesions were selected. Three different MRI sequences and textural features were extracted using gray level co-occurrence matrix and gray level run length. All diagnoses were confirmed by histopathology, laboratorial analysis or MRI. Results:The features extracted were processed for the application of ML methods that performed the classification. T1-weighted images proved to be the best sequence for classification, in which the differentiation between inflammatory and tumoral lesions presented high accuracy (0.827), area under ROC curve (0.906), precision (0.837), and recall (0.912). Conclusion:The algorithm obtained textures capable of differentiating brain tumors from inflammatory lesions, on T1-weghted images without contrast medium using the Random Forest machine learning classifier.

Comparison between convolutional neural networks and random forest for local climate zone classification in mega urban areas using Landsat images

The Local Climate Zone (LCZ) scheme is a classification system providing a standardization framework to present the characteristics of urban forms and functions, especially for urban heat island (UHI) research. Landsat-based 100 m resolution LCZ maps have been classified by the World Urban Database and Portal Tool (WUDAPT) method using a random forest (RF) machine learning classifier. Some studies have proposed modified RF and convolutional neural network (CNN) approaches. This study aims to compare CNN with an RF classifier for LCZ mapping in great detail. We designed five schemes (three RF-based schemes (S1–S3) and two CNN-based ones (S4–S5)), which consist of various combinations of input features from bitemporal Landsat 8 data over four global mega cities: Rome, Hong Kong, Madrid, and Chicago. Among the five schemes, the CNN-based one with the incorporation of a larger neighborhood information showed the best classification performance. When compared to the WUDAPT workflow, the overall accuracies for entire land cover classes (OA) and for urban LCZ types (i.e., LCZ1-10; OAurb) increased by about 6–8% and 10–13%, respectively, for the four cities. The transferability of LCZ models for the four cities were evaluated, showing that CNN consistently resulted in higher accuracy (increased by about 7–18% and 18–29% for OA and OAurb, respectively) than RF. This study revealed that the CNN classifier classified particularly well for the specific LCZ classes in which buildings were mixed with trees or buildings or plants were sparsely distributed. The research findings can provide a basis for guidance of future LCZ classification using deep learning.

Comparison of EMD, DWT and WPD for the localization of epileptogenic foci using Random Forest classifier

Localization of epileptogenic foci is an essential phase in surgical treatment planning using the earliest time detection of the seizure onset in the recordings of electroencephalogram (EEG). These recordings are defined as the areas of the brain which can be surgically removed to reach control of seizure. The characteristics of the brain area affected by partial epilepsy can be analyzed using focal and non-focal EEG signals. In this work, a method for the classification of focal and non-focal EEG signals is presented to compare different feature extraction methods combined with multi-scale principal component analysis (MSPCA) denoising and Random Forest (RF) as a machine learning technique. After de-noising, different feature extraction methods (EMD, DWT and WPD) are applied. The performance of RF classifier is measured according to accuracy, the area under the Receiver Operating Characteristics (ROC) curve (AUC) and F-measure. EEG based focal region localization with Wavelet Packet Decomposition (WPD) feature extraction and RF classifier reach 99.92% accuracy. Hence, WPD combined with RF machine learning classifier can be used to differentiate the focal and non-focal EEG signals.

Abstract 1672: Clinical validation of a genomics-based classifier to predict tissue of origin from targeted tumor sequencing

Abstract Introduction Various classes of genomic alterations can now be robustly detected from targeted Next Generation Sequencing (NGS)-based tumor profiling assays. Despite recent advances in genome-directed cancer therapies and lineage-agnostic basket clinical trials, tissue of origin remains a critical determinant of tumor biology and therapeutic sensitivity. Therefore, efficiently harnessing the mutational information to predict tissue of origin from NGS data can further aid diagnosis and treatment. Methods We have developed a novel random forest machine learning classifier that infers tissue of origin from the mutational features of each tumor NGS profile. We trained and validated the model using prospective sequencing data from &amp;gt;30,000 patients profiled by MSK-IMPACT, a custom FDA-authorized clinical sequencing assay. An initial version of this classifier trained on a smaller cohort, presented previously, has been further optimized to incorporate additional tumor types, genomic features, mutational signatures, and accessibility features for clinicians. Altogether, the model evaluates a comprehensive range of mutational features to generate predictions. In addition, we developed a framework for the prospective clinical implementation of our method that allows for extension to the expanding MSK-IMPACT cohort and utilization at the point of care. Results Overall, we predicted the correct cancer type in 74% of cases, with nearly half of cases predicted with high confidence (&amp;gt;95%). In order to make this tool accessible to pathologists for real-time diagnostic and treatment decisions, we also have implemented APIs that transmit our classifier predictions to the cBioPortal for Cancer Genomics and the MPath console from MSK Molecular Diagnostics Service. With the additional samples and genomic features, our model predictions have been improved and are considered during clinical review and sign-out. This practice has brought about critical diagnostic changes and orthogonal validation in several cases. In one case, a patient referred to MSK with a misdiagnosis of metastatic breast cancer to the bladder underwent MSK-IMPACT sequencing for her primary breast tumor and bladder lesion. The model predictions, confirmed by orthogonal testing, classified two lesions as independent primaries of breast and bladder origins, leading to significant changes in treatment regimens. Conclusion Our work delineates the framework of utilizing and optimizing machine learning models on NGS-based sequencing data to aid diagnosis and treatment. The same framework can be applied to cell-free DNA sequencing, which will capture tumor spatial heterogeneity and improve classifier performance. These results indicate that leveraging machine learning to predict tissue of origin complements conventional histologic review to provide integrated diagnoses, often with critical therapeutic implications. Citation Format: Youyun Zheng, Alexander V. Penson, Niedzica Camacho, Evan Biederstedt, Ahmet Zehir, Cyriac Kandoth, Aijazuddin Syed, Anna M. Varghese, Hikmat A. Al-Ahmadie, Nikolaus Schultz, Marc Ladanyi, David B. Solit, David S. Klimstra, David Hyman, Barry Taylor, Michael F. Berger. Clinical validation of a genomics-based classifier to predict tissue of origin from targeted tumor sequencing [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 1672.

Mapping Coarse Woody Debris with Random Forest Classification of Centimetric Aerial Imagery

Coarse woody debris (CWD; large parts of dead trees) is a vital element of forest ecosystems, playing an important role in nutrient cycling, carbon storage, fire fuel, microhabitats, and overall forest structure. However, there is a lack of effective tools for identifying and mapping both standing (snags) and downed (logs) CWD in complex natural settings. We applied a random forest machine learning classifier to detect CWD in centimetric aerial imagery acquired over a 270-hectare study area in the boreal forest of Alberta, Canada. We used a geographic object-based image analysis (GEOBIA) approach in the classification with spectral, spatial, and LiDAR (light detection and ranging)-derived height predictor variables. We found CWD to be detected with great accuracy (93.4 ± 4.2% completeness and 94.5 ± 3.2% correctness) when training samples were located within the application area, and with very good accuracy (84.2 ± 5.2% completeness and 92.2 ± 3.2% correctness) when training samples were located outside the application area. The addition of LiDAR-derived variables did not increase the accuracy of CWD detection overall (&lt;2%), but aided significantly (p &lt; 0.001) in the distinction between logs and snags. Foresters and researchers interested in CWD can take advantage of these novel methods to produce accurate maps of logs and snags, which will contribute to the understanding and management of forest ecosystems.

Combining newborn metabolic and DNA analysis for second-tier testing of methylmalonic acidemia

PurposeImproved second-tier tools are needed to reduce false-positive outcomes in newborn screening (NBS) for inborn metabolic disorders on the Recommended Universal Screening Panel (RUSP).MethodsWe designed an assay for multiplex sequencing of 72 metabolic genes (RUSPseq) from newborn dried blood spots. Analytical and clinical performance was evaluated in 60 screen-positive newborns for methylmalonic acidemia (MMA) reported by the California Department of Public Health NBS program. Additionally, we trained a Random Forest machine learning classifier on NBS data to improve prediction of true and false-positive MMA cases.ResultsOf 28 MMA patients sequenced, we found two pathogenic or likely pathogenic (P/LP) variants in a MMA-related gene in 24 patients, and one pathogenic variant and a variant of unknown significance (VUS) in 1 patient. No such variant combinations were detected in MMA false positives and healthy controls. Random Forest–based analysis of the entire NBS metabolic profile correctly identified the MMA patients and reduced MMA false-positive cases by 51%. MMA screen-positive newborns were more likely of Hispanic ethnicity.ConclusionOur two-pronged approach reduced false positives by half and provided a reportable molecular finding for 89% of MMA patients. Challenges remain in newborn metabolic screening and DNA variant interpretation in diverse multiethnic populations.

The Classification of Grassland Types Based on Object-Based Image Analysis with Multisource Data

The spatial distribution of different grassland types is important for effectively analyzing spatial patterns, obtaining key vegetation parameters using remote sensing (e.g., biomass, leaf area index, net primary production), and using and protecting grasslands. Existing classifications of grasslands by remote sensing are mostly divided according to the fractional vegetation cover or biomass, but classifications according to grassland types are scarce. In this study, we focused on the classification of different grassland types using remote sensing based on object-based image analysis (OBIA) with multitemporal images in combination with a 30-m digital elevation model (DEM) and the normalized difference vegetation index (NDVI). The grasslands were located in Hulunber, Inner Mongolia, and an autonomous region of China. The support vector machine (SVM) and random forest (RF) machine learning classifiers were selected for the classification. The results revealed the following: 1) It is feasible to generally extract different grassland types on the basis of OBIA with multisource data; the overall classification accuracy and Kappa value exceeded 90% and 0.9, respectively, using the SVM and RF machine learning classifiers, and the classification accuracy of the different grassland types ranged from 61.64% to 98.71%; 2) Multitemporal images and auxiliary data (DEM and NDVI) improved the separability of different grassland types. The information in the growing season was conducive for distinguishing temperate meadow steppe from temperate steppe and was favorable for extracting lowland meadow and swamp in the nongrowing season. The DEM and NDVI also effectively reduced the number of image segmentation objects and improved the segmentation effects; 3) Spectral and textural features were more important than geometric features in this study. A few main variables played a major role in the classification, while a large number of variables had either no significant effect or a negative effect on the classification results when the optimal feature subset was determined. This study provides a scientific basis and reference for the classification of various grassland types by remote sensing, including the data selection, image segmentation, feature selection, classifier selection, and parameter settings.