Binary Classification Performance Research Articles

Large language models for data extraction from unstructured and semi-structured electronic health records: a multiple model performance evaluation

ObjectivesWe aimed to evaluate the performance of multiple large language models (LLMs) in data extraction from unstructured and semi-structured electronic health records.Methods50 synthetic medical notes in English, containing a structured...

Densenet Model for Binary Glaucoma Classification Performance Assessment with Texture Feature

تعتبر شبكية العين جزءًا مهمًا من العين لأن الأطباء يستخدمون صورها لتشخيص العديد من أمراض العيون مثل الجلوكوما واعتلال الشبكية السكري وإعتام عدسة العين. في الواقع، يعد تصوير الشبكية المجزأ أداة قوية للكشف عن النمو غير العادي في منطقة العين بالإضافة إلى تحديد حجم وبنية القرص البصري. يمكن أن يؤدي الجلوكوما إلى إتلاف القرص البصري، مما يغير مظهر القرص البصري للعين. تعمل تقنيتنا على الكشف عن الجلوكوما وتصنيفها في هذه الدراسة. تستخدم المرحلة الأولى من هذا النظام الأنماط الثنائية المحلية (LBP) للحصول على خرائط للميزات التركيبية. يتم تنفيذ التصنيف الثنائي لصور الجلوكوما بشكل منفصل عن الصور العادية باستخدام تقنية التعلم العميق في الخطوة الثانية من هذا النظام. بالنسبة لتحديات التصنيف، استخدمنا تقنية الشبكة الكثيفة، وهي تقنية التعلم العميق. وتبقى دقتها في حدود 96% حسب نتائج الطريقة المقترحة.

Toward Foundation Models in Radiology? Quantitative Assessment of GPT-4V's Multimodal and Multianatomic Region Capabilities.

Background Large language models have already demonstrated potential in medical text processing. GPT-4V, a large vision-language model from OpenAI, has shown potential for medical imaging, yet a quantitative analysis is lacking. Purpose To quantitatively assess the performance of GPT-4V in interpreting radiologic images using unseen data. Materials and Methods This retrospective study included single representative abnormal and healthy control images from neuroradiology, cardiothoracic radiology, and musculoskeletal radiology (CT, MRI, radiography) to generate reports using GPT-4V via the application programming interface from February to March 2024. The factual correctness of free-text reports and the performance in detecting abnormalities in binary classification tasks were assessed using accuracy, sensitivity, and specificity. The binary classification performance was compared with that of a first-year nonradiologist in training and four board-certified radiologists. Results A total of 515 images in 470 patients (median age, 61 years [IQR, 44-71 years]; 267 male) were included, of which 345 images were abnormal. GPT-4V correctly identified the imaging modality and anatomic region in 100% (515 of 515) and 99.2% (511 of 515) of images, respectively. Diagnostic accuracy in free-text reports was between 0% (0 of 33 images) for pneumothorax (CT and radiography) and 90% (45 of 50 images) for brain tumor (MRI). In binary classification tasks, GPT-4V showed sensitivities between 56% (14 of 25 images) for ischemic stroke and 100% (25 of 25 images) for brain hemorrhage and specificities between 8% (two of 25 images) for brain hemorrhage and 52% (13 of 25 images) for pneumothorax, compared with a pooled sensitivity of 97.2% (1103 of 1135 images) and pooled specificity of 97.2% (1084 of 1115 images) for the human readers across all tasks. The model exhibited a clear tendency to overdiagnose abnormalities, with 86.5% (147 of 170 images) and 67.7% (151 of 223 images) false-positive rates for the free-text and binary classification tasks, respectively. Conclusion GPT-4V, in its earliest version, recognized medical image content and reliably determined the modality and anatomic region from single images. However, GPT-4V failed to detect, classify, or rule out abnormalities in image interpretation. © RSNA, 2024 Supplemental material is available for this article.

The Future of Surgical Diagnostics: AI-Enhanced Detection of Ganglion Cells for Hirschsprung Disease

Hirschsprung disease (HD), a congenital disease characterized by the absence of ganglion cells, presents significant surgical challenges. Addressing a critical gap in intraoperative diagnostics, we introduce transformative artificial-intelligence (AI) approach that significantly enhances the detection of ganglion cells in frozen-sections (FSs). The dataset comprises 366 frozen and 302 formalin-fixed-paraffin-embedded (FFPE) hematoxylin and eosin-stained slides obtained from 164 patients from three centers. Three pathologists annotated the ganglion cells on the whole-slide-images (WSIs) using bounding boxes. Tissue regions within WSIs were segmented and split into patches of 2000x2000 pixels. A deep-learning pipeline utilizing ResNet-50 model for feature extraction and Grad-CAM algorithm to generate heatmaps for ganglion cell localization was employed. The binary classification performance of the model was evaluated on independent test cohorts. In the multi-reader study, 10-pathologists assessed 50-frozen WSIs, with 25-slides containing ganglion cells, and 25-slides without. In the first phase of the study, pathologists evaluated the slides as in their routine practice. After two-week washout period, pathologists reevaluated the same WSIs along with the four patches with the highest probability of containing ganglion cells. The proposed deep-learning approach achieved an accuracy of 91.3%, 92.8%, 90.1% in detecting ganglion cells within WSIs in the test dataset obtained from centers. In the reader study, on average, the pathologists' diagnostic accuracy increased from 77% to 85.8% with the model’s heatmap support while the diagnosis time decreased from an average of 139.7 seconds to 70.5 seconds. Notably, when applied in real-world settings with group of pathologists, our model’s integration brought about substantial improvement in diagnosis precision and reduced the time required for diagnoses by half. This notable advance in AI-driven diagnostics not only sets new-standard for surgical decision-making in HD but also creates opportunities for its wider implementation in various clinical settings, highlighting its pivotal role in enhancing the efficacy and accuracy of FS analyses.

Enhancing Binary Classification Performance in Biomedical Datasets: Regularized ELM with SMOTE and Quantile Transforms Focused on Breast Cancer Analysis

Enhancing Binary Classification Performance in Biomedical Datasets: Regularized ELM with SMOTE and Quantile Transforms Focused on Breast Cancer Analysis

Spark: A Statistical Comparison and Evaluation of Classification Algorithms for Fault Prediction in Electrical Secondary Distribution Network

Managing faults in the electrical secondary distribution network is a challenging task given the nature, size, and complexity. Predicting faults early before they occur helps in increasing the safety and reliability of the power distribution system. Various statistical and machine learning techniques are being used to predict different types of faults. This study applies classification algorithms available in the big data framework Apache Spark through its python interface PySpark to predict electrical secondary distribution network faults. The study evaluates and compares nine algorithms: Decision tree, Gradient-boosted tree, Logistic regression, Naïve Bayes, Multilayer perceptron, Random forest, Linear Support Vector Machine, One-versus-rest and Factorization machines. The research uses Friedman’s test followed by the Nemenyi post hoc test to find the significance of performance differences among the algorithms. The results show significant differences among the algorithms. Gradient-boosted tree and One-versus-rest with Gradient-boosted tree had the best performance for binary and multiclass classification, respectively, while Naïve Bayes had the worst performance.

Targeting COVID-19 and Human Resources for Health News Information Extraction: Algorithm Development and Validation.

Global pandemics like COVID-19 put a high amount of strain on health care systems and health workers worldwide. These crises generate a vast amount of news information published online across the globe. This extensive corpus of articles has the potential to provide valuable insights into the nature of ongoing events and guide interventions and policies. However, the sheer volume of information is beyond the capacity of human experts to process and analyze effectively. The aim of this study was to explore how natural language processing (NLP) can be leveraged to build a system that allows for quick analysis of a high volume of news articles. Along with this, the objective was to create a workflow comprising human-computer symbiosis to derive valuable insights to support health workforce strategic policy dialogue, advocacy, and decision-making. We conducted a review of open-source news coverage from January 2020 to June 2022 on COVID-19 and its impacts on the health workforce from the World Health Organization (WHO) Epidemic Intelligence from Open Sources (EIOS) by synergizing NLP models, including classification and extractive summarization, and human-generated analyses. Our DeepCovid system was trained on 2.8 million news articles in English from more than 3000 internet sources across hundreds of jurisdictions. Rules-based classification with hand-designed rules narrowed the data set to 8508 articles with high relevancy confirmed in the human-led evaluation. DeepCovid's automated information targeting component reached a very strong binary classification performance of 98.98 for the area under the receiver operating characteristic curve (ROC-AUC) and 47.21 for the area under the precision recall curve (PR-AUC). Its information extraction component attained good performance in automatic extractive summarization with a mean Recall-Oriented Understudy for Gisting Evaluation (ROUGE) score of 47.76. DeepCovid's final summaries were used by human experts to write reports on the COVID-19 pandemic. It is feasible to synergize high-performing NLP models and human-generated analyses to benefit open-source health workforce intelligence. The DeepCovid approach can contribute to an agile and timely global view, providing complementary information to scientific literature.

Ensemble approach of deep learning models for binary and multiclass classification of histopathological images for breast cancer

Breast cancer (BC) is the most frequently occurring cancer disease observed in women after lung cancer. Out of different stages, invasive ductal BC causes maximum deaths in women. In this work, three deep learning (DL) models such as Vision Transformer (ViT), Convmixer, and Visual Geometry Group-19 (VGG-19) are implemented for the detection and classification of different breast cancer tumors with the help of Breast cancer histopathological (Break His) image database. The performance of each model is evaluated using an 80:20 training scheme and measured in terms of accuracy, precision, recall, loss, F1-score, and area under the curve (AUC). From the simulation result, ViT showed the best performance for binary classification of breast cancer tumors with accuracy, precision, recall, and F1-score of 99.89 %, 98.29 %, 98.29 %, and 98.29 %, respectively. Also, ViT showed the best performance in terms of accuracy (98.21 %), average Precision (89.84 %), recall (89.97 %), and F1-score (88.75) for eight class classifications. Moreover, we have also ensemble the ViT-Convmixer model and observed that the performance of the ensemble model is reduced as compared to the ViT model. We have also compared the performance of the proposed best model with other existing models reported by several research groups. The study will help find suitable models that will increase accuracy in early diagnoses of BC. We hope the study will also help to minimize human errors in the early diagnosis of this fatal disease and administer appropriate treatment. The proposed model may also be implemented for the detection of other diseases with improved accuracy.

Diagnosis and Screening of Velocardiofacial Syndrome by Evaluating Facial Photographs Using a Deep Learning-Based Algorithm.

Early detection of rare genetic diseases, including velocardiofacial syndrome (VCFS), is essential for patient well-being. However, their rarity and limited clinical experience of physicians make diagnosis challenging. Deep learning algorithms have emerged as promising tools for efficient and accurate diagnosis. This study investigates the use of a deep learning algorithm to develop a face recognition model for diagnosing VCFS. The study employed publicly available labeled face datasets to train the multitask cascaded convolutional neural networks (MTCNN) model. Subsequently, we examined the binary classification performance for diagnosing VCFS using the most efficient face recognition model. A total of 98 VCFS patients (920 facial photographs) and 91 non-VCFS controls (463 facial photographs) were randomly divided into training and test sets. Additionally, we analyzed whether the classification results matched the known facial phenotype of VCFS. The face recognition model demonstrated high accuracy, ranging from 94% to 99%, depending on the training dataset. The accuracy of the binary classification diagnostic model varied from 81% to 88% when evaluating with photographs taken at various angles, but reached 95% evaluating with frontal photographs only. Gradient-weighted class activation mapping heatmap revealed the high importance level of perinasal and periorbital areas, exhibiting consistency with the conventional facial phenotypes of VCFS. This study shows the feasibility and effectiveness of MTCNN-based model for detecting VCFS solely from facial photographs. The high accuracy underscores the potential of deep learning in aiding early diagnosis of rare genetic diseases, facilitating timely interventions for patient care.

Pulsar classification: comparing quantum convolutional neural networks and quantum support vector machines

Well-known quantum machine learning techniques, specifically quantum kernel-assisted support vector machines (QSVMs) and quantum convolutional neural networks (QCNNs), are applied to the binary classification of pulsars. In this comparative study, it is illustrated with simulations that both quantum methods successfully achieve effective classification of the HTRU-2 data set that connects pulsar class labels to eight separate features. While QCNNs are superior in terms of training and prediction speed compared to QSVMs, the preference shifts toward QSVMs when the present noisy NISQ-era devices are incorporated into the comparison. QSVMs demonstrate superior overall performance compared to QCNNs when assessed using binary classification performance metrics. Classical methods are implemented to serve as a benchmark for comparison with the quantum approaches.

Multi-modality radiomics of conventional T1 weighted and diffusion tensor imaging for differentiating Parkinson’s disease motor subtypes in early-stages

This study aimed to develop and validate a multi-modality radiomics approach using T1-weighted and diffusion tensor imaging (DTI) to differentiate Parkinson's disease (PD) motor subtypes, specifically tremor-dominant (TD) and postural instability gait difficulty (PIGD), in early disease stages. We analyzed T1-weighted and DTI scans from 140 early-stage PD patients (70 TD, 70 PIGD) and 70 healthy controls from the Parkinson's Progression Markers Initiative database. Radiomics features were extracted from 16 brain regions of interest. After harmonization and feature selection, four machine learning classifiers were trained and evaluated for both three-class (HC vs TD vs PIGD) and binary (TD vs PIGD) classification tasks. The light gradient boosting machine (LGBM) classifier demonstrated the best overall performance. For the three-class classification, LGBM achieved an accuracy of 85% and an area under the receiver operating characteristic curve (AUC) of 0.94 using combined T1 and DTI features. In the binary classification task, LGBM reached an accuracy of 95% and AUC of 0.95. Key discriminative features were identified in the Thalamus, Amygdala, Hippocampus, and Substantia Nigra for the three-group classification, and in the Pallidum, Amygdala, Hippocampus, and Accumbens for binary classification. The combined T1 + DTI approach consistently outperformed single-modality classifications, with DTI alone showing particularly low performance (AUC 0.55–0.62) in binary classification. The high accuracy and AUC values suggest that this approach could significantly improve early diagnosis and subtyping of PD. These findings have important implications for clinical management, potentially enabling more personalized treatment strategies based on early, accurate subtype identification.

Machine learning-driven feature importance appraisal of seismic parameters on tunnel damage and seismic fragility prediction

This study proposes a machine learning-driven approach for the analysis of the feature importance of seismic parameters on tunnel damage and seismic fragility prediction. The Incremental Dynamic Analysis (IDA) method serves as the fundamental database for vulnerability analysis. Strength and deformation yield criteria are chosen to comprehensively assess the impact of different seismic parameters on the vulnerability of tunnels to seismic events. Three machine learning algorithms, namely Extreme Gradient Boosting (XGBoost), Random Forest (RF), and Support Vector Machine (SVM), are utilized to develop models for classifying and regressing tunnel damage under seismic conditions. Following parameter tuning, the models' performance in multi-classification, binary classification, and regression prediction is assessed, with XGBoost and RF models exhibiting outstanding performance. Feature importance analysis of seismic parameters in XGBoost and RF models for multi-classification, binary classification, and regression is performed using Shapley additive explanations (SHAP). The correlation analysis between SHAP-based feature values and predictions reveals that Peak Ground Displacement (PGD) has the highest influence in the regression model. Utilizing the interaction dependencies among crucial features in the regression model, fragility curves for tunnels based on these key features are effectively derived. The predicted fragility curves closely align with those derived from IDA, illustrating the time-saving and high-performance capabilities of machine learning in nonlinear dynamic computations.

Modelling child comprehension: A case of suffixal passive construction in Korean

The present study investigates a computational model's ability to capture monolingual children's language behaviour during comprehension in Korean, an understudied language in the field. Specifically, we test whether and how two neural network architectures (LSTM, GPT-2) cope with a suffixal passive construction involving verbal morphology and required interpretive procedures (i.e., revising the mapping between thematic roles and case markers) driven by that morphology. To this end, we fine-tune our models via patching (i.e., pre-trained model + caregiver input) and hyperparameter adjustments, and measure their binary classification performance on the test sentences used in a behavioural study manifesting scrambling and omission of sentential components to varying degrees. We find that, while these models’ performance converges with the children's response patterns found in the behavioural study to some extent, the models do not faithfully simulate the children's comprehension behaviour pertaining to the suffixal passive, yielding by-model, by-condition, and by-hyperparameter asymmetries. This points to the limits of the neural networks’ capacity to address child language features. The implications of this study invite subsequent inquiries on the extent to which computational models reveal developmental trajectories of children's linguistic knowledge that have been unveiled through corpus-based or experimental research.

Depression recognition using voice-based pre-training model

The early screening of depression is highly beneficial for patients to obtain better diagnosis and treatment. While the effectiveness of utilizing voice data for depression detection has been demonstrated, the issue of insufficient dataset size remains unresolved. Therefore, we propose an artificial intelligence method to effectively identify depression. The wav2vec 2.0 voice-based pre-training model was used as a feature extractor to automatically extract high-quality voice features from raw audio. Additionally, a small fine-tuning network was used as a classification model to output depression classification results. Subsequently, the proposed model was fine-tuned on the DAIC-WOZ dataset and achieved excellent classification results. Notably, the model demonstrated outstanding performance in binary classification, attaining an accuracy of 0.9649 and an RMSE of 0.1875 on the test set. Similarly, impressive results were obtained in multi-classification, with an accuracy of 0.9481 and an RMSE of 0.3810. The wav2vec 2.0 model was first used for depression recognition and showed strong generalization ability. The method is simple, practical, and applicable, which can assist doctors in the early screening of depression.

A novel method for predicting ecological interactions with an unsupervised machine learning algorithm

Abstract This gap in knowledge regarding ecological interactions has prompted the development of various predictive approaches. Traditionally, ecological interactions have been inferred using traits. However, the lack of trait information for numerous organisms necessitates using phylogenetic data and statistical insights from interaction matrices for prediction. Previous studies have overlooked the validation of model‐predicted interactions. This study used a novel method in predicting ecological interactions using a self‐organizing map (SOM), an unsupervised machine learning algorithm. The SOM learns from the input interaction matrix by grouping the nodes into output layers based on their interactions. Subsequently, the trained model predicts the interactions as scores. To distinguish between interactions and non‐interactions, we employed F1 score maximization, setting scores above a specific threshold as interactions and the remainder as non‐interactions. We applied this method to three unipartite metawebs and one bipartite metaweb and subsequently validated the predicted interactions using two innovative approaches: taxonomic and interaction recovery validation. Our method exhibited outstanding predictive performance, particularly for large networks. Various binary classification performance indicators, including F1 score (0.84–0.97) and accuracy (0.97–0.99), indicated high performance. Moreover, the method generated minimal predicted interactions, signifying low noise in the predictions, particularly for large networks. Taxonomic validation excels in metawebs with a connectance >0.1 but performs poorly in metawebs with very low connectance. In contrast, interaction recovery was most effective in larger metawebs. Our proposed method excels at making highly accurate predictions of ecological interactions with minimal noise, solely utilizing input interaction data without relying on traits or phylogenetic information regarding interacting nodes. These predictions are particularly precise for large networks, underscoring their potential to address knowledge gaps in emerging extensive metawebs. Notably, taxonomic validation and interaction recovery methods are sensitive to connectance and network size, respectively, suggesting prospects for developing robust interaction validation methods.

Machine Learning-Based Fatigue Level Prediction for Exoskeleton-Assisted Trunk Flexion Tasks Using Wearable Sensors

Monitoring physical demands during task execution with exoskeletons can be instrumental in understanding their suitability for industrial tasks. This study aimed at developing a fatigue level prediction model for Back-Support Industrial Exoskeletons (BSIEs) using wearable sensors. Fourteen participants performed a set of intermittent trunk-flexion task cycles consisting of static, sustained, and dynamic activities, until they reached medium-high fatigue levels, while wearing BSIEs. Three classification algorithms, Support Vector Machine (SVM), Random Forest (RF), and XGBoost (XGB), were implemented to predict perceived fatigue level in the back and leg regions using features from four wearable wireless Electromyography (EMG) sensors with integrated Inertial Measurement Units (IMUs). We examined the best grouping and sensor combinations by comparing prediction performance. The findings showed best performance in binary classification of leg and back fatigue with 95% (2 EMG + IMU sensors) and 82% (single IMU sensor) accuracy, respectively. Tertiary classification for back and leg fatigue level prediction required four sensor setups with both EMG and IMU measures to perform at 79% and 67% accuracy, respectively. The efforts presented in our article demonstrate the feasibility of an accessible fatigue level detection system, which can be beneficial for objective fatigue assessment, design selection, and implementation of BSIEs in real-world scenarios.

Hospital Length of Stay Prediction for Planned Admissions Using Observational Medical Outcomes Partnership Common Data Model: Retrospective Study.

Accurate hospital length of stay (LoS) prediction enables efficient resource management. Conventional LoS prediction models with limited covariates and nonstandardized data have limited reproducibility when applied to the general population. In this study, we developed and validated a machine learning (ML)-based LoS prediction model for planned admissions using the Observational Medical Outcomes Partnership Common Data Model (OMOP CDM). Retrospective patient-level prediction models used electronic health record (EHR) data converted to the OMOP CDM (version 5.3) from Seoul National University Bundang Hospital (SNUBH) in South Korea. The study included 137,437 hospital admission episodes between January 2016 and December 2020. Covariates from the patient, condition occurrence, medication, observation, measurement, procedure, and visit occurrence tables were included in the analysis. To perform feature selection, we applied Lasso regularization in the logistic regression. The primary outcome was an LoS of 7 days or longer, while the secondary outcome was an LoS of 3 days or longer. The prediction models were developed using 6 ML algorithms, with the training and test set split in a 7:3 ratio. The performance of each model was evaluated based on the area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (AUPRC). Shapley Additive Explanations (SHAP) analysis measured feature importance, while calibration plots assessed the reliability of the prediction models. External validation of the developed models occurred at an independent institution, the Seoul National University Hospital. The final sample included 129,938 patient entry events in the planned admissions. The Extreme Gradient Boosting (XGB) model achieved the best performance in binary classification for predicting an LoS of 7 days or longer, with an AUROC of 0.891 (95% CI 0.887-0.894) and an AUPRC of 0.819 (95% CI 0.813-0.826) on the internal test set. The Light Gradient Boosting (LGB) model performed the best in the multiclassification for predicting an LoS of 3 days or more, with an AUROC of 0.901 (95% CI 0.898-0.904) and an AUPRC of 0.770 (95% CI 0.762-0.779). The most important features contributing to the models were the operation performed, frequency of previous outpatient visits, patient admission department, age, and day of admission. The RF model showed robust performance in the external validation set, achieving an AUROC of 0.804 (95% CI 0.802-0.807). The use of the OMOP CDM in predicting hospital LoS for planned admissions demonstrates promising predictive capabilities for stays of varying durations. It underscores the advantage of standardized data in achieving reproducible results. This approach should serve as a model for enhancing operational efficiency and patient care coordination across health care settings.

Linearithmic and unbiased implementation of DeLong’s algorithm for comparing the areas under correlated ROC curves

Receiver operating characteristic (ROC) analysis is a widespread framework in two-group classification problems. The area under the ROC curve (AUC) has been a popular figure of merit to summarize the performance of binary classifiers. Because of its simplicity of implementation, DeLong’s algorithm (DLA) has been widely used in practice. However, DLA suffers two drawbacks of time inefficiency and biasedness. The authors’ previous work has overcome the first drawback. Specifically, the time complexity of DLA is reduced from the quadratic order down to a linearithmic one, by exploiting the relationship between the Heaviside function and the mid-ranks of samples. Unfortunately, the second drawback of DLA remains unaddressed, to the best of our knowledge. Motivated by such unsatisfactory situation, in this paper, the authors correct the bias of DLA with the assistance of an efficient implementation of Kendall’s tau (KT). The overall computational load of the improved DLA (being strictly unbiased) is in linearithmic time. Theoretical analyses and numerical results verified the proposed algorithmic findings in this work.

Detecting Fake Online Reviews: An Unsupervised Detection Method With a Novel Performance Evaluation

ABSTRACT Fake reviews are critical issues in the online world, as they affect the credibility of e-commerce platforms and undermine consumers’ trust. Therefore, fake-review detection is of great significance. Since fake-review detection is an unsupervised problem, most existing methods and performance metrics cannot be applied. In addition, contemporary review manipulations are much more difficult to detect than before. To address these two problems, we first propose a fully unsupervised method with steps of survey research, fake-review feature analysis, fake index estimation, and fake-review selection. Fake-review features can be accurately derived from existing studies and survey research. Second, we propose a recommendation-based performance metric for evaluating fake-review detection methods. This metric differs from traditional binary classification performance metrics, as it can be used on review data with no objective review authenticity classifications. In this research, we utilize Dianping as a case study to evaluate the effectiveness of the proposed detection method and performance metric.

Multicategory Survival Outcomes Classification via Overlapping Group Screening Process Based on Multinomial Logistic Regression Model With Application to TCGA Transcriptomic Data.

Under the classification of multicategory survival outcomes of cancer patients, it is crucial to identify biomarkers that affect specific outcome categories. The classification of multicategory survival outcomes from transcriptomic data has been thoroughly investigated in computational biology. Nevertheless, several challenges must be addressed, including the ultra-high-dimensional feature space, feature contamination, and data imbalance, all of which contribute to the instability of the diagnostic model. Furthermore, although most methods achieve accurate predicted performance for binary classification with high-dimensional transcriptomic data, their extension to multi-class classification is not straightforward. We employ the One-versus-One strategy to transform multi-class classification into multiple binary classification, and utilize the overlapping group screening procedure with binary logistic regression to include pathway information for identifying important genes and gene-gene interactions for multicategory survival outcomes. A series of simulation studies are conducted to compare the classification accuracy of our proposed approach with some existing machine learning methods. In practical data applications, we utilize the random oversampling procedure to tackle class imbalance issues. We then apply the proposed method to analyze transcriptomic data from various cancers in The Cancer Genome Atlas, such as kidney renal papillary cell carcinoma, lung adenocarcinoma, and head and neck squamous cell carcinoma. Our aim is to establish an accurate microarray-based multicategory cancer diagnosis model. The numerical results illustrate that the new proposal effectively enhances cancer diagnosis compared to approaches that neglect pathway information. We showcase the effectiveness of the proposed method in terms of class prediction accuracy through evaluations on simulated synthetic datasets as well as real dataset applications. We also identified the cancer-related gene-gene interaction biomarkers and reported the corresponding network structure. According to the identified major genes and gene-gene interactions, we can predict for each patient the probabilities that he/she belongs to each of the survival outcome classes.