Feature selection method based on support vector machine and shape analysis for high-throughput medical data

MRI-Based Radiomics and Delta-Radiomics Models of the Patella Predict the Radiographic Progression of Osteoarthritis: Data From the FNIH OA Biomarkers Consortium

RATIONALE AND OBJECTIVES

To analyse the MRI-based radiomics and delta-radiomics features to establish radiomics models for predicting the radiographic progression of osteoarthritis (OA).

MATERIALS AND METHODS

The data used in this research come from the dataset of the FNIH Biomarker Consortium Project within the Osteoarthritis Initiative (OAI). 565 participants randomly divided into training and validation groups at a 7:3 ratio. The training cohort consisted of 395 participants and included 202 cases. The validation cohort consisted of 170 participants and included 87 cases. Least absolute shrinkage and selection operator (LASSO) was used for feature selection. Support vector machine (SVM) was used to establish radiomics models and clinical and biomarker models for predicting the radiographic progression of OA. The predictive ability of the model was evaluated by the area under the curve (AUC).

RESULTS

The baseline, 24 M, Delta, and two combination radiomics models (Baseline and Delta, 24 M and Delta) all showed good predictive performance in the training and validation cohorts, with the combination model exhibiting the best performance. In the training cohort, the AUCs were 0.851 (95% CI: 0.812-0.890), 0.825 (95% CI: 0.784-0.865), 0.804 (95% CI: 0.761-0.847), 0.892 (95% CI: 0.860-0.924) and 0.884 (95% CI: 0.851-0.917), respectively. The AUCs in the validation cohort were 0.741 (95% CI: 0.667-0.814), 0.786 (95% CI: 0.716-0.856), 0.745 (95% CI: 0.671-0.819), 0.781 (95% CI: 0.711-0.851) and 0.802 (95% CI: 0.736-0.869), respectively. As compared, the clinical and biomarker models have AUC < 0.74. The DeLong test showed that the predictive performance of the radiomics models in the training and validation cohorts was significantly better than that of the clinical and biomarker models (P < 0.001).

CONCLUSION

The MRI-based radiomics models of the patella all showed good predictive performance performed better than the clinical and biomarker models in predicting the radiographic progression of OA. Delta radiomics can improve the predictive performance of the single time model, the combined model of 24 M and Delta provided the best predictive performance.

Corneal biomechanics in early diagnosis of keratoconus using artificial intelligence

Keratoconus is a blinding eye disease that affects activities of daily living; therefore, early diagnosis is crucial. Great efforts have been made toward an early diagnosis of keratoconus. Recent studies have shown that corneal biomechanics is associated with the occurrence and progression of keratoconus. Hence, detecting changes in corneal biomechanics may provide a novel strategy for early diagnosis. However, an early keratoconus diagnosis remains challenging due to the subtle and localized nature of its lesions. Artificial intelligence has been used to help address this problem. Herein, we reviewed the literature regarding three aspects of keratoconus (keratoconus, early keratoconus, and keratoconus grading) based on corneal biomechanical properties using artificial intelligence. Furthermore, we summarized the current research progress, limitations, and possible prospects.

Union With Recursive Feature Elimination: A Feature Selection Framework to Improve the Classification Performance of Multicategory Causes of Death in Colorectal Cancer

Despite the use of machine learning tools, it is challenging to properly model cause-specific deaths in colorectal cancer (CRC) patients and choose appropriate treatments. Here, we propose an interesting feature selection framework, namely union with recursive feature elimination (U-RFE), to select the union feature sets that are crucial in CRC progression-specific mortality using The Cancer Genome Atlas (TCGA) dataset. Based on the union feature sets, we compared the performance of 5 classification algorithms, including logistic regression (LR), support vector machines (SVM), random forest (RF), eXtreme gradient boosting (XGBoost), and Stacking, to identify the best model for classifying 4-category deaths. In the first stage of U-RFE, LR, SVM, and RF were used as base estimators to obtain subsets containing the same number of features but not exactly the same specific features. Union analysis of the subsets was then performed to determine the final union feature set, effectively combining the advantages of different algorithms. We found that the U-RFE framework could improve various models' performance. Stacking outperformed LR, SVM, RF, and XGBoost in most scenarios. When the target feature number of the RFE was set to 50 and the union feature set contained 298 deterministic features, the Stacking model achieved F1_weighted, Recall_weighted, Precision_weighted, Accuracy, and Matthews correlation coefficient of 0.851, 0.864, 0.854, 0.864, and 0.717, respectively. The performance of the minority categories was also significantly improved. Therefore, this recursive feature elimination-based approach of feature selection improves performances of classifying CRC deaths using clinical and omics data or those using other data with high feature redundancy and imbalance.

Identifying potential biomarkers for non-obstructive azoospermia using WGCNA and machine learning algorithms

Objective

The cause and mechanism of non-obstructive azoospermia (NOA) is complicated; therefore, an effective therapy strategy is yet to be developed. This study aimed to analyse the pathogenesis of NOA at the molecular biological level and to identify the core regulatory genes, which could be utilised as potential biomarkers.

Methods

Three NOA microarray datasets (GSE45885, GSE108886, and GSE145467) were collected from the GEO database and merged into training sets; a further dataset (GSE45887) was then defined as the validation set. Differential gene analysis, consensus cluster analysis, and WGCNA were used to identify preliminary signature genes; then, enrichment analysis was applied to these previously screened signature genes. Next, 4 machine learning algorithms (RF, SVM, GLM, and XGB) were used to detect potential biomarkers that are most closely associated with NOA. Finally, a diagnostic model was constructed from these potential biomarkers and visualised as a nomogram. The differential expression and predictive reliability of the biomarkers were confirmed using the validation set. Furthermore, the competing endogenous RNA network was constructed to identify the regulatory mechanisms of potential biomarkers; further, the CIBERSORT algorithm was used to calculate immune infiltration status among the samples.

Results

A total of 215 differentially expressed genes (DEGs) were identified between NOA and control groups (27 upregulated and 188 downregulated genes). The WGCNA results identified 1123 genes in the MEblue module as target genes that are highly correlated with NOA positivity. The NOA samples were divided into 2 clusters using consensus clustering; further, 1027 genes in the MEblue module, which were screened by WGCNA, were considered to be target genes that are highly correlated with NOA classification. The 129 overlapping genes were then established as signature genes. The XGB algorithm that had the maximum AUC value (AUC=0.946) and the minimum residual value was used to further screen the signature genes. IL20RB, C9orf117, HILS1, PAOX, and DZIP1 were identified as potential NOA biomarkers. This 5 biomarker model had the highest AUC value, of up to 0.982, compared to other single biomarker models; additionally, the results of this biomarker model were verified in the validation set.

Conclusions

As IL20RB, C9orf117, HILS1, PAOX, and DZIP1 have been determined to possess the strongest association with NOA, these five genes could be used as potential therapeutic targets for NOA patients. Furthermore, the model constructed using these five genes, which possessed the highest diagnostic accuracy, may be an effective biomarker model that warrants further experimental validation.

Artificial intelligence in cancer diagnosis and therapy: Current status and future perspective

Artificial intelligence (AI) in healthcare plays a pivotal role in combating many fatal diseases, such as skin, breast, and lung cancer. AI is an advanced form of technology that uses mathematical-based algorithmic principles similar to those of the human mind for cognizing complex challenges of the healthcare unit. Cancer is a lethal disease with many etiologies, including numerous genetic and epigenetic mutations. Cancer being a multifactorial disease is difficult to be diagnosed at an early stage. Therefore, genetic variations and other leading factors could be identified in due time through AI and machine learning (ML). AI is the synergetic approach for mining the drug targets, their mechanism of action, and drug-organism interaction from massive raw data. This synergetic approach is also facing several challenges in data mining but computational algorithms from different scientific communities for multi-target drug discovery are highly helpful to overcome the bottlenecks in AI for drug-target discovery. AI and ML could be the epicenter in the medical world for the diagnosis, treatment, and evaluation of almost any disease in the near future. In this comprehensive review, we explore the immense potential of AI and ML when integrated with the biological sciences, specifically in the context of cancer research. Our goal is to illuminate the many ways in which AI and ML are being applied to the study of cancer, from diagnosis to individualized treatment. We highlight the prospective role of AI in supporting oncologists and other medical professionals in making informed decisions and improving patient outcomes by examining the intersection of AI and cancer control. Although AI-based medical therapies show great potential, many challenges must be overcome before they can be implemented in clinical practice. We critically assess the current hurdles and provide insights into the future directions of AI-driven approaches, aiming to pave the way for enhanced cancer interventions and improved patient care.

Exploring patient medication adherence and data mining methods in clinical big data: A contemporary review

BACKGROUND

Increasingly, patient medication adherence data are being consolidated from claims databases and electronic health records (EHRs). Such databases offer an indirect avenue to gauge medication adherence in our data-rich healthcare milieu. The surge in data accessibility, coupled with the pressing need for its conversion to actionable insights, has spotlighted data mining, with machine learning (ML) emerging as a pivotal technique. Nonadherence poses heightened health risks and escalates medical costs. This paper elucidates the synergistic interaction between medical database mining for medication adherence and the role of ML in fostering knowledge discovery.

METHODS

We conducted a comprehensive review of EHR applications in the realm of medication adherence, leveraging ML techniques. We expounded on the evolution and structure of medical databases pertinent to medication adherence and harnessed both supervised and unsupervised ML paradigms to delve into adherence and its ramifications.

RESULTS

Our study underscores the applications of medical databases and ML, encompassing both supervised and unsupervised learning, for medication adherence in clinical big data. Databases like SEER and NHANES, often underutilized due to their intricacies, have gained prominence. Employing ML to excavate patient medication logs from these databases facilitates adherence analysis. Such findings are pivotal for clinical decision-making, risk stratification, and scholarly pursuits, aiming to elevate healthcare quality.

CONCLUSION

Advanced data mining in the era of big data has revolutionized medication adherence research, thereby enhancing patient care. Emphasizing bespoke interventions and research could herald transformative shifts in therapeutic modalities.

Artificial Intelligence-Based Diagnostic Model for Detecting Keratoconus Using Videos of Corneal Force Deformation

Purpose

To develop a novel method based on biomechanical parameters calculated from raw corneal dynamic deformation videos to quickly and accurately diagnose keratoconus using machine learning.

Methods

The keratoconus group was included according to Rabinowitz's criteria, and the normal group included corneal refractive surgery candidates. Independent biomechanical parameters were calculated from dynamic corneal deformation videos. A novel neural network model was trained to diagnose keratoconus. Tenfold cross-validation was performed, and the sample set was divided into a training set for training, a validation set for parameter validation, and a testing set for performance evaluation. External validation was performed to evaluate the model's generalizability.

Results

A novel intelligent diagnostic model for keratoconus based on a five-layer feedforward network was constructed by calculating four biomechanical characteristics, including time of the first applanation, deformation amplitude at the highest concavity, central corneal thickness, and radius at the highest concavity. The model was able to diagnose keratoconus with 99.6% accuracy, 99.3% sensitivity, 100% specificity, and 100% precision in the sample set (n = 276), and it achieved an accuracy of 98.7%, sensitivity of 97.4%, specificity of 100%, and precision of 100% in the external validation set (n = 78).

Conclusions

In the absence of corneal topographic examination, rapid and accurate diagnosis of keratoconus is possible with the aid of machine learning. Our study provides a new potential approach and sheds light on the diagnosis of keratoconus from a purely corneal biomechanical perspective.

Translational Relevance

Our findings could help improve the diagnosis of keratoconus based on corneal biomechanical properties.

A multicenter prospective, randomized, placebo-controlled phase II/III trial for preemptive acute graft-versus-host disease therapy

Acute graft-versus-host disease (aGvHD) contributes to about 50% of transplant-related mortality (non-relapse mortality) after allogeneic hematopoietic stem cell transplantation (HSCT). Here the predictive value of a urinary proteomic profile (aGvHD_MS17) was tested together with preemptive prednisolone therapy. Two-hundred and fifty-nine of 267 patients were eligible for analysis. Ninety-two patients were randomized upon aGvHD_MS17 classification factor above 0.1 to receive either prednisolone (2–2.5 mg/kg, N = 44) or placebo (N = 47; N = 1 randomization failure) for 5 days followed by tapering. The remaining 167 patients formed the observation group. The primary endpoint of the randomized trial was incidence of aGvHD grade II between randomization and day +100 post HSCT. Analysis of the short-term preemptive prednisolone therapy in the randomized patients showed no significant difference in incidence or severity of acute GvHD (HR: 1.69, 95% CI: 0.66–4.32, P = 0.27). Prednisolone as preemptive treatment did not lead to an increase in relapse (20.2% in the placebo and 14.0% in the prednisolone group (P = 0.46)). The frequency of adverse events was slightly higher in the placebo group (64.4% versus 50%, respectively). Taken together, the results of the Pre-GvHD trial demonstrated the feasibility and safety of preemptive prednisolone treatment in the randomized patients.

Improving the Diagnosis of Phenylketonuria by Using a Machine Learning-Based Screening Model of Neonatal MRM Data

Phenylketonuria (PKU) is a common genetic metabolic disorder that affects the infant's nerve development and manifests as abnormal behavior and developmental delay as the child grows. Currently, a triple–quadrupole mass spectrometer (TQ-MS) is a common high-accuracy clinical PKU screening method. However, there is high false-positive rate associated with this modality, and its reduction can provide a diagnostic and economic benefit to both pediatric patients and health providers. Machine learning methods have the advantage of utilizing high-dimensional and complex features, which can be obtained from the patient's metabolic patterns and interrogated for clinically relevant knowledge. In this study, using TQ-MS screening data of more than 600,000 patients collected at the Newborn Screening Center of Shanghai Children's Hospital, we derived a dataset containing 256 PKU-suspected cases. We then developed a machine learning logistic regression analysis model with the aim to minimize false-positive rates in the results of the initial PKU test. The model attained a 95–100% sensitivity, the specificity was improved 53.14%, and positive predictive value increased from 19.14 to 32.16%. Our study shows that machine learning models may be used as a pediatric diagnosis aid tool to reduce the number of suspected cases and to help eliminate patient recall. Our study can serve as a future reference for the selection and evaluation of computational screening methods.

Clinical risk assessment in early pregnancy for preeclampsia in nulliparous women: A population based cohort study

Objective

To evaluate the capacity of multivariable prediction of preeclampsia during pregnancy, based on detailed routinely collected early pregnancy data in nulliparous women.

Design and setting

A population-based cohort study of 62 562 pregnancies of nulliparous women with deliveries 2008–13 in the Stockholm-Gotland Counties in Sweden.

Methods

Maternal social, reproductive and medical history and medical examinations (including mean arterial pressure, proteinuria, hemoglobin and capillary glucose levels) routinely collected at the first visit in antenatal care, constitute the predictive variables. Predictive models for preeclampsia were created by three methods; logistic regression models using 1) pre-specified variables (similar to the Fetal Medicine Foundation model including maternal factors and mean arterial pressure), 2) backward selection starting from the full suite of variables, and 3) a Random forest model using the same candidate variables. The performance of the British National Institute for Health and Care Excellence (NICE) binary risk classification guidelines for preeclampsia was also evaluated. The outcome measures were diagnosis of preeclampsia with delivery <34, <37, and ≥37 weeks’ gestation.

Results

A total of 2 773 (4.4%) nulliparous women subsequently developed preeclampsia. The pre-specified variables model was superior the other two models, regarding prediction of preeclampsia with delivery <34 and <37 weeks, both with areas under the curve of 0.68, and sensitivity of 30.6% (95% CI 24.5–37.2) and 29.2% (95% CI 25.2–33.4) at a 10% false positive rate, respectively. The performance of these customizable multivariable models at the chosen false positive rate, was significantly better than the binary NICE-guidelines for preeclampsia with delivery <37 and ≥37 weeks’ gestation.

Conclusion

Multivariable models in early pregnancy had a modest performance, although providing advantages over the NICE-guidelines, in predicting preeclampsia in nulliparous women. Use of a machine learning algorithm (Random forest) did not result in superior prediction.

Sparse support vector machines with L0 approximation for ultra-high dimensional omics data

Omics data usually have ultra-high dimension (p) and small sample size (n). Standard support vector machines (SVMs), which minimize the L₂ norm for the primal variables, only lead to sparse solutions for the dual variables. L₁ based SVMs, directly minimizing the L₁ norm, have been used for feature selection with omics data. However, most current methods directly solve the primal formulations of the problem, which are not computationally scalable. The computational complexity increases with the number of features. In addition, L₁ norm is known to be asymptotically biased and not consistent for feature selection. In this paper, we develop an efficient method for sparse support vector machines with L₀ norm approximation. The proposed method approximates the L₀ minimization through solving a series of L₂ optimization problems, which can be formulated with dual variables. It finds the optimal solution for p primal variables through estimating n dual variables, which is more efficient as long as the sample size is small. L₀ approximation leads to sparsity in both dual and primal variables, and can be used for both feature and sample selections. The proposed method identifies much less number of features and achieves similar performances in simulations. We apply the proposed method to feature selections with metagenomic sequencing and gene expression data. It can identify biologically important genes and taxa efficiently.