Analysis of Cross-Combinations of Feature Selection and Machine-Learning Classification Methods Based on [18F]F-FDG PET/CT Radiomic Features for Metabolic Response Prediction of Metastatic Breast Cancer Lesions

Machine Learning Model Development for Malignant Prostate Lesion Prediction Using Texture Analysis Features from Ultrasound Shear-Wave Elastography

Introduction: Artificial intelligence (AI) is increasingly utilized for texture analysis and the development of machine learning (ML) techniques to enhance diagnostic accuracy. ML algorithms are trained to differentiate between normal and malignant conditions based on provided data. Texture feature analysis, including first-order and second-order features, is a critical step in ML development. This study aimed to evaluate quantitative texture features of normal and prostate cancer tissues identified through ultrasound B-mode and shear-wave elastography (SWE) imaging and to develop and assess ML models for predicting and classifying normal versus malignant prostate tissues. Methodology: First-order and second-order texture features were extracted from B-mode and SWE imaging, including four reconstructed regions of interest (ROIs) from SWE images for normal and malignant tissues. A total of 94 texture features were derived, including features for intensity, Gray-Level Co-Occurrence Matrix (GLCM), Gray-Level Dependence Length Matrix (GLDLM), Gray-Level Run Length Matrix (GLRLM), and Gray-Level Size Zone Matrix (GLSZM). Five ML models were developed and evaluated using 5-fold cross-validation to predict normal and malignant tissues. Results: Data from 62 patients were analyzed. All ROIs, except those derived from B-mode imaging, exhibited statistically significant differences in features between normal and malignant tissues. Among the developed models, Support Vector Machines (SVM), Random Forest (RF), and Naive Bayes (NB) demonstrated the highest performance across all ROIs. These models consistently achieved strong predictive accuracy for classifying normal versus malignant tissues. Gray Pure SWE and Gray Reconstructed images Provided the highest sensitivity and specificity in PCa prediction by 82%, 90%, and 98%, 96%, respectively. Conclusions: Texture analysis with machine learning on SWE-US and reconstructed images effectively differentiates malignant from benign prostate lesions, with features like contrast, entropy, and correlation playing a key role. Random Forest, SVM, and Naïve Bayes showed the highest classification performance, while grayscale reconstructions (GPSWE and GRRI) enhanced detection accuracy.

Tlalpan 2020 Case Study: Enhancing Uric Acid Level Prediction with Machine Learning Regression and Cross-Feature Selection

Background/Objectives: Uric acid is a key metabolic byproduct of purine degradation and plays a dual role in human health. At physiological levels, it acts as an antioxidant, protecting against oxidative stress. However, excessive uric acid can lead to hyperuricemia, contributing to conditions like gout, kidney stones, and cardiovascular diseases. Emerging evidence also links elevated uric acid levels with metabolic disorders, including hypertension and insulin resistance. Understanding its regulation is crucial for preventing associated health complications. Methods: This study, part of the Tlalpan 2020 project, aimed to predict uric acid levels using advanced machine learning algorithms. The dataset included clinical, anthropometric, lifestyle, and nutritional characteristics from a cohort in Mexico City. We applied Boosted Decision Trees (Boosted DTR), eXtreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), and Shapley Additive Explanations (SHAP) to identify the most relevant variables associated with hyperuricemia. Feature engineering techniques improved model performance, evaluated using Mean Squared Error (MSE), Root-Mean-Square Error (RMSE), and the coefficient of determination (R2). Results: Our study showed that XGBoost had the highest accuracy for anthropometric and clinical predictors, while CatBoost was the most effective at identifying nutritional risk factors. Distinct predictive profiles were observed between men and women. In men, uric acid levels were primarily influenced by renal function markers, lipid profiles, and hereditary predisposition to hyperuricemia, particularly paternal gout and diabetes. Diets rich in processed meats, high-fructose foods, and sugary drinks showed stronger associations with elevated uric acid levels. In women, metabolic and cardiovascular markers, family history of metabolic disorders, and lifestyle factors such as passive smoking and sleep quality were the main contributors. Additionally, while carbohydrate intake was more strongly associated with uric acid levels in women, fructose and sugary beverages had a greater impact in men. To enhance model robustness, a cross-feature selection approach was applied, integrating top features from multiple models, which further improved predictive accuracy, particularly in gender-specific analyses. Conclusions: These findings provide insights into the metabolic, nutritional characteristics, and lifestyle determinants of uric acid levels, supporting targeted public health strategies for hyperuricemia prevention.

Prediction of Chemotherapy Response in Locally Advanced Breast Cancer Patients at Pre-Treatment Using CT Textural Features and Machine Learning: Comparison of Feature Selection Methods

RATIONALE

Neoadjuvant chemotherapy (NAC) is a key element of treatment for locally advanced breast cancer (LABC). Predicting the response of NAC for patients with LABC before initiating treatment would be valuable to customize therapies and ensure the delivery of effective care.

OBJECTIVE

Our objective was to develop predictive measures of tumor response to NAC prior to starting for LABC using machine learning and textural computed tomography (CT) features in different level of frequencies.

MATERIALS AND METHODS

A total of 851 textural biomarkers were determined from CT images and their wavelet coefficients for 117 patients with LABC to evaluate the response to NAC. A machine learning pipeline was designed to classify response to NAC treatment for patients with LABC. For training predictive models, three models including all features (wavelet and original image features), only wavelet and only original-image features were considered. We determined features from CT images in different level of frequencies using wavelet transform. Additionally, we conducted a comparison of feature selection methods including mRMR, Relief, Rref QR decomposition, nonnegative matrix factorization and perturbation theory feature selection techniques.

RESULTS

Of the 117 patients with LABC evaluated, 82 (70%) had clinical-pathological response to chemotherapy and 35 (30%) had no response to chemotherapy. The best performance for hold-out data splitting was obtained using the KNN classifier using the Top-5 features, which were obtained by mRMR, for all features (accuracy = 77%, specificity = 80%, sensitivity = 56%, and balanced-accuracy = 68%). Likewise, the best performance for leave-one-out data splitting could be obtained by the KNN classifier using the Top-5 features, which was obtained by mRMR, for all features (accuracy = 75%, specificity = 76%, sensitivity = 62%, and balanced-accuracy = 72%).

CONCLUSIONS

The combination of original textural features and wavelet features results in a greater predictive accuracy of NAC response for LABC patients. This predictive model can be utilized to predict treatment outcomes prior to starting, and clinicians can use it as a recommender system to modify treatment.

Robust vs. Non-robust radiomic features: the quest for optimal machine learning models using phantom and clinical studies

PURPOSE

This study aimed to select robust features against lung motion in a phantom study and use them as input to feature selection algorithms and machine learning classifiers in a clinical study to predict the lymphovascular invasion (LVI) of non-small cell lung cancer (NSCLC). The results of robust features were also compared with conventional techniques without considering the robustness of radiomic features.

METHODS

An in-house developed lung phantom was developed with two 22mm lesion sizes based on a clinical study. A specific motor was built to simulate motion in two orthogonal directions. Lesions of both clinical and phantom studies were segmented using a Fuzzy C-means-based segmentation algorithm. After inducing motion and extracting 105 radiomic features in 4 feature sets, including shape, first-, second-, and higher-order statistics features from each region of interest (ROI) of the phantom image, statistical analyses were performed to select robust features against motion. Subsequently, these robust features and a total of 105 radiomic features were extracted from 126 clinical data. Various feature selection (FS) and multiple machine learning (ML) classifiers were implemented to predict the LVI of NSCLC, followed by comparing the results of predicting LVI using robust features with common conventional techniques not considering the robustness of radiomic features.

RESULTS

Our results demonstrated that selecting robust features as input to FS algorithms and ML classifiers surges the sensitivity, which has a gentle negative effect on the accuracy and the area under the curve (AUC) of predictions compared with commonly used methods in 12 of 15 outcomes. The top performance of the LVI prediction was achieved by the NB classifier and RFE FS without considering the robustness of radiomic features with 95% area under the curve of AUC, 67% accuracy, and 100% sensitivity. Moreover, the top performance of the LVI prediction using robust features belonged to the NB classifier and Boruta feature selection with 92% AUC, 86% accuracy, and 100% sensitivity.

CONCLUSION

Robustness over various influential factors is critical and should be considered in a radiomic study. Selecting robust features is a solution to overcome the low reproducibility of radiomic features. Although setting robust features against motion in a phantom study has a minor negative impact on the accuracy and AUC of LVI prediction, it boosts the sensitivity of prediction to a large extent.

Cross-Combination Analyses of Random Forest Feature Selection and Decision Tree Model for Predicting Intraoperative Hypothermia in Total Joint Arthroplasty

BACKGROUND

In total joint arthroplasty patients, intraoperative hypothermia (IOH) is associated with perioperative complications and an increased economic burden. Previous models have some limitations and mainly focus on regression modeling. Random forest (RF) algorithms and decision tree modeling are effective for eliminating irrelevant features and making predictions that aid in accelerating modeling and reducing application difficulty.

METHODS

We conducted this prospective observational study using convenience sampling and collected data from 327 total joint arthroplasty patients in a tertiary hospital from March 4, 2023, to September 11, 2023. Of those, 229 patients were assigned to the training and 98 to the testing sets. The Chi-square, Mann-Whitney U, and t-tests were used for baseline analyses. The feature variables selection used the RF algorithms, and the decision tree model was trained on 299 examples and validated on 98. The sensitivity, specificity, recall, F1 score, and area under the curve were used to test the model's performance.

RESULTS

The RF algorithms identified the preheating time, the volume of flushing fluids, the intraoperative infusion volume, the anesthesia time, the surgical time, and the core temperature after intubation as risk factors for IOH. The decision tree was grown to 5 levels with 9 terminal nodes. The overall incidence of IOH was 42.13%. The sensitivity, specificity, recall, F1 score, and area under the curve were 0.651, 0.907, 0.916, 0.761, and 0.810, respectively. The model indicated strong internal consistency and predictive ability.

CONCLUSIONS

The preheating time, the volume of flushing fluids, the intraoperative infusion volume, the anesthesia time, the surgical time, and the core temperature after intubation could accurately predict IOH in total joint arthroplasty patients. By monitoring these factors, the clinical staff could achieve early detection and intervention of IOH in total joint arthroplasty patients.

AI tool for predicting MGMT methylation in glioblastoma for clinical decision support in resource limited settings

Glioblastoma is an aggressive brain cancer with a poor prognosis. The O6-methylguanine-DNA methyltransferase (MGMT) gene methylation status is crucial for treatment stratification, yet economic constraints often limit access. This study aims to develop an artificial intelligence (AI) framework for predicting MGMT methylation. Diagnostic magnetic resonance (MR) images in public repositories were used for training. The algorithm created was validated in data from a single institution. All images were segmented according to widely used guidelines for radiotherapy planning and combined with clinical evaluations from neuroradiology experts. Radiomic features and clinical impressions were extracted, tabulated, and used for modeling. Feature selection methods were used to identify relevant phenotypes. A total of 100 patients were used for training and 46 for validation. A total of 343 features were extracted. Eight feature selection methods produced seven independent predictive frameworks. The top-performing ML model was a model post-Least Absolute Shrinkage and Selection Operator (LASSO) feature selection reaching accuracy (ACC) of 0.82, an area under the curve (AUC) of 0.81, a recall of 0.75, and a precision of 0.75. This study demonstrates that integrating clinical and radiotherapy-derived AI-driven phenotypes can predict MGMT methylation. The framework addresses constraints that limit molecular diagnosis access.

Optimizing radiomics for prostate cancer diagnosis: feature selection strategies, machine learning classifiers, and MRI sequences

OBJECTIVES

Radiomics-based analyses encompass multiple steps, leading to ambiguity regarding the optimal approaches for enhancing model performance. This study compares the effect of several feature selection methods, machine learning (ML) classifiers, and sources of radiomic features, on models' performance for the diagnosis of clinically significant prostate cancer (csPCa) from bi-parametric MRI.

METHODS

Two multi-centric datasets, with 465 and 204 patients each, were used to extract 1246 radiomic features per patient and MRI sequence. Ten feature selection methods, such as Boruta, mRMRe, ReliefF, recursive feature elimination (RFE), random forest (RF) variable importance, L1-lasso, etc., four ML classifiers, namely SVM, RF, LASSO, and boosted generalized linear model (GLM), and three sets of radiomics features, derived from T2w images, ADC maps, and their combination, were used to develop predictive models of csPCa. Their performance was evaluated in a nested cross-validation and externally, using seven performance metrics.

RESULTS

In total, 480 models were developed. In nested cross-validation, the best model combined Boruta with Boosted GLM (AUC = 0.71, F1 = 0.76). In external validation, the best model combined L1-lasso with boosted GLM (AUC = 0.71, F1 = 0.47). Overall, Boruta, RFE, L1-lasso, and RF variable importance were the top-performing feature selection methods, while the choice of ML classifier didn't significantly affect the results. The ADC-derived features showed the highest discriminatory power with T2w-derived features being less informative, while their combination did not lead to improved performance.

CONCLUSION

The choice of feature selection method and the source of radiomic features have a profound effect on the models' performance for csPCa diagnosis.

CRITICAL RELEVANCE STATEMENT

This work may guide future radiomic research, paving the way for the development of more effective and reliable radiomic models; not only for advancing prostate cancer diagnostic strategies, but also for informing broader applications of radiomics in different medical contexts.

KEY POINTS

Radiomics is a growing field that can still be optimized. Feature selection method impacts radiomics models' performance more than ML algorithms. Best feature selection methods: RFE, LASSO, RF, and Boruta. ADC-derived radiomic features yield more robust models compared to T2w-derived radiomic features.

Explainable 18F-FDG PET/CT radiomics model for predicting EGFR mutation status in lung adenocarcinoma: a two-center study

PURPOSE

To establish an explainable 18F-FDG PET/CT-derived prediction model to identify EGFR mutation status and subtypes (EGFR wild, EGFR-E19, and EGFR-E21) in lung adenocarcinoma (LUAD).

METHODS

Baseline 18F-FDG PET/CT images of 478 patients with LUAD from 2 hospitals were collected. Data from hospital A (n = 390) was randomly split into a training group (n = 312) and an internal test group (n = 78), with data from hospital B (n = 88) utilized for external test. Further, a total of 4,760 handcrafted radiomics features (HRFs) were extracted from PET/CT scans. Candidates for the prediction model were constructed by cross-combinations of 11 feature selection methods and 7 classifiers. The optimal model was determined by combining the results of cross-center data validation and model visualization (Yellowbrick). The predictive performance was assessed via receiver operating characteristic curve, confusion matrix and classification report. Four explainable artificial intelligence technologies were used for optimal model interpretation.

RESULTS

Sex and SUVmax were selected as clinical risk factors, which were then combined with 8 robust PET/CT HRFs to establish the models. The optimal performance was obtained by combining a light gradient boosting machine classifier with random forest feature selection method achieving an optimal performance with a macro-average AUC of 0.75 in the internal test group and 0.81 in the external test group.

CONCLUSION

The explainable EGFR mutation status prediction model have certain clinical practicability and good generalization performance, which may help in the timely selection of treatment options and prognosis prediction in patients with LUAD.

Clinical application of 18F-fluorodeoxyglucose positron emission tomography/computed tomography radiomics-based machine learning analyses in the field of oncology

Machine learning (ML) analyses using 18F-fluorodeoxyglucose (18F-FDG) positron emission tomography (PET)/computed tomography (CT) radiomics features have been applied in the field of oncology. The current review aimed to summarize the current clinical articles about 18F-FDG PET/CT radiomics-based ML analyses to solve issues in classifying or constructing prediction models for several types of tumors. In these studies, lung and mediastinal tumors were the most commonly evaluated lesions, followed by lymphatic, abdominal, head and neck, breast, gynecological, and other types of tumors. Previous studies have commonly shown that 18F-FDG PET radiomics-based ML analysis has good performance in differentiating benign from malignant tumors, predicting tumor characteristics and stage, therapeutic response, and prognosis by examining significant differences in the area under the receiver operating characteristic curves, accuracies, or concordance indices (> 0.70). However, these studies have reported several ML algorithms. Moreover, different ML models have been applied for the same purpose. Thus, various procedures were used in 18F-FDG PET/CT radiomics-based ML analysis in oncology, and 18F-FDG PET/CT radiomics-based ML models, which are easy and universally applied in clinical practice, would be expected to be established.

Methodological evaluation of original articles on radiomics and machine learning for outcome prediction based on positron emission tomography (PET)

AIM

Despite a vast number of articles on radiomics and machine learning in positron emission tomography (PET) imaging, clinical applicability remains limited, partly owing to poor methodological quality. We therefore systematically investigated the methodology described in publications on radiomics and machine learning for PET-based outcome prediction.

METHODS

A systematic search for original articles was run on PubMed. All articles were rated according to 17 criteria proposed by the authors. Criteria with >2 rating categories were binarized into "adequate" or "inadequate". The association between the number of "adequate" criteria per article and the date of publication was examined.

RESULTS

One hundred articles were identified (published between 07/2017 and 09/2023). The median proportion of articles per criterion that were rated "adequate" was 65% (range: 23-98%). Nineteen articles (19%) mentioned neither a test cohort nor cross-validation to separate training from testing. The median number of criteria with an "adequate" rating per article was 12.5 out of 17 (range, 4-17), and this did not increase with later dates of publication (Spearman's rho, 0.094; p = 0.35). In 22 articles (22%), less than half of the items were rated "adequate". Only 8% of articles published the source code, and 10% made the dataset openly available.

CONCLUSION

Among the articles investigated, methodological weaknesses have been identified, and the degree of compliance with recommendations on methodological quality and reporting shows potential for improvement. Better adherence to established guidelines could increase the clinical significance of radiomics and machine learning for PET-based outcome prediction and finally lead to the widespread use in routine clinical practice.

Nuclear medicine radiomics in digestive system tumors: Concept, applications, challenges, and future perspectives

Radiomics aims to develop novel biomarkers and provide relevant deeper subvisual information about pathology, immunophenotype, and tumor microenvironment. It uses automated or semiautomated quantitative analysis of high-dimensional images to improve characterization, diagnosis, and prognosis. Recent years have seen a rapid increase in radiomics applications in nuclear medicine, leading to some promising research results in digestive system oncology, which have been driven by big data analysis and the development of artificial intelligence. Although radiomics advances one step further toward the non-invasive precision medical analysis, it is still a step away from clinical application and faces many challenges. This review article summarizes the available literature on digestive system tumors regarding radiomics in nuclear medicine. First, we describe the workflow and steps involved in radiomics analysis. Subsequently, we discuss the progress in clinical application regarding the utilization of radiomics for distinguishing between various diseases and evaluating their prognosis, and demonstrate how radiomics advances this field. Finally, we offer our viewpoint on how the field can progress by addressing the challenges facing clinical implementation.

Radiomics approaches to predict PD-L1 and PFS in advanced non-small cell lung patients treated with immunotherapy: a multi-institutional study

With the increasing use of immune checkpoint inhibitors (ICIs), there is an urgent need to identify biomarkers to stratify responders and non-responders using programmed death-ligand (PD-L1) expression, and to predict patient-specific outcomes such as progression free survival (PFS). The current study is aimed to determine the feasibility of building imaging-based predictive biomarkers for PD-L1 and PFS through systematically evaluating a combination of several machine learning algorithms with different feature selection methods. A retrospective, multicenter study of 385 advanced NSCLC patients amenable to ICIs was undertaken in two academic centers. Radiomic features extracted from pretreatment CT scans were used to build predictive models for PD-L1 and PFS (short-term vs. long-term survivors). We first employed the LASSO methodology followed by five feature selection methods and seven machine learning approaches to build the predictors. From our analyses, we found several combinations of feature selection methods and machine learning algorithms to achieve a similar performance. Logistic regression with ReliefF feature selection (AUC = 0.64, 0.59 in discovery and validation cohorts) and SVM with Anova F-test feature selection (AUC = 0.64, 0.63 in discovery and validation datasets) were the best-performing models to predict PD-L1 and PFS. This study elucidates the application of suitable feature selection approaches and machine learning algorithms to predict clinical endpoints using radiomics features. Through this study, we identified a subset of algorithms that should be considered in future investigations for building robust and clinically relevant predictive models.

Comparing mechanism-based and machine learning models for predicting the effects of glucose accessibility on tumor cell proliferation

Glucose plays a central role in tumor metabolism and development and is a target for novel therapeutics. To characterize the response of cancer cells to blockade of glucose uptake, we collected time-resolved microscopy data to track the growth of MDA-MB-231 breast cancer cells. We then developed a mechanism-based, mathematical model to predict how a glucose transporter (GLUT1) inhibitor (Cytochalasin B) influences the growth of the MDA-MB-231 cells by limiting access to glucose. The model includes a parameter describing dose dependent inhibition to quantify both the total glucose level in the system and the glucose level accessible to the tumor cells. Four common machine learning models were also used to predict tumor cell growth. Both the mechanism-based and machine learning models were trained and validated, and the prediction error was evaluated by the coefficient of determination (R²). The random forest model provided the highest accuracy predicting cell dynamics (R² = 0.92), followed by the decision tree (R² = 0.89), k-nearest-neighbor regression (R² = 0.84), mechanism-based (R² = 0.77), and linear regression model (R² = 0.69). Thus, the mechanism-based model has a predictive capability comparable to machine learning models with the added benefit of elucidating biological mechanisms.

Optimal 18F-FDG PET/CT radiomics model development for predicting EGFR mutation status and prognosis in lung adenocarcinoma: a multicentric study

Purpose

To develop and interpret optimal predictive models to identify epidermal growth factor receptor (EGFR) mutation status and subtypes in patients with lung adenocarcinoma based on multicentric ¹⁸F-FDG PET/CT data, and further construct a prognostic model to predict their clinical outcome.

Methods

The ¹⁸F-FDG PET/CT imaging and clinical characters of 767 patients with lung adenocarcinoma from 4 cohorts were collected. Seventy-six radiomics candidates using cross-combination method to identity EGFR mutation status and subtypes were built. Further, Shapley additive explanations and local interpretable model-agnostic explanations were used for optimal models' interpretation. Moreover, in order to predict the overall survival, a multivariate Cox proportional hazard model based on handcrafted radiomics features and clinical characteristics was constructed. The predictive performance and clinical net benefit of the models were evaluated via area under receiver operating characteristic (AUC), C-index and decision curve analysis.

Results

Among the 76 radiomics candidates, light gradient boosting machine classifier (LGBM) combined with recursive feature elimination wrapped LGBM feature selection method achieved best performance in predicting EGFR mutation status (AUC reached 0.80, 0.61, 0.71 in the internal test cohort and two external test cohorts, respectively). And extreme gradient boosting classifier combined with support vector machine feature selection method achieved best performance in predicting EGFR subtypes (AUC reached 0.76, 0.63, 0.61 in the internal test cohort and two external test cohorts, respectively). The C-index of the Cox proportional hazard model achieved 0.863.

Conclusions

The integration of cross-combination method and the external validation from multi-center data achieved a good prediction and generalization performance in predicting EGFR mutation status and its subtypes. The combination of handcrafted radiomics features and clinical factors achieved good performance in predicting prognosis. With the urgent needs of multicentric ¹⁸F-FDG PET/CT trails, robust and explainable radiomics models have great potential in decision making and prognosis prediction of lung adenocarcinoma.

Identification of Resting-State Network Functional Connectivity and Brain Structural Signatures in Fibromyalgia Using a Machine Learning Approach

Abnormal resting-state functional connectivity (rs-FC) and brain structure have emerged as pathological hallmarks of fibromyalgia (FM). This study investigated and compared the accuracy of network rs-FC and brain structural features in identifying FM with a machine learning (ML) approach. Twenty-six FM patients and thirty healthy controls were recruited. Clinical presentation was measured by questionnaires. After MRI acquisitions, network rs-FC z-score and network-based gray matter volume matrices were exacted and preprocessed. The performance of feature selection and classification methods was measured. Correlation analyses between predictive features in final models and clinical data were performed. The combination of the recursive feature elimination (RFE) selection method and support vector machine (rs-FC data) or logistic regression (structural data), after permutation importance feature selection, showed high performance in distinguishing FM patients from pain-free controls, in which the rs-FC ML model outperformed the structural ML model (accuracy: 0.91 vs. 0.86, AUC: 0.93 vs. 0.88). The combined rs-FC and structural ML model showed the best performance (accuracy: 0.95, AUC: 0.95). Additionally, several rs-FC features in the final ML model correlated with FM's clinical data. In conclusion, ML models based on rs-FC and brain structural MRI features could effectively differentiate FM patients from pain-free subjects.