Imbalanced Data Correction Based PET/CT Radiomics Model for Predicting Lymph Node Metastasis in Clinical Stage T1 Lung Adenocarcinoma

Predicting the spread through air spaces in lung adenocarcinoma from preoperative 18 F-FDG PET/CT radiomics

OBJECTIVE

This study aimed to develop an effective radiomics-clinical model to preoperatively discriminate the spread through air spaces (STAS) in lung adenocarcinoma (ADC).

METHODS

Data from 192 ADC patients were enrolled, with 2/3 ( n  = 128) allocated as the training cohort and the remaining 1/3 ( n  = 64) designated as the validation cohort. A total of 2212 radiomics features were extracted from PET/computed tomography (PET/CT) images. The least absolute shrinkage and selection operator regression method was applied to select features. Logistic regression was used to construct radiomics and clinical models. Finally, a radiomics-clinical model that combined clinical with radiomics features was developed. The models were evaluated by receiver operating characteristic (ROC) curve and decision curve analysis.

RESULTS

The area under the ROC curve (AUC) of the radiomics-clinical model was 0.924 (95% confidence interval, 0.878-0.969) in the training cohort and 0.919 (0.833-1.000) in the validation cohort. The AUC of the radiomics model was 0.885 (0.825-0.945) in the training cohort and 0.877 (0.766-0.988) in the validation cohort. The AUC of the clinical model was 0.883 (0.814-0.951) in the training cohort and 0.896 (0.7706-1.000) in the validation cohort. The decision curve analysis indicated its clinical usefulness.

CONCLUSION

The PET/CT-based radiomics-clinical model achieved satisfactory performance in discriminating the STAS in ADC preoperatively.

Impact of harmonization and oversampling methods on radiomics analysis of multi-center imbalanced datasets: application to PET-based prediction of lung cancer subtypes

BACKGROUND

Medical imaging data frequently encounter image-generation heterogeneity and class imbalance properties, challenging strong generalized predictive performances with data-driven machine-learning methods. The purpose of this study was to investigate the impact of harmonization and oversampling methods on multi-center imbalanced datasets, with specific application to PET-based radiomics modeling for histologic subtype prediction in non-small cell lung cancer (NSCLC).

METHODS

The retrospective study included 245 patients with adenocarcinoma (ADC) and 78 patients with squamous cell carcinoma (SCC) from 4 centers. Utilizing 1502 radiomics features per patient, we trained, validated, and tested 4 machine-learning classifiers, to investigate the effect of no harmonization (NoH) or 4 feature harmonization methods, paired with no oversampling (NoO) or 5 oversampling methods on subtype prediction. Model performance was evaluated using the average area under the ROC curve (AUROC) and G-mean via 5 times 5-fold cross-validations. Statistical comparisons of the combined models against baseline (NoH + NoO) were performed for each fold of cross-validation using the DeLong test.

RESULTS

The number of cross-combinations with both AUROC and G-mean outperforming baseline in validation and testing was 15, 4, 2, and 7 (out of 29) for random forest (RF), linear discriminant analysis (LDA), logistic regression (LR), and support vector machine (SVM), respectively. ComBat harmonization combined with oversampling (SMOTE) via RF yielded better performance than baseline (AUROC and G-mean of validation: 0.725 vs. 0.608 and 0.625 vs. 0.398; testing: 0.637 vs. 0.567 and 0.506 vs. 0.287), though statistical significances were not observed.

CONCLUSIONS

Applying harmonization and oversampling methods in multi-center imbalanced datasets can improve NSCLC-subtype prediction, but the effect varies widely across classifiers. We have created open-source comparisons of harmonization and oversampling on different classifiers for comprehensive evaluations in different studies.

A Novel Data Augmentation Method for Radiomics Analysis Using Image Perturbations

Radiomics extracts hundreds of features from medical images to quantitively characterize a region of interest (ROI). When applying radiomics, imbalanced or small dataset issues are commonly addressed using under or over-sampling, the latter being applied directly to the extracted features. Aim of this study is to propose a novel balancing and data augmentation technique by applying perturbations (erosion, dilation, contour randomization) to the ROI in cardiac computed tomography images. From the perturbed ROIs, radiomic features are extracted, thus creating additional samples. This approach was tested addressing the clinical problem of distinguishing cardiac amyloidosis (CA) from aortic stenosis (AS) and hypertrophic cardiomyopathy (HCM). Twenty-one CA, thirty-two AS and twenty-one HCM patients were included in the study. From each original and perturbed ROI, 107 radiomic features were extracted. The CA-AS dataset was balanced using the perturbation-based method along with random over-sampling, adaptive synthetic (ADASYN) and the synthetic minority oversampling technique (SMOTE). The same methods were tested to perform data augmentation dealing with CA and HCM. Features were submitted to robustness, redundancy, and relevance analysis testing five feature selection methods (p-value, least absolute shrinkage and selection operator (LASSO), semi-supervised LASSO, principal component analysis (PCA), semi-supervised PCA). Support vector machine performed the classification tasks, and its performance were evaluated by means of a 10-fold cross-validation. The perturbation-based approach provided the best performances in terms of f1 score and balanced accuracy in both CA-AS (f1 score: 80%, AUC: 0.91) and CA-HCM (f1 score: 86%, AUC: 0.92) classifications. These results suggest that ROI perturbations represent a powerful approach to address both data balancing and augmentation issues.

Value of Presurgical 18F-FDG PET/CT Radiomics for Predicting Mediastinal Lymph Node Metastasis in Patients with Lung Adenocarcinoma

Objective: The aim of this study was to develop a 18F-fluorodeoxyglucose (18F-FDG) positron emission tomography/computed tomography (PET/CT) radiomic model for predicting mediastinal lymph node metastasis (LNM) in presurgical patients with lung adenocarcinoma. Methods: The study enrolled 320 patients with lung adenocarcinoma (288 internal and 32 external cases) and extracted 190 radiomic features using the LIFEx package. Optimal radiomic features to build a radiomic model were selected using the least absolute shrinkage and selection operator algorithm. Logistic regression was used to build the clinical and complex (combined radiomic and clinical variables) models. Results: Ten radiomic features were selected. In the training group, the area under the receiver operating characteristic curve of the complex model was significantly higher than that of the radiomic and clinical models [0.924 (95% CI: 0.887-0.961) vs. 0.863 (95% CI: 0.814-0.912; p = 0.001) and 0.838 (95% CI: 0.783-0.894; p = 0.000), respectively]. The sensitivity, specificity, accuracy, and positive and negative predictive values of the radiomic model were 0.857, 0.790, 0.811, and 0.651 and 0.924, respectively, which were better than that of visual evaluation (0.539, 0.724, 0.667, and 0.472 and 0.775, respectively) and PET semiquantitative analyses (0.619, 0.732, 0.697, and 0.513 and 0.808, respectively). Conclusions: 18F-FDG PET/CT radiomics showed good predictive performance for LNM and improved the N-stage accuracy of lung adenocarcinoma.

Predictive value of radiomic features extracted from primary lung adenocarcinoma in forecasting thoracic lymph node metastasis: a systematic review and meta-analysis

BACKGROUND

The application of radiomics in thoracic lymph node metastasis (LNM) of lung adenocarcinoma is increasing, but diagnostic performance of radiomics from primary tumor to predict LNM has not been systematically reviewed. Therefore, this study sought to provide a general overview regarding the methodological quality and diagnostic performance of using radiomic approaches to predict the likelihood of LNM in lung adenocarcinoma.

METHODS

Studies were gathered from literature databases such as PubMed, Embase, the Web of Science Core Collection, and the Cochrane library. The Radiomic Quality Score (RQS) and the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) were both used to assess the quality of each study. The pooled sensitivity, specificity, and area under the curve (AUC) of the best radiomics models in the training and validation cohorts were calculated. Subgroup and meta-regression analyses were also conducted.

RESULTS

Seventeen studies with 159 to 1202 patients each were enrolled between the years of 2018 to 2022, of which ten studies had sufficient data for the quantitative evaluation. The percentage of RQS was between 11.1% and 44.4% and most of the studies were considered to have a low risk of bias and few applicability concerns in QUADAS-2. Pyradiomics and logistic regression analysis were the most commonly used software and methods for radiomics feature extraction and selection, respectively. In addition, the best prediction models in seventeen studies were mainly based on radiomics features combined with non-radiomics features (semantic features and/or clinical features). The pooled sensitivity, specificity, and AUC of the training cohorts were 0.84 (95% confidence interval (CI) [0.73-0.91]), 0.88 (95% CI [0.81-0.93]), and 0.93(95% CI [0.90-0.95]), respectively. For the validation cohorts, the pooled sensitivity, specificity, and AUC were 0.89 (95% CI [0.82-0.94]), 0.86 (95% CI [0.74-0.93]) and 0.94 (95% CI [0.91-0.96]), respectively.

CONCLUSIONS

Radiomic features based on the primary tumor have the potential to predict preoperative LNM of lung adenocarcinoma. However, radiomics workflow needs to be standardized to better promote the applicability of radiomics.

TRIAL REGISTRATION

CRD42022375712.

Use of CT-derived radiomic features to preoperatively identify invasive mucinous adenocarcinoma in solitary pulmonary nodules ≤3 cm

Objective

In this study, we aimed to utilize computed tomography (CT)-derived radiomics and various machine learning approaches to differentiate between invasive mucinous adenocarcinoma (IMA) and invasive non-mucinous adenocarcinoma (INMA) preoperatively in solitary pulmonary nodules (SPN) ≤3 cm.

Methods

A total of 538 patients with SPNs measuring ≤3 cm were enrolled, categorized into either the IMA group (n = 50) or INMA group (n = 488) based on postoperative pathology. Radiomic features were extracted from non-contrast-enhanced CT scans and identified using the least absolute shrinkage and selection operator (LASSO) algorithm. In constructing radiomics-based models, logistic regression, support vector machines, classification and regression trees, and k-nearest neighbors were employed. Additionally, a clinical model was developed, focusing on CT radiological features. Subsequently, this clinical model was integrated with the most effective radiomic model to create a combined model. Performance assessments of these models were conducted, utilizing metrics such as the area under the receiver operating characteristic curve (AUC), DeLong's test, net reclassification index (NRI), and integrated discrimination improvement (IDI).

Results

The support vector machine approach showed superior predictive efficiency, with AUCs of 0.829 and 0.846 in the training and test cohorts, respectively. The clinical model had AUCs of 0.760 and 0.777 in the corresponding cohorts. The combined model had AUCs of 0.847 and 0.857 in the corresponding cohorts. Furthermore, compared to the radiomic model, the combined model significantly improved performance in both the training (DeLong test P = 0.045, NRI 0.206, IDI 0.024) and test cohorts (P = 0.029, NRI 0.125, IDI 0.032), as well as compared to the clinical model in both the training (P = 0.01, NRI 0.310, IDI 0.09) and test cohorts (P = 0.047, NRI 0.382, IDI 0.085).

Conclusion

the combined model exhibited excellent performance in distinguishing between IMA and INMA in SPNs ≤3 cm.

AI/ML advances in non-small cell lung cancer biomarker discovery

Lung cancer is the leading cause of cancer deaths among both men and women, representing approximately 25% of cancer fatalities each year. The treatment landscape for non-small cell lung cancer (NSCLC) is rapidly evolving due to the progress made in biomarker-driven targeted therapies. While advancements in targeted treatments have improved survival rates for NSCLC patients with actionable biomarkers, long-term survival remains low, with an overall 5-year relative survival rate below 20%. Artificial intelligence/machine learning (AI/ML) algorithms have shown promise in biomarker discovery, yet NSCLC-specific studies capturing the clinical challenges targeted and emerging patterns identified using AI/ML approaches are lacking. Here, we employed a text-mining approach and identified 215 studies that reported potential biomarkers of NSCLC using AI/ML algorithms. We catalogued these studies with respect to BEST (Biomarkers, EndpointS, and other Tools) biomarker sub-types and summarized emerging patterns and trends in AI/ML-driven NSCLC biomarker discovery. We anticipate that our comprehensive review will contribute to the current understanding of AI/ML advances in NSCLC biomarker research and provide an important catalogue that may facilitate clinical adoption of AI/ML-derived biomarkers.

Automated Classification of Lung Cancer Subtypes Using Deep Learning and CT-Scan Based Radiomic Analysis

Artificial intelligence and emerging data science techniques are being leveraged to interpret medical image scans. Traditional image analysis relies on visual interpretation by a trained radiologist, which is time-consuming and can, to some degree, be subjective. The development of reliable, automated diagnostic tools is a key goal of radiomics, a fast-growing research field which combines medical imaging with personalized medicine. Radiomic studies have demonstrated potential for accurate lung cancer diagnoses and prognostications. The practice of delineating the tumor region of interest, known as segmentation, is a key bottleneck in the development of generalized classification models. In this study, the incremental multiple resolution residual network (iMRRN), a publicly available and trained deep learning segmentation model, was applied to automatically segment CT images collected from 355 lung cancer patients included in the dataset "Lung-PET-CT-Dx", obtained from The Cancer Imaging Archive (TCIA), an open-access source for radiological images. We report a failure rate of 4.35% when using the iMRRN to segment tumor lesions within plain CT images in the lung cancer CT dataset. Seven classification algorithms were trained on the extracted radiomic features and tested for their ability to classify different lung cancer subtypes. Over-sampling was used to handle unbalanced data. Chi-square tests revealed the higher order texture features to be the most predictive when classifying lung cancers by subtype. The support vector machine showed the highest accuracy, 92.7% (0.97 AUC), when classifying three histological subtypes of lung cancer: adenocarcinoma, small cell carcinoma, and squamous cell carcinoma. The results demonstrate the potential of AI-based computer-aided diagnostic tools to automatically diagnose subtypes of lung cancer by coupling deep learning image segmentation with supervised classification. Our study demonstrated the integrated application of existing AI techniques in the non-invasive and effective diagnosis of lung cancer subtypes, and also shed light on several practical issues concerning the application of AI in biomedicine.

Automated data preparation for in vivo tumor characterization with machine learning

Background

This study proposes machine learning-driven data preparation (MLDP) for optimal data preparation (DP) prior to building prediction models for cancer cohorts.

Methods

A collection of well-established DP methods were incorporated for building the DP pipelines for various clinical cohorts prior to machine learning. Evolutionary algorithm principles combined with hyperparameter optimization were employed to iteratively select the best fitting subset of data preparation algorithms for the given dataset. The proposed method was validated for glioma and prostate single center cohorts by 100-fold Monte Carlo (MC) cross-validation scheme with 80-20% training-validation split ratio. In addition, a dual-center diffuse large B-cell lymphoma (DLBCL) cohort was utilized with Center 1 as training and Center 2 as independent validation datasets to predict cohort-specific clinical endpoints. Five machine learning (ML) classifiers were employed for building prediction models across all analyzed cohorts. Predictive performance was estimated by confusion matrix analytics over the validation sets of each cohort. The performance of each model with and without MLDP, as well as with manually-defined DP were compared in each of the four cohorts.

Results

Sixteen of twenty established predictive models demonstrated area under the receiver operator characteristics curve (AUC) performance increase utilizing the MLDP. The MLDP resulted in the highest performance increase for random forest (RF) (+0.16 AUC) and support vector machine (SVM) (+0.13 AUC) model schemes for predicting 36-months survival in the glioma cohort. Single center cohorts resulted in complex (6-7 DP steps) DP pipelines, with a high occurrence of outlier detection, feature selection and synthetic majority oversampling technique (SMOTE). In contrast, the optimal DP pipeline for the dual-center DLBCL cohort only included outlier detection and SMOTE DP steps.

Conclusions

This study demonstrates that data preparation prior to ML prediction model building in cancer cohorts shall be ML-driven itself, yielding optimal prediction models in both single and multi-centric settings.

CT-based radiomics in predicting pathological response in non-small cell lung cancer patients receiving neoadjuvant immunotherapy

Objectives

In radiomics, high-throughput algorithms extract objective quantitative features from medical images. In this study, we evaluated CT-based radiomics features, clinical features, in-depth learning features, and a combination of features for predicting a good pathological response (GPR) in non-small cell lung cancer (NSCLC) patients receiving immunotherapy-based neoadjuvant therapy (NAT).

Materials and methods

We reviewed 62 patients with NSCLC who received surgery after immunotherapy-based NAT and collected clinicopathological data and CT images before and after immunotherapy-based NAT. A series of image preprocessing was carried out on CT scanning images: tumor segmentation, conventional radiomics feature extraction, deep learning feature extraction, and normalization. Spearman correlation coefficient, principal component analysis (PCA), and least absolute shrinkage and selection operator (LASSO) were used to screen features. The pretreatment traditional radiomics combined with clinical characteristics (before_rad_cil) model and pretreatment deep learning characteristics (before_dl) model were constructed according to the data collected before treatment. The data collected after NAT created the after_rad_cil model and after_dl model. The entire model was jointly constructed by all clinical features, conventional radiomics features, and deep learning features before and after neoadjuvant treatment. Finally, according to the data obtained before and after treatment, the before_nomogram and after_nomogram were constructed.

Results

In the before_rad_cil model, four traditional radiomics features ("original_shape_flatness," "wavelet hhl_firer_skewness," "wavelet hlh_firer_skewness," and "wavelet lll_glcm_correlation") and two clinical features ("gender" and "N stage") were screened out to predict a GPR. The average prediction accuracy (ACC) after modeling with k-nearest neighbor (KNN) was 0.707. In the after_rad_cil model, nine features predictive of GPR were obtained after feature screening, among which seven were traditional radiomics features: "exponential_firer_skewness," "exponential_glrlm_runentropy," "log- sigma-5-0-mm-3d_firer_kurtosis," "logarithm_skewness," "original_shape_elongation," "original_shape_brilliance," and "wavelet llh_glcm_clustershade"; two were clinical features: "after_CRP" and "after lymphocyte percentage." The ACC after modeling with support vector machine (SVM) was 0.682. The before_dl model and after_dl model were modeled by SVM, and the ACC was 0.629 and 0.603, respectively. After feature screening, the entire model was constructed by multilayer perceptron (MLP), and the ACC of the GPR was the highest, 0.805. The calibration curve showed that the predictions of the GPR by the before_nomogram and after_nomogram were in consensus with the actual GPR.

Conclusion

CT-based radiomics has a good predictive ability for a GPR in NSCLC patients receiving immunotherapy-based NAT. Among the radiomics features combined with the clinicopathological information model, deep learning feature model, and the entire model, the entire model had the highest prediction accuracy.

MRI-based radiomics analysis for preoperative evaluation of lymph node metastasis in hypopharyngeal squamous cell carcinoma

Objective

To investigate the role of pre-treatment magnetic resonance imaging (MRI) radiomics for the preoperative prediction of lymph node (LN) metastasis in patients with hypopharyngeal squamous cell carcinoma (HPSCC).

Methods

A total of 155 patients with HPSCC were eligibly enrolled from single institution. Radiomics features were extracted from contrast-enhanced axial T-1 weighted (CE-T1WI) sequence. The most relevant features of LN metastasis were selected by the least absolute shrinkage and selection operator (LASSO) method. Univariate and multivariate logistic regression analysis was adopted to determine the independent clinical risk factors. Three models were constructed to predict the LN metastasis status: one using radiomics only, one using clinical factors only, and the other one combined radiomics and clinical factors. Receiver operating characteristic (ROC) curves and calibration curve were used to evaluate the discrimination and the accuracy of the models, respectively. The performances were tested by an internal validation cohort (n=47). The clinical utility of the models was assessed by decision curve analysis.

Results

The nomogram consisted of radiomics scores and the MRI-reported LN status showed satisfactory discrimination in the training and validation cohorts with AUCs of 0.906 (95% CI, 0.840 to 0.972) and 0.853 (95% CI, 0.739 to 0.966), respectively. The nomogram, i.e., the combined model, outperformed the radiomics and MRI-reported LN status in both discrimination and clinical usefulness.

Conclusions

The MRI-based radiomics nomogram holds promise for individual and non-invasive prediction of LN metastasis in patients with HPSCC.

Predicting Local Failure after Partial Prostate Re-Irradiation Using a Dosiomic-Based Machine Learning Model

The aim of this study is to predict local failure after partial prostate re-irradiation for the treatment of isolated locally recurrent prostate cancer by using a machine learning classifier based on radiomic features from pre-treatment computed tomography (CT), positron-emission tomography (PET) and biological effective dose distribution (BED) of the radiotherapy plan. The analysis was conducted on a monocentric dataset of 43 patients with evidence of isolated intraprostatic recurrence of prostate cancer after primary external beam radiotherapy. All patients received partial prostate re-irradiation delivered by volumetric modulated arc therapy. The gross tumor volume (GTV) of each patient was manually contoured from planning CT, choline-PET and dose maps. An ensemble machine learning pipeline including unbalanced data correction and feature selection was trained using the radiomic and dosiomic features as input for predicting occurrence of local failure. The model performance was assessed using sensitivity, specificity, accuracy and area under receiver operating characteristic curves of the score function in 10-fold cross validation repeated 100 times. Local failure was observed in 13 patients (30%), with a median time to recurrence of 36.7 months (range = 6.1-102.4 months). A four variables ensemble machine learning model resulted in accuracy of 0.62 and AUC 0.65. According to our results, a dosiomic machine learning classifier can predict local failure after partial prostate re-irradiation.

Radiomics in Oncological PET Imaging: A Systematic Review-Part 1, Supradiaphragmatic Cancers

Radiomics is an upcoming field in nuclear oncology, both promising and technically challenging. To summarize the already undertaken work on supradiaphragmatic neoplasia and assess its quality, we performed a literature search in the PubMed database up to 18 February 2022. Inclusion criteria were: studies based on human data; at least one specified tumor type; supradiaphragmatic malignancy; performing radiomics on PET imaging. Exclusion criteria were: studies only based on phantom or animal data; technical articles without a clinically oriented question; fewer than 30 patients in the training cohort. A review database containing PMID, year of publication, cancer type, and quality criteria (number of patients, retrospective or prospective nature, independent validation cohort) was constructed. A total of 220 studies met the inclusion criteria. Among them, 119 (54.1%) studies included more than 100 patients, 21 studies (9.5%) were based on prospectively acquired data, and 91 (41.4%) used an independent validation set. Most studies focused on prognostic and treatment response objectives. Because the textural parameters and methods employed are very different from one article to another, it is complicated to aggregate and compare articles. New contributions and radiomics guidelines tend to help improving quality of the reported studies over the years.