[18 F]-FDG PET radiomic model as prognostic biomarker in diffuse large B-cell lymphoma.

To evaluate the association between radiomic features (RFs) extracted from 18 F-FDG PET/CT (18 F-FDG-PET) with progression-free survival (PFS) and overall survival (OS) in diffuse large-B-cell lymphoma (DLBCL) patients eligible to first-line chemotherapy. DLBCL patients who underwent 18 F-FDG-PET prior to first-line chemotherapy were retrospectively analyzed. RFs were extracted from the lesion showing the highest uptake. A radiomic score to predict PFS and OS was obtained by multivariable Elastic Net Cox model. Radiomic univariate model, clinical and combined clinical-radiomic multivariable models to predict PFS and OS were obtained. 112 patients were analyzed. Median follow-up was 34.7 months (Inter-Quartile Range (IQR) 11.3-66.3 months) for PFS and 41.1 (IQR 18.4-68.9) for OS. Radiomic score resulted associated with PFS and OS (p < 0.001), outperforming conventional PET parameters. C-index (95% CI) for PFS prediction were 0.67 (0.58-0.76), 0.81 (0.75-0.88) and 0.84 (0.77-0.91) for clinical, radiomic and combined clinical-radiomic model, respectively. C-index for OS were 0.77 (0.66-0.89), 0.84 (0.76-0.91) and 0.90 (0.81-0.98). In the Kaplan-Meier analysis (low-IPI vs. high-IPI), the radiomic score was significant predictor of PFS (p < 0.001). The radiomic score was an independent prognostic biomarker of survival in DLBCL patients. The extraction of RFs from baseline 18 F-FDG-PET might be proposed in DLBCL to stratify high-risk versus low-risk patients of relapse after first-line therapy, especially in low-IPI patients.

Comparison of automated segmentation techniques for magnetic resonance images of the prostate.

BACKGROUND: Contouring of anatomical regions is a crucial step in the medical workflow and is both time-consuming and prone to intra- and inter-observer variability. This study compares different strategies for automatic segmentation of the prostate in T2-weighted MRIs. METHODS: This study included 100 patients diagnosed with prostate adenocarcinoma who had undergone multi-parametric MRI and prostatectomy. From the T2-weighted MR images, ground truth segmentation masks were established by consensus from two expert radiologists. The prostate was then automatically contoured with six different methods: (1) a multi-atlas algorithm, (2) a proprietary algorithm in the Syngo.Via medical imaging software, and four deep learning models: (3) a V-net trained from scratch, (4) a pre-trained 2D U-net, (5) a GAN extension of the 2D U-net, and (6) a segmentation-adapted EfficientDet architecture. The resulting segmentations were compared and scored against the ground truth masks with one 70/30 and one 50/50 train/test data split. We also analyzed the association between segmentation performance and clinical variables. RESULTS: The best performing method was the adapted EfficientDet (model 6), achieving a mean Dice coefficient of 0.914, a mean absolute volume difference of 5.9%, a mean surface distance (MSD) of 1.93 pixels, and a mean 95th percentile Hausdorff distance of 3.77 pixels. The deep learning models were less prone to serious errors (0.854 minimum Dice and 4.02 maximum MSD), and no significant relationship was found between segmentation performance and clinical variables. CONCLUSIONS: Deep learning-based segmentation techniques can consistently achieve Dice coefficients of 0.9 or above with as few as 50 training patients, regardless of architectural archetype. The atlas-based and Syngo.via methods found in commercial clinical software performed significantly worse (0.855[Formula: see text]0.887 Dice).

Brain metastases from NSCLC treated with stereotactic radiotherapy: prediction mismatch between two different radiomic platforms.

BACKGROUND AND PURPOSE: Radiomics enables the mining of quantitative features from medical images. The influence of the radiomic feature extraction software on the final performance of models is still a poorly understood topic. This study aimed to investigate the ability of radiomic features extracted by two different radiomic platforms to predict clinical outcomes in patients treated with radiosurgery for brain metastases from non-small cell lung cancer. We developed models integrating pre-treatment magnetic resonance imaging (MRI)-derived radiomic features and clinical data. MATERIALS AND METHODS: Pre-radiotherapy gadolinium enhanced axial T1-weighted MRI scans were used. MRI images were re-sampled, intensity-shifted, and histogram-matched before radiomic extraction by means of two different platforms (PyRadiomics and SOPHiA Radiomics). We adopted LASSO Cox regression models for multivariable analyses by creating radiomic, clinical, and combined models using three survival clinical endpoints (local control, distant progression, and overall survival). The statistical analysis was repeated 50 times with different random seeds and the median concordance index was used as performance metric of the models. RESULTS: We analysed 276 metastases from 148 patients. The use of the two platforms resulted in differences in both the quality and the number of extractable features. That led to mismatches in terms of end-to-end performance, statistical significance of radiomic scores, and clinical covariates found significant in combined models. CONCLUSION: This study shed new light on how extracting radiomic features from the same images using two different platforms could yield several discrepancies. That may lead to acute consequences on drawing conclusions, comparing results across the literature, and translating radiomics into clinical practice.

Gene expression profiling and FDG-PET radiomics uncover radiometabolic signatures associated with outcome in DLBCL.

Emerging evidence indicates that chemoresistance is closely related to altered metabolism in cancer. Here, we hypothesized that distinct metabolic gene expression profiling (GEP) signatures might be correlated with outcome and with specific fluorodeoxyglucose positron emission tomography (FDG-PET) radiomic profiles in diffuse large B-cell lymphoma (DLBCL). We retrospectively analyzed a discovery cohort of 48 consecutive patients with DLBCL treated at our center with standard first-line chemoimmunotherapy by performing targeted GEP (T-GEP)- and FDG-PET radiomic analyses on the same target lesions at baseline. T-GEP-based metabolic profiling identified a 6-gene signature independently associated with outcomes in univariate and multivariate analyses. This signature included genes regulating mitochondrial oxidative metabolism (SCL25A1, PDK4, PDPR) that were upregulated and was inversely associated with genes involved in hypoxia and glycolysis (MAP2K1, HIF1A, GBE1) that were downregulated. These data were validated in 2 large publicly available cohorts. By integrating FDG-PET radiomics and T-GEP, we identified a radiometabolic signature (RadSig) including 4 radiomic features (histo kurtosis, histo energy, shape sphericity, and neighboring gray level dependence matrix contrast), significantly associated with the metabolic GEP-based signature (r = 0.43, P = .0027) and with progression-free survival (P = .028). These results were confirmed using different target lesions, an alternative segmentation method, and were validated in an independent cohort of 64 patients. RadSig retained independent prognostic value in relation to the International Prognostic Index score and metabolic tumor volume (MTV). Integration of RadSig and MTV further refined prognostic stratification. This study provides the proof of principle for the use of FDG-PET radiomics as a tool for noninvasive assessment of cancer metabolism and prognostic stratification in DLBCL.

Association between Contrast-Enhanced Computed Tomography Radiomic Features, Genomic Alterations and Prognosis in Advanced Lung Adenocarcinoma Patients.

Non-invasive methods to assess mutational status, as well as novel prognostic biomarkers, are warranted to foster therapy personalization of patients with advanced non-small cell lung cancer (NSCLC). This study investigated the association of contrast-enhanced Computed Tomography (CT) radiomic features of lung adenocarcinoma lesions, alone or integrated with clinical parameters, with tumor mutational status (EGFR, KRAS, ALK alterations) and Overall Survival (OS). In total, 261 retrospective and 48 prospective patients were enrolled. A Radiomic Score (RS) was created with LASSO-Logistic regression models to predict mutational status. Radiomic, clinical and clinical-radiomic models were trained on retrospective data and tested (Area Under the Curve, AUC) on prospective data. OS prediction models were trained and tested on retrospective data with internal cross-validation (C-index). RS significantly predicted each alteration at training (radiomic and clinical-radiomic AUC 0.95-0.98); validation performance was good for EGFR (AUC 0.86), moderate for KRAS and ALK (AUC 0.61-0.65). RS was also associated with OS at univariate and multivariable analysis, in the latter with stage and type of treatment. The validation C-index was 0.63, 0.79, and 0.80 for clinical, radiomic, and clinical-radiomic models. The study supports the potential role of CT radiomics for non-invasive identification of gene alterations and prognosis prediction in patients with advanced lung adenocarcinoma, to be confirmed with independent studies.

Impact of image filtering and assessment of volume-confounding effects on CT radiomic features and derived survival models in non-small cell lung cancer.

Background: No evidence supports the choice of specific imaging filtering methodologies in radiomics. As the volume of the primary tumor is a well-recognized prognosticator, our purpose is to assess how filtering may impact the feature/volume dependency in computed tomography (CT) images of non-small cell lung cancer (NSCLC), and if such impact translates into differences in the performance of survival modeling. The role of lesion volume in model performances was also considered and discussed. Methods: Four-hundred seventeen CT images NSCLC patients were retrieved from the NSCLC-Radiomics public repository. Pre-processing and features extraction were implemented using Pyradiomics v3.0.1. Features showing high correlation with volume across original and filtered images were excluded. Cox proportional hazards (PH) with least absolute shrinkage and selection operator (LASSO) regularization and CatBoost models were built with and without volume, and their concordance (C-) indices were compared using Wilcoxon signed-ranked test. The Mann Whitney U test was used to assess model performances after stratification into two groups based on low- and high-volume lesions. Results: Radiomic models significantly outperformed models built on only clinical variables and volume. However, the exclusion/inclusion of volume did not generally alter the performances of radiomic models. Overall, performances were not substantially affected by the choice of either imaging filter (overall C-index 0.539-0.590 for Cox PH and 0.589-0.612 for CatBoost). The separation of patients with high-volume lesions resulted in significantly better performances in 2/10 and 7/10 cases for Cox PH and CatBoost models, respectively. Both low- and high-volume models performed significantly better with the inclusion of radiomic features (P<0.0001), but the improvement was largest in the high-volume group (+10.2% against +8.7% improvement for CatBoost models and +10.0% against +5.4% in Cox PH models). Conclusions: Radiomic features complement well-known prognostic factors such as volume, but their volume-dependency is high and should be managed with vigilance. The informative content of radiomic features may be diminished in small lesion volumes, which could limit the applicability of radiomics in early-stage NSCLC, where tumors tend to be small. Our results also suggest an advantage of CatBoost models over the Cox PH models.

Quality assurance for automatically generated contours with additional deep learning.

OBJECTIVE: Deploying an automatic segmentation model in practice should require rigorous quality assurance (QA) and continuous monitoring of the model's use and performance, particularly in high-stakes scenarios such as healthcare. Currently, however, tools to assist with QA for such models are not available to AI researchers. In this work, we build a deep learning model that estimates the quality of automatically generated contours. METHODS: The model was trained to predict the segmentation quality by outputting an estimate of the Dice similarity coefficient given an image contour pair as input. Our dataset contained 60 axial T2-weighted MRI images of prostates with ground truth segmentations along with 80 automatically generated segmentation masks. The model we used was a 3D version of the EfficientDet architecture with a custom regression head. For validation, we used a fivefold cross-validation. To counteract the limitation of the small dataset, we used an extensive data augmentation scheme capable of producing virtually infinite training samples from a single ground truth label mask. In addition, we compared the results against a baseline model that only uses clinical variables for its predictions. RESULTS: Our model achieved a mean absolute error of 0.020 ± 0.026 (2.2% mean percentage error) in estimating the Dice score, with a rank correlation of 0.42. Furthermore, the model managed to correctly identify incorrect segmentations (defined in terms of acceptable/unacceptable) 99.6% of the time. CONCLUSION: We believe that the trained model can be used alongside automatic segmentation tools to ensure quality and thus allow intervention to prevent undesired segmentation behavior.

The Challenge of Choosing the Best Classification Method in Radiomic Analyses: Recommendations and Applications to Lung Cancer CT Images.

Radiomics uses high-dimensional sets of imaging features to predict biological characteristics of tumors and clinical outcomes. The choice of the algorithm used to analyze radiomic features and perform predictions has a high impact on the results, thus the identification of adequate machine learning methods for radiomic applications is crucial. In this study we aim to identify suitable approaches of analysis for radiomic-based binary predictions, according to sample size, outcome balancing and the features-outcome association strength. Simulated data were obtained reproducing the correlation structure among 168 radiomic features extracted from Computed Tomography images of 270 Non-Small-Cell Lung Cancer (NSCLC) patients and the associated to lymph node status. Performances of six classifiers combined with six feature selection (FS) methods were assessed on the simulated data using AUC (Area Under the Receiver Operating Characteristics Curves), sensitivity, and specificity. For all the FS methods and regardless of the association strength, the tree-based classifiers Random Forest and Extreme Gradient Boosting obtained good performances (AUC ≥ 0.73), showing the best trade-off between sensitivity and specificity. On small samples, performances were generally lower than in large-medium samples and with larger variations. FS methods generally did not improve performances. Thus, in radiomic studies, we suggest evaluating the choice of FS and classifiers, considering specific sample size, balancing, and association strength.