Robustness of radiomics within photon-counting detector CT: impact of acquisition and reconstruction factors

European Radiology, Год журнала: 2025, Номер unknown

Опубликована: Янв. 31, 2025

To assess the impact of acquisition and reconstruction factors on robustness radiomics within photon-counting detector CT (PCD-CT). A phantom with twenty-eight texture materials was scanned different including reposition, scan mode (standard vs high-pitch), tube voltage (120 kVp 140 kVp), slice thickness (1.0 mm 0.4 mm), radiation dose level (0.5 mGy, 1.0 3.0 5.0 10.0 mGy), quantum iterative (0/4, 2/4, 4/4), kernel (Qr40, Qr44, Qr48). Thirteen sets virtual monochromatic images at 70-keV were reconstructed. The regions interest drawn rigid registrations. Ninety-three features extracted from each material. reproducibility evaluated using intraclass correlation coefficient (ICC) concordance (CCC). variability assessed by variation (CV) quartile dispersion (QCD). percentage ICC > 0.90 CCC high when repositioned (88.2% 88.2%) changed (87.1% 87.1%), but none high-pitch used. CV < 10% QCD (47.3% 68.8%) (64.2% 71.0%), that low between standard scans (16.1% 26.9%) (19.4% 29.0%). PCD-CT robust to voltage, dose, strength level, kernel, brittle thickness. Question stability against should be fully determined before academic research clinical application. Findings are Clinical relevance influence voxel size set careful attention PCD-CT, allow a higher implementation analysis in routine.

Язык: Английский

---

The Avocado Sign: A novel imaging marker for nodal staging in rectal cancer

European Radiology, Год журнала: 2025, Номер unknown

Опубликована: Фев. 26, 2025

Язык: Английский

---

Predicting Surgical Difficulty in Rectal Cancer Surgery: A Systematic Review of Artificial Intelligence Models Applied to Pre-Operative MRI

Cancers, Год журнала: 2025, Номер 17(5), С. 812 - 812

Опубликована: Фев. 26, 2025

Introduction: Following the rapid advances in minimally invasive surgery, there are a multitude of surgical modalities available for resecting rectal cancers. Robotic resections represent current pinnacle approaches. Currently, decisions on modality depend local resources and expertise team. Given limited access to robotic developing tools based pre-operative data that can predict difficulty surgery would streamline efficient utilisation resources. This systematic review aims appraise existing literature artificial intelligence (AI)-driven preoperative MRI analysis prediction identify knowledge gaps promising models warranting further clinical evaluation. Methods: A narrative synthesis were undertaken accordance with PRISMA SWiM guidelines. Systematic searches performed Medline, Embase, CENTRAL Trials register. Studies published between 2012 2024 included where AI was applied imaging adult cancer patients undergoing surgeries, any approach, purpose stratifying difficulty. Data extracted according pre-specified protocol capture study characteristics design; objectives performance outcome metrics summarised. Results: database returned 568 articles, 40 ultimately this review. support assessments identified across eight domains (direct grading, extramural vascular invasion (EMVI), lymph node metastasis (LNM), lymphovascular (LVI), perineural (PNI), T staging, requirement multiple linear stapler firings. For each, at least one model very good (AUC scores >0.80), several showing excellent considerably above threshold. Conclusions: assessment surgeries emerging, progressing development strong many models. These warrant evaluation, which aid personalised approaches ensure adequate

Язык: Английский

---

MRI radiomic study on prediction of nonenlarged lymph node metastasis of rectal cancer: reduced field-of-view versus conventional DWI

European Radiology Experimental, Год журнала: 2025, Номер 9(1)

Опубликована: Март 22, 2025

Abstract Background Nonenlarged lymph node metastasis (NELNM) of rectal cancer is easily overlooked because these apparently normal nodes are sometimes too small to measure directly using imaging techniques. Radiomic-based multiparametric sequences could predict NELNM based on the primary lesion cancer. We aimed study performance magnetic resonance (MRI) radiomics derived from reduced field-of-view diffusion-weighted (rDWI) and conventional DWI (cDWI) for prediction NELNM. Methods A total 86 patients (60 26 in training test cohorts, respectively), underwent MRI. Radiomic features were extracted whole segmented T2-weighted (T2WI), rDWI, cDWI, both with b -value 800 s/mm 2 apparent diffusion coefficient (ADC) maps (rADC cADC). The radiomic models above methods built assessment status. Their diagnostic performances evaluated comparison subjective evaluation by radiologists. Results rADC demonstrated a significant advantage over predicting cohorts ( p ≤ 0.002). In cohort, exhibited significantly higher area under receiver operating characteristics curve than cADC, cDWIb800, T2WI 0.020) assessing region-of-interest (ROI) delineation while excelling rDWIb800 = 0.0498). Conclusion outperformed those fDWI status cancer, was more advantageous Relevance statement Advanced rDWI excelled cDWI best observed rADC, contrast rDWIb800, T2WI. Key Points help assess For radiomics, ADC map accurate reliable region interest delineation. Graphical

Язык: Английский

---

Artificial intelligence in magnetic resonance imaging for predicting lymph node metastasis in rectal cancer patients: a meta-analysis

European Radiology, Год журнала: 2025, Номер unknown

Опубликована: Апрель 12, 2025

Язык: Английский

---

Non-invasive Prediction of Lymph Node Metastasis in NSCLC Using Clinical, Radiomics, and Deep Learning Features From 18F-FDG PET/CT Based on Interpretable Machine Learning

Academic Radiology, Год журнала: 2024, Номер unknown

Опубликована: Дек. 1, 2024

PurposeThis study aimed to develop and evaluate a machine learning model combining clinical, radiomics, deep features derived from PET/CT imaging predict lymph node metastasis (LNM) in patients with non-small cell lung cancer (NSCLC). The model's interpretability was enhanced using Shapley additive explanations (SHAP).MethodsA total of 248 NSCLC who underwent preoperative scans were included divided into training, test, external validation sets. Radiomics extracted segmented tumor regions on images, generated the ResNet50 architecture. Feature selection performed minimum-redundancy maximum-relevance (mRMR), least absolute shrinkage operator (LASSO) algorithm. Four models—clinical, radiomics (DL_radiomics), combined model—were constructed XGBoost algorithm evaluated based diagnostic performance metrics, including area under receiver operating characteristic curve (AUC), accuracy, F1 score, sensitivity, specificity. Additive exPlanations (SHAP) used for interpretability.ResultsThe achieved highest AUC test set (AUC=0.853), outperforming clinical (AUC=0.758), (AUC=0.831), DL_radiomics (AUC=0.834) models. Decision analysis (DCA) demonstrated that offered greater net benefits. SHAP global interpretation, summary plot indicated ct_original_glrlm_LongRunHighGrayLevelEmphasis, pet_gradient_glcm_lmc1 most important predictions.ConclusionThe model, PET/CT, significantly improved accuracy LNM prediction patients. SHAP-based provided valuable insights decision-making process, enhancing its potential application NSCLC.

Язык: Английский

---

Prediction of regional lymph node status in rectal cancer with radiomics features based on deep learning segmented tumor area

Research Square (Research Square), Год журнала: 2024, Номер unknown

Опубликована: Дек. 17, 2024

Background: To predict regional lymph node metastasis (LNM) in rectal cancer (RC) using deep learning-based tumor auto-segmentation and radiomics. Methods: This single-center research retrospectively analyzed 282 patients with RC from two MR vendors. The models were constructed on T2WI DWI 3D U-Net, V-Net, nnU-Net v2 assessed the Dice Similarity Coefficient (DSC). Radiomics features manual-based VOI (MbV) (DbV, highest DSC) extracted respectively. After feature normalization selection, five classifiers used for radiomics model building then LNM prediction. optimal was selected a 5-fold cross-validation strategy evaluated area under curve (AUC), accuracy, specificity, sensitivity. Results: The DSC of significantly higher than that U-Net V-Net (T2WI: 0.886 vs 0.548 0.616, p < 0.001; DWI: 0.906 0.583 0.433, test set). AUC DbV based-radiomics (0.700 T2WI, 0.667 DWI, 0.800 + DWI) comparable to those corresponding MbV-based (0.633 p = 0.638; 0.700 0.544; 0.833 DWI, 0.248) Conclusions: Radiomics based nnU-net segmented showed reliable performance predicting RC.

Язык: Английский

---