The Biological Meaning of Radiomic Features

Radiology, Journal Year: 2021, Volume and Issue: 299(2), P. E256 - E256

Published: April 26, 2021

Language: Английский

---

Radiomics in Oncology: A Practical Guide

Radiographics, Journal Year: 2021, Volume and Issue: 41(6), P. 1717 - 1732

Published: Oct. 1, 2021

Radiomics refers to the extraction of mineable data from medical imaging and has been applied within oncology improve diagnosis, prognostication, clinical decision support, with goal delivering precision medicine. The authors provide a practical approach for successfully implementing radiomic workflow planning conceptualization through manuscript writing. Applications in typically are either classification tasks that involve computing probability sample belonging category, such as benign versus malignant, or prediction events time-to-event analysis, overall survival. is multidisciplinary, involving radiologists scientists, follows stepwise process tumor segmentation, image preprocessing, feature extraction, model development, validation. Images curated processed before which can be performed on tumors, subregions, peritumoral zones. Extracted features describe distribution signal intensities spatial relationship pixels region interest. To performance reduce overfitting, redundant nonreproducible removed. Validation essential estimate new iteratively samples dataset (cross-validation) separate hold-out by using internal external data. A variety noncommercial commercial software applications used. Guidelines artificial intelligence checklists useful when writing up studies. Although interest field continues grow, should familiar potential pitfalls ensure meaningful conclusions drawn. Online supplemental material available this article. Published under CC BY 4.0 license.

Language: Английский

---

Diagnosis of Coronavirus Disease 2019 (COVID-19) With Structured Latent Multi-View Representation Learning

IEEE Transactions on Medical Imaging, Journal Year: 2020, Volume and Issue: 39(8), P. 2606 - 2614

Published: May 5, 2020

Recently, the outbreak of Coronavirus Disease 2019 (COVID-19) has spread rapidly across world. Due to large number affected patients and heavy labor for doctors, computer-aided diagnosis with machine learning algorithm is urgently needed, could largely reduce efforts clinicians accelerate process. Chest computed tomography (CT) been recognized as an informative tool disease. In this study, we propose conduct COVID-19 a series features extracted from CT images. To fully explore multiple describing images different views, unified latent representation learned which can completely encode information aspects endowed promising class structure separability. Specifically, completeness guaranteed group backward neural networks (each one type features), while by using labels enforced be compact within COVID-19/community-acquired pneumonia (CAP) also margin between types pneumonia. way, our model well avoid overfitting compared case directly projecting highdimensional into classes. Extensive experimental results show that proposed method outperforms all comparison methods, rather stable performances are observed when varying numbers training data.

Language: Английский

---

CheckList for EvaluAtion of Radiomics research (CLEAR): a step-by-step reporting guideline for authors and reviewers endorsed by ESR and EuSoMII

Insights into Imaging, Journal Year: 2023, Volume and Issue: 14(1)

Published: May 4, 2023

Even though radiomics can hold great potential for supporting clinical decision-making, its current use is mostly limited to academic research, without applications in routine practice. The workflow of complex due several methodological steps and nuances, which often leads inadequate reporting evaluation, poor reproducibility. Available guidelines checklists artificial intelligence predictive modeling include relevant good practices, but they are not tailored radiomic research. There a clear need complete checklist study planning, manuscript writing, evaluation during the review process facilitate repeatability reproducibility studies. We here present documentation standard research that guide authors reviewers. Our motivation improve quality reliability and, turn, name CLEAR (CheckList EvaluAtion Radiomics research), convey idea being more transparent. With 58 items, should be considered standardization tool providing minimum requirements presenting In addition dynamic online version checklist, public repository has also been set up allow community comment on items adapt future versions. Prepared revised by an international group experts using modified Delphi method, we hope will serve well as single scientific reviewers literature.

Language: Английский

---

Assessing robustness of radiomic features by image perturbation

Scientific Reports, Journal Year: 2019, Volume and Issue: 9(1)

Published: Jan. 24, 2019

Image features need to be robust against differences in positioning, acquisition and segmentation ensure reproducibility. Radiomic models that only include can used analyse new images, whereas with non-robust may fail predict the outcome of interest accurately. Test-retest imaging is recommended assess robustness, but not available for phenotype interest. We therefore investigated 18 methods determine feature robustness based on image perturbations. perturbation were compared 4032 computed from gross tumour volume two cohorts tomography imaging: I) 31 non-small-cell lung cancer (NSCLC) patients; II): 19 head-and-neck squamous cell carcinoma (HNSCC) patients. Robustness was measured using intraclass correlation coefficient (1,1) (ICC). Features ICC$\geq0.90$ considered robust. The NSCLC cohort contained more test-retest than HNSCC ($73.5\%$ vs. $34.0\%$). A chain consisting noise addition, affine translation, growth/shrinkage supervoxel-based contour randomisation identified fewest false positive (NSCLC: $3.3\%$; HNSCC: $10.0\%$). Thus, this robustness.

Language: Английский

---

Standardization of brain MR images across machines and protocols: bridging the gap for MRI-based radiomics

Scientific Reports, Journal Year: 2020, Volume and Issue: 10(1)

Published: July 23, 2020

Abstract Radiomics relies on the extraction of a wide variety quantitative image-based features to provide decision support. Magnetic resonance imaging (MRI) contributes personalization patient care but suffers from being highly dependent acquisition and reconstruction parameters. Today, there are no guidelines regarding optimal pre-processing MR images in context radiomics, which is crucial for generalization published signatures. This study aims assess impact three different intensity normalization methods (Nyul, WhiteStripe, Z-Score) typically used MRI together with two discretization (fixed bin size fixed number). The these was evaluated first- second-order radiomics extracted brain MRI, establishing unified methodology future studies. Two independent datasets were used. first one (DATASET1) included 20 institutional patients WHO grade II III gliomas who underwent post-contrast 3D axial T1-weighted (T1w-gd) T2-weighted fluid attenuation inversion recovery (T2w-flair) sequences devices (1.5 T 3.0 T) 1-month delay. Jensen–Shannon divergence compare pairs histograms before after normalization. stability first-order across acquisitions analysed using concordance correlation coefficient intra-class coefficient. second dataset (DATASET2) public TCIA database 108 135 IV glioblastomas. based tumour classification task (balanced accuracy measurement) five well-established machine learning algorithms. Intensity improved robustness performances subsequent models. For T1w-gd sequence, mean balanced increased 0.67 (95% CI 0.61–0.73) 0.82 0.79–0.84, P = .006), 0.79 0.76–0.82, .021) 0.80–0.85, . 005), respectively, Nyul, WhiteStripe Z-Score compared relative makes unnecessary use features. Even if number had small performances, good compromise obtained 32 bins considering both T2w-flair sequences. No significant improvements observed feature selection. A standardized pipeline proposed tumours. models features, we recommend normalizing method adopting an absolute approach. feature-based signatures, can be without prior In cases, recommended. may pave way multicentric development validation MR-based biomarkers.

Language: Английский

---

Radiomics in predicting treatment response in non-small-cell lung cancer: current status, challenges and future perspectives

European Radiology, Journal Year: 2020, Volume and Issue: 31(2), P. 1049 - 1058

Published: Aug. 18, 2020

Abstract Objectives Radiomics is the extraction of quantitative data from medical imaging, which has potential to characterise tumour phenotype. The radiomics approach capacity construct predictive models for treatment response, essential pursuit personalised medicine. In this literature review, we summarise current status and evaluate scientific reporting quality research in prediction response non-small-cell lung cancer (NSCLC). Methods A comprehensive search was conducted using PubMed database. total 178 articles were screened eligibility 14 peer-reviewed included. score (RQS), a radiomics-specific metric emulating TRIPOD guidelines, used assess quality. Results Included studies reported several markers including first-, second- high-order features, such as kurtosis, grey-level uniformity wavelet HLL mean respectively, well PET-based metabolic parameters. Quality assessment demonstrated low median + 2.5 (range − 5 9), mainly reflecting lack reproducibility clinical evaluation. There extensive heterogeneity between due differences patient population, stage, modality, follow-up timescales workflow methodology. Conclusions not yet been translated into use. Efforts towards standardisation collaboration are needed identify reproducible radiomic predictors response. Promising must be externally validated their impact evaluated within pathway before they can implemented decision-making tool facilitate patients with NSCLC. Key Points • included promising cancer; however, there studies. (RQS) 9). Future should focus on implementation standardised features software, together external validation prospective setting.

Language: Английский

---

Multimodal data integration using machine learning improves risk stratification of high-grade serous ovarian cancer

Nature Cancer, Journal Year: 2022, Volume and Issue: 3(6), P. 723 - 733

Published: June 28, 2022

Abstract Patients with high-grade serous ovarian cancer suffer poor prognosis and variable response to treatment. Known prognostic factors for this disease include homologous recombination deficiency status, age, pathological stage residual status after debulking surgery. Recent work has highlighted important information captured in computed tomography histopathological specimens, which can be exploited through machine learning. However, little is known about the capacity of combining features from these disparate sources improve prediction treatment response. Here, we assembled a multimodal dataset 444 patients primarily late-stage discovered quantitative features, such as tumor nuclear size on staining hematoxylin eosin omental texture contrast-enhanced tomography, associated prognosis. We found that contributed complementary relative one another clinicogenomic features. By fusing histopathological, radiologic machine-learning models, demonstrate promising path toward improved risk stratification data integration.

Language: Английский

---

Robustness and Reproducibility of Radiomics in Magnetic Resonance Imaging

Investigative Radiology, Journal Year: 2018, Volume and Issue: 54(4), P. 221 - 228

Published: Nov. 15, 2018

The aim of this study was to investigate the robustness and reproducibility radiomic features in different magnetic resonance imaging sequences.A phantom scanned on a clinical 3 T system using fluid-attenuated inversion recovery (FLAIR), T1-weighted (T1w), T2-weighted (T2w) sequences with low high matrix size. For retest data, scans were repeated after repositioning phantom. Test datasets segmented semiautomated approach. Intraobserver interobserver comparison performed. Radiomic extracted standardized preprocessing images. Test-retest assessed concordance correlation coefficients, dynamic range, Bland-Altman analyses. Reproducibility by intraclass coefficients.The number robust (concordance coefficient range ≥ 0.90) higher for calculated from FLAIR than T1w T2w High-resolution images provided highest percentage (n = 37/45, 81%). No considerable difference observed between low- high-resolution (T1w low: n 26/45, 56%; high: 25/45, 54%; T2 21/45, 46%; 24/45, 52%). A total 15 (33%) 45 showed excellent across all demonstrated intraobserver (intraclass 0.75).FLAIR delivers most substrate Only sequences. Care must be taken interpretation studies nonrobust features.

Language: Английский

---

A review in radiomics: Making personalized medicine a reality via routine imaging

Medicinal Research Reviews, Journal Year: 2021, Volume and Issue: 42(1), P. 426 - 440

Published: July 26, 2021

Radiomics is the quantitative analysis of standard-of-care medical imaging; information obtained can be applied within clinical decision support systems to create diagnostic, prognostic, and/or predictive models. performed by extracting hand-crafted radiomics features or via deep learning algorithms. has evolved tremendously in last decade, becoming a bridge between imaging and precision medicine. exploits sophisticated image tools coupled with statistical elaboration extract wealth hidden inside images, such as computed tomography (CT), magnetic resonance (MR), Positron emission (PET) scans, routinely everyday practice. Many efforts have been devoted recent years standardization validation approaches, demonstrate their usefulness robustness beyond any reasonable doubts. However, booming publications commercial applications approaches warrant caution proper understanding all factors involved avoid "scientific pollution" overly enthusiastic claims researchers clinicians alike. For these reasons present review aims guidebook sorts, describing process radiomics, its pitfalls, challenges, opportunities, along ability improve decision-making, from oncology respiratory medicine pharmacological genotyping studies.

Language: Английский

---