Radiomic Machine-Learning Classifiers for Prognostic Biomarkers of Head and Neck Cancer

Frontiers in Oncology, Journal Year: 2015, Volume and Issue: 5

Published: Dec. 3, 2015

"Radiomics" extracts and mines a large number of medical imaging features in non-invasive cost-effective way. The underlying assumption radiomics is that these quantify phenotypic characteristics an entire tumor. In order to enhance applicability clinical oncology, highly accurate reliable machine-learning approaches are required. this radiomic study, 13 feature selection methods 11 classification were evaluated terms their performance stability for predicting overall survival head neck cancer patients.Two independent cohorts investigated. Training cohort HN1 consisted 101 patients. Cohort HN2 (n = 95) was used validation. A total 440 extracted from the segmented tumor regions CT images. Feature compared using unbiased evaluation framework.We observed three minimum redundancy maximum relevance (AUC 0.69, Stability 0.66), mutual information 0.66, 0.69), conditional infomax extraction 0.68, 0.7) had high prognostic stability. classifiers BY 0.67, RSD 11.28), RF 0.61, 7.36), NN 0.62, 10.52) also showed Analysis investigating variability indicated choice method major factor driving variation (29.02% variance).Our study identified prediction Identification optimal radiomics-based analyses could broaden scope precision oncology care.

Language: Английский

---

Residual Convolutional Neural Network for the Determination of IDH Status in Low- and High-Grade Gliomas from MR Imaging

Clinical Cancer Research, Journal Year: 2017, Volume and Issue: 24(5), P. 1073 - 1081

Published: Nov. 22, 2017

Purpose: Isocitrate dehydrogenase (IDH) mutations in glioma patients confer longer survival and may guide treatment decision making. We aimed to predict the IDH status of gliomas from MR imaging by applying a residual convolutional neural network preoperative radiographic data.Experimental Design: Preoperative was acquired for 201 Hospital University Pennsylvania (HUP), 157 Brigham Women's (BWH), 138 The Cancer Imaging Archive (TCIA) divided into training, validation, testing sets. trained each sequence (FLAIR, T2, T1 precontrast, postcontrast) built predictive model outputs. To increase size training set prevent overfitting, we augmented images introducing random rotations, translations, flips, shearing, zooming.Results: With our model, achieved prediction accuracies 82.8% (AUC = 0.90), 83.0% 0.93), 85.7% 0.94) within sets, respectively. When age at diagnosis incorporated increased 87.3% 87.6% 0.95), 89.1% respectively.Conclusions: developed deep learning technique noninvasively genotype grade II-IV using conventional multi-institutional data set. Clin Res; 24(5); 1073-81. ©2017 AACR.

Language: Английский

---

Quantitative radiomics studies for tissue characterization: a review of technology and methodological procedures

British Journal of Radiology, Journal Year: 2016, Volume and Issue: 90(1070)

Published: Dec. 12, 2016

Quantitative analysis of tumour characteristics based on medical imaging is an emerging field research. In recent years, quantitative features derived from CT, positron emission tomography and MR scans were shown to be added value in the prediction outcome parameters oncology, what called radiomics field. However, results might difficult compare owing a lack standardized methodologies conduct image analyses. this review, we aim present overview current challenges, technical routines protocols that are involved studies. The first issue should overcome dependency several scan acquisition reconstruction parameters. Adopting consistent methods subsequent target segmentation step evenly crucial. To further establish robust analyses, standardization or at least calibration different feature extraction settings required, especially for texture- filter-based features. Several open-source commercial software packages perform currently available, all with slightly functionalities, which makes benchmarking quite challenging. number calculated typically larger than patients studied, emphasizes importance proper selection model-building prevent overfitting. Even though many these challenges still need addressed before can brought into daily clinical practice, expected critical component integration image-derived information personalize treatment future.

Language: Английский

---

Radiomics: a new application from established techniques

Expert Review of Precision Medicine and Drug Development, Journal Year: 2016, Volume and Issue: 1(2), P. 207 - 226

Published: March 3, 2016

The increasing use of biomarkers in cancer have led to the concept personalized medicine for patients. Personalized provides better diagnosis and treatment options available clinicians. Radiological imaging techniques provide an opportunity deliver unique data on different types tissue. However, obtaining useful information from all radiological is challenging era 'big data'. Recent advances computational power genomics generated a new area research termed Radiomics. Radiomics defined as high throughput extraction quantitative features or texture (radiomics) decode tissue pathology creating dimensional set feature extraction. Radiomic about gray-scale patterns, inter-pixel relationships. In addition, shape spectral properties can be extracted within same regions interest images. Moreover, these further used develop models using advanced machine learning algorithms that may serve tool guidance.

Language: Английский

---

Radiomic phenotype features predict pathological response in non-small cell lung cancer

Radiotherapy and Oncology, Journal Year: 2016, Volume and Issue: 119(3), P. 480 - 486

Published: April 14, 2016

Language: Английский

---