چاپ صفحه |  صفحه نخست سامانه |  نویسندگان |  اطلاعات تفضیلی |  دانلود مقاله | دانشگاه علوم پزشکی تبریز |
| نویسنده ثبت کننده مقاله | داود خضرلو |
| مرحله جاری مقاله | تایید نهایی |
| دانشکده/مرکز مربوطه | دانشکده پیراپزشکی |
| کد مقاله | 82807 |
| عنوان فارسی مقاله | تجزیه و تحلیل کمی تصاویر تشدید مغناطیسی با کمک یادگیری ماشین برای درجه بندی گلیوما: یک مطالعه رادیومیکس |
| عنوان لاتین مقاله | Quantitative analysis of magnetic resonance images by the aid of machine learning for glioma grading: a radiomics study |
| نوع ارائه | سخنرانی |
| عنوان کنگره / همایش | کنگره بین المللی رویکردهای نوین سبک زندگی، پیشگیری و درمان سرطان |
| نوع کنگره / همایش | بین المللی |
| کشور محل برگزاری کنگره/ همایش | Iran (Islamic Republic) |
| شهر محل برگزاری کنگره/ همایش | آمل |
| سال انتشار/ ارائه شمسی | 1402 |
| سال انتشار/ارائه میلادی | 2023 |
| تاریخ شمسی شروع و خاتمه کنگره/همایش | 1402/07/13 الی 1402/07/15 |
| آدرس لینک مقاله/ همایش در شبکه اینترنت | https://clpt.ir/2023 |
| آدرس علمی (Affiliation) نویسنده متقاضی | Department of Radiology, Faculty of Allied Medical Sciences, Tabriz University of Medical Science, Tabriz, Iran. |
| نویسنده | نفر چندم مقاله |
|---|---|
| داود خضرلو | اول |
| عنوان | متن |
|---|---|
| کلمات کلیدی | : 'Glioma Tumors,' 'Gross Tumor Volumes,' 'Magnetic Resonance Imaging,' 'Radiomics,' 'Machine Learning' |
| خلاصه مقاله | Introduction: One of the most critical challenges in diagnosing gliomas is differentiating low-grade from high-grade tumors. Classification of patient groups in terms of disease grading can significantly impact the process of diagnosis, treatment, and follow-up, which requires logical and personalized methods specific to each patient. Radiomics is a newly emerging machine learning-based technology that could resolve the problem of subjective judgments by radiologists that are vulnerable to inter-observer variability by converting encrypted medical images into usable data and extracting high throughput imaging features, and relating feature data to clinical outcomes. This study aimed to improve the differentiation performance of grades II and III gliomas using a completely non-invasive and quantitative method by radiomics analysis of magnetic resonance imaging (MRI) based on a radiation oncology tumor border identification perspective. Methods: In this study, MRI images of 120 patients with glioma were obtained and divided into a training cohort (n=80, 40 grade II and 40 grade III) and an independent validation cohort (n=40, 20 grade II and 20 grade III). All images were first resampled to 512*512 matrix size and 1*1*1 mm3 voxel size. In each image, the gross tumor volume (GTV) area was drawn by an oncologist and selected as the region of interest (ROI) for analysis. Then, we extracted quantitative information from the selected tumor volumes. More than one hundred features were extracted from each patient’s images. They included three main groups intensity histogram (mean, median, skewness, kurtosis, etc.), shape (volume, surface area density, maximum diameter, etc.), and textural information (homogeneity, heterogeneity, correlation, cluster tendency, contrast, etc.). The segmentation and feature extraction process was done in 3D Slicer software with the aid of the pyradiomics extension. Machine learning-based models were trained and then validated on the independent cohort using the support vector machine (SVM) algorithm in MATLAB 2021b software. The models' performance was assessed by calculating the area under the curve (AUC) of the receiver operating characteristic (ROC). Results: Our findings showed that using a linear SVM algorithm in conjunction with extracting radiomics features from GTV regions is an efficient approach for differentiating glioma grade II and III tumors completely non-invasive. Machine learning-based modeling showed high performance in both training (AUC=0.90, Accuracy=0.82) and validation (AUC=0.84, Accuracy=0.85) cohorts. Homogeneity, correlation, kurtosis, and surface area density were the four most predictive radiomics features in terms of glioma grading in this study (p-value<0.05). Conclusion: The complexity of semi-automated and automated AI-based methods of segmentation has been always a big obstacle in the way of radiomics being practically implemented in the routine clinical diagnosis and management of patients and specifically glioma patients. Simple volume delineation such as gross tumor volume (GTV) may improve not only the segmentation workflow but also radiomics inter-and intra-observer reproducibility. Radiomics analysis of glioma tumors based on GTV regions and training of linear SVM models can lead to highly non-invasive and accurate diagnoses of patients. It has the potential to complement or replace tumor biopsies and develop novel prognostic markers for glioma patients in the near future. |
| نام فایل | تاریخ درج فایل | اندازه فایل | دانلود |
|---|---|---|---|
| article 1-Oral-Certicication.pdf | 1402/07/22 | 1266259 | دانلود |
| oral.pdf | 1403/01/05 | 515127 | دانلود |
