سامانه همایش‌ها TSTA | ثبت‌نام و ارسال مقاله کنفرانس‌ها و کنگره‌های ملی

صفحه اصلی / سی و دومین کنفرانس ملی و دهمین کنفرانس بین المللی مهندسی زیست پزشکی ایران

Added value of synthetic T1/T2-weighted MR images in the segmentation and staging of meningioma

نویسندگان :

Masoud Noroozi¹ Sayna Jamaati² Peyman Sharifian³ Mahsa Karbasi⁴ Esmaeil Gharepapagh⁵ Alireza Karimian⁶ Hossein Arabi⁷ Sahar Rezaei⁸

1- گروه مهندسی پزشکی، دانشکده مهندسی، دانشگاه اصفهان 2- دانشگاه صنعتی شریف 3- گروه مهندسی پزشکی، دانشکده مهندسی، دانشگاه اصفهان 4- دانشگاه علوم پزشکی تبریز 5- دانشگاه علوم پزشکی تبریز 6- گروه مهندسی پزشکی، دانشکده مهندسی، دانشگاه اصفهان 7- دانشگاه ژنو 8- دانشگاه علوم پزشکی تبریز

کلمات کلیدی :

Meningioma،T1/T2 Ratio Imaging،nnU-Net،Segmentation،Automated WHO Grading،Deep Learning

چکیده :

Accurate pre-operative classification and volumetric definition of intracranial meningiomas are paramount to the development of appropriate surveillance, surgical, and radiotherapeutic approaches. The traditional post-contrast T1-weighted MRI (T1c) is used clinically but is time-consuming to contour and not always available or against indications. To assess how a bias-corrected native-T1/ T2-weighted ratio (T1(n)/T2(w)) map, combined with fully automated segmentation and grading networks, can enhance meningioma work-up without gadolinium. The novelty of this research lies in the use of T1n/T2w information to generate synthetic images, replacing the need for four separate MRI sequences. The BraTS-MEN multi-centre dataset (685 scans to be segmented and 868 scans to be graded) was skull-stripped and registered to the atlas. A 3-D nnU-Net V2 was trained to segment tumors using (i) T1c and (ii) T1n/T2w volumes. The resulting masks were either presented directly or supplemented with the four mpMRI channels in a 3-D ResNet-18 to predict WHO grades 1-2. Performance was measured as Dice, IoU, accuracy, and class-based sensitivity/specificity. T1c performed the best in terms of geometry Dice (92.12 ± 4.14 %) and IoU (86.35 ± 11.3 %). The T1n/T2w map nonetheless maintained a clinically satisfactory Dice of 82.1 ± 11.68 with decreased false-positive voxels in neighboring dura. The ratio-based pipeline was superior to the T1c model in all global measures (accuracy 0.61 vs 0.427; mean Dice 0.558 vs 0.425) and the sensitivity of high-grade (>= WHO II) tumor was over twofold higher (0.70 vs 0.31). The T1n/T2w ratio map, as a gadolinium-free contrast agent, and the nnU-Net V2 in segmentation and ResNet-18 in the classifier demonstrate strength in automated grade assessment accuracy and preventing the under-treatment of aggressive meningiomas. An open-source, hardware-light, and easily reproducible workflow indicates a potential avenue of non-invasive, contrast-sparing pre-operative assessment that could be validated in a multi-institutional prospective setting.

لیست مقالات

لیست مقالات بایگانی شده

Physics-Informed Neural Networks for Cardiac Flow Estimation in 2D Simplified Human Right Ventricular Geometry

Mohammadmahdi Sekhavatpisheh - Nasser Fatouraee

Injectability Enhancement and Optimization of a Biphasic Calcium Phosphate Bone Cement

Sepehr Larijani - Mitra Asadi-Eydivand - Nabiollah Abolfathi - Mehran Solati-Hashjin

EEG-based Schizophrenia Detection Using Spectral, Entropy, and Graph Connectivity Features with Machine Learning

Nazila Ahmadi Daryakenari - Seyed Kamaledin Setarehdan

علیت بین توسعه مالی و نوع ابزارهای مالی از نظر مدیریت پرتفولیو

فاطمه خسروی

بررسی تأثیر ریسک اطلاعات مالی و بندهای تعدیلی حسابرسی بر وجوه هیئت‌مدیره مشترک در شرکت‌های پذیرفته شده در بورس اوراق بهادار تهران

حمیدرضا عزیزی - عرفان تخستین حلم

Assessing the Risk of Musculoskeletal Injuries of Workers at the Warehousing Workstation of Iran Tire Company

Mahshad Nazari Jeirani - Amirhossein Mohammadzadeh - Seyedeh Shokouh Azam Mirdamadi - Mohadeseh Sadat Shahangian - Navid Arjmand

نقش هوش مصنوعی در شخصی‌سازی تجربه مشتری: بررسی رفتار مصرف‌کننده در فروشگاه‌های آنلاین

بهزاد بالازاده - حسین بوداقی - نازلی قراچورلو

Investigation of microbubble motion in a microvessel with various obstructions filled with viscous fluid: A finite element modeling study

Mahdi Mirzaei - Afsaneh Mojra

Document Clustering Using Deep Pre-trained Language Model Embeddings for Information Retrieval

Mahdi Mohammadiha - Mohammad Hassan Sadreddini - Morteza Mohammadi Zanjireh

TransFuse++: A Hybrid CNN-Transformer Architecture with Cross-Attention, Temporal Modeling, and Uncertainty Estimation for Medical Image Segmentation

Masoud Noroozi - Sayna Jamaati - Hamed Aghapanah - Ali Saeeidi Rad - Mahsa Asadi Anar - Ali Darzi - Mahla Shokouhfar - Helia Sadat Kazemi - Mohammadreza Ghahari - Mohammad Saeed Soleimani Meigoli - Jafar Majidpour - Hossein Arabi - Ali Reza Karimian

بیشتر