Pengembangan Automatic Windowing Pada MRI T2 Kuantitatif Berbasis Convolutional Neural Network

Wakhidah, Anisa Nur (2026) Pengembangan Automatic Windowing Pada MRI T2 Kuantitatif Berbasis Convolutional Neural Network. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5001221065-Undergaduate_thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Magnetic Resonance Imaging (MRI) dapat menghasilkan citra dengan rentang intensitas tinggi yang membutuhkan penyesuaian Window Level (WL) dan Window Width (WW) agar optimal pada monitor. Penyesuain manual dalam praktik klinis memerlukan waktu dan bersifat objektif. Penelitian ini mengevaluasi performa Automatic Windowing Convolutional Neural Network (AWCNN) untuk memprediksi parameter WW dan WL secara otomatis, serta dapat membandingkan kualitas citra sebelum dan sesudah windowing. Pemindaian dilakukan dengan menggunakan MRI 3Tesla dengan TR 1000 ms dan TE 4 variasi 75 ms, 100 ms, 150 ms, dan 200 ms. Model CNN menggunakan STD rasio r = 4 dan 5 blok konvolusi. Prediksi dievaluasi dengan Mean Absolute Error (MAE), Mean Relative Error (MRE), dan Pearson Product Moment Correlation Coefficient (PMCC). Kualitas citra dievaluasi melalui Signal-to-Noise Ratio (SNR) dan Contrast-to-Noise Ratio (CNR). Hasil menunjukkan AWCNN sangat akurat dan dapat mengungguli metode Global Stretching Contrast (GCS). Nilai MAE mencapai 0,43 untuk WL dan 2,58 untuk WW, dengan MRE 0,05% (WL) dan 0,14% (WW). Nilai korelasi (p) mencapai 0,9992 (WL) dan 0,9979 (WW). Penerapan AWCNN ini dapat meningkatkan sinyal dan kontras jaringan, didukung oleh peningkatan nilai SNR dan CNR di seluruh Region of Interest (ROI) dibandingkan citra tanpa automatic windowing.
=======================================================================================================================================
Magnetic Resonance Imaging (MRI) produces high-intensity images requiring precise Window Width (WW) and Window Level (WL) adjustments for optimal display. Manual windowing is often subjective and time-consuming. This study evaluates an Automatic Windowing Convolutional Neural Network (AWCNN) to predict WW and WL automatically and analyzes image quality before and after windowing. A gel phantom was scanned using a GE Signa 3T MRI with a Repetition Time (TR) of 1000 ms and 4 Echo Time (TE) variations is TE 75 ms, 100 ms, 150 ms, and 200 ms. The CNN utilized a Space-to-Depth (STD) layer with a ratio of r=4 and five convolution blocks. Accuracy was evaluated using Mean Absolute Error (MAE), Mean Relative Error (MRE), and the Pearson Product-Moment Correlation Coefficient (PMCC). Image quality was assessed via Signal-to-Noise Ratio (SNR) and Contrast-to-Noise Ratio (CNR). The results showed AWCNN is highly accurate, outperforming the Global Contrast Stretching (GCS) method. The AWCNN achieved an MAE of 0.43 for WL and 2.58 for WW, with MREs of 0.05% (WL) and 0.14% (WW). Correlation values (

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Automatic Windowing, Convolutional Neural Network, Contrast to Noise Ratio, Magnetic Resonance Imaging, Signal Noise to Ratio, Automatic Windowing, Contrast-to-Noise Ratio, Convolutional Neural Network, Magnetic Resonance Imaging, phantom Gel, Signal-to-Noise Ratio.
Subjects:	Q Science R Medicine > R Medicine (General) > R858 Deep Learning R Medicine > RC Internal medicine > RC78.7.N83 Magnetic resonance imaging.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Physics > 45201-(S1) Undergraduate Thesis
Depositing User:	Anisa Nur Wakhidah
Date Deposited:	02 Aug 2026 00:38
Last Modified:	02 Aug 2026 00:38
URI:	http://repository.its.ac.id/id/eprint/141696

Actions (login required)

View Item